Biomechanical Analysis of Proximal Humerus Plate for Spatial Subchondral Support

Transcription

1 Biomechanical Analysis of Proximal Humerus Plate for Spatial Subchondral Support A thesis submitted to The University of Manchester for the degree of Doctor of Philosophy in the Faculty of Science and Engineering 2017 ALI JABRAN School of Mechanical, Aerospace and Civil Engineering

2 2

3 Table of Contents List of Figures... 7 List of Tables Abstract Declaration Copyright Acknowledgements Chapter 1: Introduction Proximal Humerus Fractures Treatments Spatial Subchondral Support Plate Equinoxe Fx Hybrid Blade Plate Project Aims and Objectives Thesis Overview...29 Chapter 2: Literature Review Survey Methodology Biomechanical Testing of Proximal Humerus Plates...65 Loading Type 1: Axial Compression and Tension Loading Type 2: Torsion Loading Type 3: Bending Loading Type 4: Combined Bending and Axial Loading Complex Loading Using Cadaveric Tendons Complex Loading Using Synthetic Tendons Comparison of Plate Technologies and Techniques...79 Locking vs. Non-Locking Screws Polyaxial vs. Monoaxial Locking Screws Importance of Calcar Region Rigid vs. Semi-rigid Plates Cement Augmentation Spatial Subchondral Support Plate

4 2.4 Discussion...85 In Vitro Biomechanical Testing Current Technologies and Techniques Future of Orthopaedic Implant Design Conclusion...91 Chapter 3: Mechanical Testing of Proximal Humerus Plates Protocol Design and Feasibility Assessment Test Rig Design Cyclic/Static Loading Design of Humeral Shaft Holder Design of Humeral Head Holder Plate Choice and Implantation Pilot Studies D Scanning Set-up and Feasibility Tests Using Wood Specimen Validation of Experimental Protocol Testing of Cylindrical Wood Specimen Trial Tests on Humerus With Different Setup Trial Tests Using Static Loading Final Protocol Elastic Testing Plastic Testing Data Processing and Statistical Analysis Post-Processing of 3D Scans Chapter 4: Results and Discussion of Mechanical Tests Results Elastic Testing Plastic Testing D Scanning Discussion Plate Position and Orientation Plate Dimensions

5 4.2.3 Plate Surface Profile Screw Number Screw Position Screw Orientation Screw Dimensions Screw Surface Profile Blade Limitations of In Vitro Mechanical Tests Conclusion Chapter 5: Finite Element Model Creation and Validation Introduction to Computational Framework Three-dimensional Geometry Reverse Engineering Finite Element Model Creation Finite Element Model Validation Discussion Chapter 6: Design Optimisation based on Finite Element Analysis Introduction to Optimisation Study Finite Element Model Automation Optimisation Method Results Fracture Gap Change and F5 Load Von Mises Stress Distribution Contour Plots Position of Nodes with Maximum von Mises Stress Discussion Conclusion Chapter 7: Finite Element Based Study of Screw Surface Profile Introduction Pseudo-threading in Homogenous Cylinder Bone Validation of Pseudo-threading in Bone-Plate Finite Element Model Final Study Method Results Discussion Conclusion

6 Chapter 8: Research Overview Thesis Overview Research Output Arising from this Work Future Works References Appendices Appendix I: Technical Drawings of Compression and Bending Test Rig Appendix II: Technical Drawings of Axial Rotation Test Rig Appendix III: Experimental Protocol (Proposed: August 2014) Comparison Tests Main Experiments Appendix IV: Experimental Protocol (Proposed: November 2014) Comparison Tests Main Experiments Appendix V: Technical Drawings of Testing Machine Components Load Cell Drawing Load Cell-Testing Machine Adaptor Design Humeral Shaft Holder Design Appendix VI: Source Code for Script to Automate Creation of Finite Element Model in Abaqus Appendix VII: Source Code for Script to Determine Feasible Height and Divergence Angles Word Count: 71,244 6

7 List of Figures Figure 1. Classification of proximal humerus fractures according to Neer classification system [9] Figure 2. Classification of proximal humerus fractures according to AO/ASIF classification system [254] Figure 3. Three main plates investigated in this thesis: spatial subchondral support plate, Proximal Humerus Internal Locking System (PHILOS) plate and Fx plate, represented in their correct relative scale Figure 4. Overview of the key stages (and chapters in parentheses) of the two phases in this thesis Figure 5. Literature search profile Figure 6. Four types of loading performed in humerus-only testing studies Figure 7. Overall categorisation of studies included in the literature review Figure 8. Common experimental setups used in literature for applying bending loads Figure 9. Design of the axial compression test rig that allows loading of humeral head with the distal end fixed (inset) Figure 10. Design of the axial compression test rig set in cantilever mode to allow bending of distal humerus with the distal end fixed (inset) Figure 11. Illustration showing the working principle of the pulley system designed, to allow axial rotation in both clockwise (A) and anticlockwise (B) directions with the use of a rope, wheel (blue), fixed pulleys (grey) and a moving pulley (red) onto which the weights could be hanged Figure 12. Design of the axial rotation test rig set based on the pulley system deciding, allowing axial rotation of humeral head along the shaft axis with the distal end fixed (inset) Figure 13. Cantilever bending tests performed by Huff et al. involved loading in the frontal plane to achieve varus and valgus bending (a) and in the sagittal plane for extension and flexion (b). NB. Directions presented here are based on left humerus Figure 14. Holder to allow transfer of load from the load testing machine to the humeral shaft Figure 15. Setup for experimenting with different resins, showing the wood specimen (A) and the resin cavities (B) Figure 16. Hose clamp (left) and the pipe clamp (right) [255,256] Figure 17. Pipe (left) and repair (right) clamp with rubbered grip [256,257] Figure 18. Diagram showing the setup for oriented loading from one of the experimental protocols proposed earlier. N.B. to simplify the diagram, only three screws have been shown Figure 19. A typical base plate used for scaffolding [258] Figure 20. Mould used for preparation of the cement block Figure 21. Humerus in the cement block after chiselling Figure 22. Positioning of Fx plate on the humerus

8 Figure 23. Photograph showing the configuration of external attachment and drilling guide Figure 24. Creaform ExaScan in action [259] Figure 25. 3D scanning setup, showing the location of the position targets and the specimen Figure 26. 3D scanning setup with the black board placed over the specimen Figure 27. 3D scanning setup with the black board placed behind the specimen Figure 28. Selection process of scanned vertical wall to be removed after the completion of the scan of one side of the specimen Figure 29. Set-up of the specimen implanted with S3 plate and held using the circular hole shaft holder Figure 30. Experimental results obtained for the pilot study (blue) with Huff et al results superimposed approximately (black) Figure 31. Experimental results obtained after conducting the pilot study on PHILOS plate Figure 32. Cross-section analysis of 3D humerus model Figure 33. Peak load values for different positions of humerus specimen on its own, without any fracture and implantation Figure 34. Experimental setup using cylindrical wood specimen and metallic plate Figure 35. Peak load values for different positions after testing circular wood specimen Figure 36. New specimen loader design with a semi-circular prism contact surface Figure 37. Experimental setup using the new specimen loader and placing a heavy weight directly above the cement block Figure 38. Peak load values for humeri implanted with PHILOS using the new experimental setup Figure 39. Load-displacement graph for PHILOS specimen tested to failure at 1 mm/s displacement rate. Drops in load at 1-minute scan pauses are clearly visible Figure 40. Load-displacement graph for PHILOS specimen tested to failure at 0.05 mm/s displacement rate Figure 41. Numbering and zoning of screws and blade holes on S3 (A), PHILOS (B) and Fx (C) plates based on their proximity to fracture gap Figure 42. The four loading directions for the elastic tests Figure 43. 3D scans of the setup with the unwanted regions highlight to be deleted Figure 44. 3D scan showing the specimen (left) and the reference block (right) after using the manifold tool and manually deleting unwanted regions Figure 45. Before filling any holes, the hole was inspected for spikes Figure 46. A flat opening such as the distal-most end of the humerus was cleaned using the Fill option

9 Figure 47. Folds and self-intersection were a common issue with the scan of the humerus Figure 48. Scan of the inner sides of the fracture gaps required filling and bridging Figure 49. Post-processed scans of a bone-plate construct before (A) and after (B) 50 mm varus bending Figure 50. Colour map showing the resultant 3D deviation (mm) across a specimen subjected to 50 mm varus bending load, produced using the 3D deviation analysis tool in Geomagic Control software Figure 51. Mean stiffness (S) for S3, PHILOS and Fx plate configuration groups during elastic loading of 5 mm cantilever displacement in extension, flexion, valgus and varus directions Figure 52. Mean peak load (F 5) for S3, PHILOS and Fx plate configuration groups during elastic loading of 5 mm cantilever displacement in extension, flexion, valgus and varus directions Figure 53. Mean peak loads (F) for S3, PHILOS and Fx plate configuration groups during plastic loading at 15 mm displacement before (F 15a) and after (F 15b) eight-minute intermission and at 30 mm displacement (F 30) Figure 54: Typical load-displacement curves at load point for S3, Fx and PHILOS plate control groups (S0, F0, P0) constructs during plastic loading. A drop of 4-5 N in load is noted at 15 mm displacement due to the stress relaxation of construct during the eight-minute intermission Figure 55. Overview of the different ways plates' zones differed from each other. The three plates (S3, PHILOS and Fx) were implanted at different positions and orientations (A). Their geometries were also different, in terms of their dimensions (B), surface profile (C) as well as the number (topology) and position of screw holes (D). Zones also differed in terms of the number (D), orientation (E) and the geometry of their screws. The latter differed with a changed in surface profile (F; e.g. using smooth pegs), dimensions (G) or use of a blade in place of the screw (H) Figure 56. Schematic of a simple analytical model commonly studied in the literature. Cylindrical bone is split into two fragments ('head' and 'shaft') and held together with a cuboid plate and screws. With the head fixed, shaft is loaded in the varus bending direction Figure 57. Insertion of a plate with same elastic stiffness modulus as bone shifts the neutral axis (A). With the insertion of a steel plate, this shift (arrow) is increased much further, highlighting the domination of the plate in the bending behaviour of the whole construct (Adapted from Fig in [158]) Figure 58. Percentage change in the extension, flexion, valgus and varus bending stiffness of all configurations of the S3, PHILOS and Fx plate, as compared to their respective control groups. NB: Cell background colours indicate the magnitude of the change and plates are represented in their correct relative scale Figure 59. Cross section profiles of the S3, PHILOS and the Fx plate at plate sections spanning the fracture site, just before the beginning of their respective zone 1 screw holes. NB: plates are represented in their correct relative scale

10 Figure 60. Photographs of the S3 (A), PHILOS (B) and the Fx plate, superimposed with the shaft- (red) and head- (green) screws' axes Figure 61. Schematic showing the movement of the gap in the screw insertion at a constant working length Figure 62. Cross-sectional views of the bone-plate construct, showing the changes (arrows) in the furthermost point of the screw from the extension and flexion bending neutral axis, for a single centrally aligned 3.8 mm screw (A; e.g. S3 zone 1), a single centrally aligned 6.5 mm screw (B; e.g. Fx zone 2), a near-parallel screw pair (C; e.g. PHILOS plate zone 1) and a divergent screw pair (D; S3 plate, zone 2) Figure 63. Computation framework, from the reverse engineering of the implant and bone geometry to the creation of the FE model and the subsequent automation and optimisation. Majority of the work was performed on three software: Mimics (green), Geomagic Wrap (blue) and Abaqus (red) Figure 64. Reverse engineering process of the humerus, from segmentation of bone geometry in CT scan image (A) to triangular surface mesh s preprocessing and simulation of a two-part fracture (B) Figure 65. Assembly of humerus and plate in the FE model and selection of the head boundary condition surface and the shaft surface to apply varus displacement (red arrow) Figure 66. Plot of the varus bending load values and mesh element number obtained for the five FE models included in the mesh sensitivity study Figure 67. Load at each step increment solved by the FE model compared with the mean in vitro biomechanical test loads Figure 68. Frontal view of the control FE model in loaded (A) and unloaded (B) state with the calculation of change in fracture gap Figure 69. von Mises stress (MPa) distribution across humeral head in the standard FE model, shown from the sagittal (A), transverse (B) and frontal view (C and D) Figure 70. von Mises stress (MPa) distribution across humeral shaft in the standard FE model (A-C) Figure 71. von Mises stress (MPa) distribution across plate in the standard FE model (A-C) Figure 72. von Mises stress (MPa) distribution across the bone-plate construct in the standard FE model (A-B) Figure 73. Visual representation of the divergence angle, θ d (A) and height angle, θ h (B) angle of screws 4 and 5, along with the screws divergence (large black dot) and plate s midplane (dashed grey line) Figure 74. Flowchart of the Geomagic Wrap script, highlighting the key inputs, decisions, functions and loops required for the calculation of feasible height and divergence angle combinations within the user-specified angle ranges Figure 75. Contour plot of 538 feasible combinations of height and divergence angles (yellow), as determined by Python script in Geomagic Wrap

11 Figure 76. Contour plot showing the fracture gap change of each height and divergence angle combination tested in the optimisation study, as a percentage difference from the standard FE model Figure 77. Contour plot showing the varus bending load of each height and divergence angle combination tested in the optimisation study, as a percentage difference from the standard FE model Figure 78. Frontal (A) and sagittal (B) view of the superimposition of the standard (blue) and the optimum (grey) plate Figure 79. von Mises stress (MPa) distribution across humeral head in the optimum FE model, shown from the sagittal (A), transverse (B) and frontal view (C and D) Figure 80. von Mises stress (MPa) distribution across humeral shaft in the optimum FE model (A-C) Figure 81. von Mises stress (MPa) distribution across plate in the optimum FE model (A-C) Figure 82. Contour plot showing the Maximum von Mises stress (MPa) on humeral head of each height and divergence angle combination tested in the optimisation study, as a percentage difference from the control FE model Figure 83. Contour plot showing the Maximum von Mises stress (MPa) on humeral shaft of each height and divergence angle combination tested in the optimisation study, as a percentage difference from the control FE model Figure 84. Contour plot showing the Maximum von Mises stress (MPa) on plate of each height (stress range set at MPa) and divergence angle combination tested in the optimisation study, as a percentage difference from the control FE model Figure 85. Contour plot showing the Maximum von Mises stress (MPa) on plate of each height (stress range set at MPa) and divergence angle combination tested in the optimisation study, as a percentage difference from the control FE model Figure 86. Locations of the nodes (red) with maximum von Mises stress (MPa) on the humeral head (A and B), shaft (C) and plate (D) from the 538 FE models Figure 87. Stress distribution of models with highest (A) and lowest (B and C) peak von Mises stress (MPa) in the humeral head and shaft Figure 88. Stress distribution of models with highest (A) and lowest (B and C) peak von Mises stress (MPa) in the plate Figure 89. Three-months post-surgery varus collapse of humeral head and secondary screw penetration [260] Figure 90. Smooth- (A) and threaded- (B) pegs fixation options provided by the S3 plate Figure 91. A pseudo-threaded screw (A) inserted into a cylinder bone with load applied in five directions (B) Figure 92. Absolute principal strain distributions for qualitative comparison between this study and the Inzana et al. study in five loading directions

12 Figure 93. S0 screw configuration of the S3 proximal humerus plate in reality (A) and in the FE model (B) and the numbering of the plate s head screw holes (C) Figure 94. Load at each step increment solved by the FE model compared with the mean in vitro biomechanical test loads Figure 95. Load required to apply 5 mm varus displacement (F 5) for 25%, 50%, 75% and 100% threading of all six screws in comparison with the model with all screws smooth (dashed line) Figure 96. Percentage change in the load required to apply 5 mm varus displacement (F 5) for 25%, 50%, 75% and 100% threading of all six screws in comparison with the model with all screws smooth Figure 97. Percentage change in the load required to apply 5 mm varus displacement (F 5) for 25%, 50%, 75% and 100% threading of all six screws in comparison with the model with all screws smooth and normalised using screws' lengths Figure 98. Humeral head collapse and secondary perforation of locked screws into the glenohumeral joint with one screw backing out [227] Figure 99. Far cortical locking screw (A) and the exterior and cross-sectional views of dynamic locking screws (B) [251,261] Figure. A.1. Technical drawing of front view of the test rig designed for compression and bending Figure. A.2. Technical drawing of top and bottom view of the test rig designed for compression and bending Figure. A.3. Technical drawing of left side view of the test rig designed for axial rotation Figure. A.4. Technical drawing of front view of the test rig designed for axial rotation Figure. A.5. Technical drawing of top view of the test rig designed for axial rotation Figure A.6. Set-up for axial compression, oriented loading and cantilever bending (L to R) Figure A.7. Drawing of the load cell ordered for ESH testing machine (Model 41, RDP Electronics, West Midlands, UK) [262] Figure A.8. Design of the adaptor to connect the RDP load cell to the ESH testing machine Figure A.9. Designs of the top (A) and bottom (B) parts of the specimen holder to apply load to the humerus at the shaft Figure A.10. Designs of two specimen holder for compressive loading

13 List of Tables Table 1. Loading types for humerus-only testing and the number of studies in which they were performed Table 2. Details of the experimental protocols of the 58 in vitro biomechanical studies included in the literature review Table 3. Length (mm) and descriptions of the screws and blades of S3 plate (green), PHILOS plate (blue) and Fx plate (red) configuration groups where TP stands for threaded peg, SP smooth peg, ND 90o screw, M multidirectional, CA cancellous, CO-L cortical locking, CO-C cortical compression, N/A where no such hole exists and None where the hole does exist but was left unscrewed Table 4. Mean stiffness (K) and load values (F) for all S3 plate configuration groups, along extension, flexion, valgus and varus, with their respective standard deviations (S.D.). K and F 5 denote stiffness and peak load values obtained during elastic tests while F 15a and F 15b are loads at 15 mm before and after eight-minute intermission and F 30 is the load at 30 mm during plastic tests Table 5. Mean stiffness (K) and load values (F) for all PHILOS plate configuration groups, along extension, flexion, valgus and varus, with their respective standard deviations (S.D.). K and F 5 denote stiffness and peak load values obtained during elastic tests while F 15a and F 15b are loads at 15 mm before and after eight-minute intermission and F 30 is the load at 30 mm during plastic tests Table 6. Mean stiffness (K) and load values (F) for all Fx plate configuration groups, along extension, flexion, valgus and varus, with their respective standard deviations (S.D.). K and F 5 denote stiffness and peak load values obtained during elastic tests while F 15a and F 15b are loads at 15 mm before and after eight-minute intermission and F 30 is the load at 30 mm during plastic tests Table 7. P values for elastic stiffness and peak load values of S3 plate configuration groups, obtained from their pairwise comparison statistical analysis Table 8. P values for elastic stiffness and peak load values of PHILOS plate configuration groups, obtained from their pairwise comparison statistical analysis Table 9. P values for elastic stiffness and peak load values of Fx plate configuration groups, obtained from their pairwise comparison statistical analysis Table 10. Results of pairwise comparison statistical analysis: P values for plastic loads at 15 mm displacement before (F 15a) and after (F 15b) eight-minute intermission and at 30 mm displacement (F 30), for S3 plate configuration groups Table 11. Results of pairwise comparison statistical analysis: P values for plastic loads at 15 mm displacement before (F 15a) and after (F 15b) eight-minute intermission and at 30 mm displacement (F 30), for PHILOS plate configuration groups Table 12. Results of pairwise comparison statistical analysis: P values for plastic loads at 15 mm displacement before (F 15a) and after (F 15b) eight-minute 13

14 intermission and at 30 mm displacement (F 30), for Fx plate configuration groups Table 13. Percentage change in the elastic varus bending load (F 5), F 15a, F 15b, and F 30 of all configurations of the S3, PHILOS and the Fx plate, as compared to their respective control groups. NB: Cell background colours indicate the magnitude of the change Table 14. Load required to apply 5 mm varus displacement (F 5) for 0%, 25%, 50%, 75% and 100% threading of all six screws Table 15. Percentage change in the load required to apply 5 mm varus displacement (F 5) for 25%, 50%, 75% and 100% threading of all six screws in comparison with the model with all screws smooth Table 16. Percentage change in the load required to apply 5 mm varus displacement (F 5) for 25%, 50%, 75% and 100% threading of all six screws in comparison with the model with all screws smooth and normalised for screws' lengths Table 17. Cumulated percentage change in the load required to apply 5 mm varus displacement (F 5) for each 25% increment in threading of all six screws

15 Abstract Proximal humerus fractures are the third most common fractures in the over-65 patient population and their stable fixation remains a key challenge in orthopaedic and trauma surgery. While Open Reduction Internal Fixation by plate has become a wellknown treatment modality in the last few decades, clinical studies associate high complication rate with its use. The overall aim of this project was to create a computeraided design framework for proximal humerus plates using a validated subjectspecific humerus-plate finite element model. The framework consisted of three stages: (1) reverse engineering of bone and plate geometry, (2) creation and validation of a finite element model simulating the in vitro testing of the bone-implant construct and (3) parametric optimisation study of implant design using this model. In vitro biomechanical tests were conducted to not only compare the mechanical performance of three key commercially available proximal humerus plates (S3-, Fxand PHILOS plate) but also the effect of different screw zones. Sixty-five humeri specimens with two-part surgical neck fractures were treated and grouped based on their different screw configurations. Extension, flexion, varus and valgus bending were applied in the cantilever fashion in the elastic tests whereas only varus bending was applied in the plastic tests. The load required to apply 5 mm displacement was measured to determine bone-plate construct stiffness. The S3 plate yielded the stiffest constructs and while the removal of the inferomedial support had the most impact on varus bending stiffness, type of medial support was important: inferomedial screws in the Fx plate achieved higher bending stiffness than blade insertion. Stability of constructs treated with the plate was an interplay of factors such as the plate s and screws' number, orientation and position. Next, a subject-specific finite element model of the humerus-plate construct was successfully developed that simulated the stiffest of the constructs from the in vitro varus bending tests conducted in this project. The model was validated against the in vitro results. The validated model was then used to perform a parametric optimisation study where the combination of design parameters (height and divergence angle of S3 plate s inferomedial screws) was determined that achieved optimum bone-plate construct stability (minimum fracture gap change). Out of the 538 designs tested, the optimum design (16 o divergence angle and 33 o height angle) yielded the lowest fracture gap change (0.156 mm) which was 4.686% lower than the standard finite element model while achieving 5.707% higher varus bending load ( N). The validated model was also used to investigate the issue of using smooth pegs and threaded screws. Twenty-six models with different percentages of screw threading were run to compare their bone-plate construct stiffness. While threading the smooth pegs was found to increase the varus bending stiffness by up to 4.546%, it did not affect all screws equally. Finally, the successful completion of the optimisation study of screw orientation and the clinical investigation promises the implementation of the computational framework for a range of future multi-objective optimisation studies of multiple design parameters especially for the design of implants for other parts of the human body and also for investigations into other clinically relevant questions. 15

16 Declaration No portion of the work referred to in the Thesis has been submitted in support of an application for another degree or qualification of this or any other university or other institute of learning. 16

17 Copyright The author of this Thesis (including any appendices and/or schedules to this Thesis) owns certain Copyright or related rights in it (the Copyright ) and he has given The University of Manchester certain rights to use such Copyright, including for administrative purposes. Copies of this Thesis, either in full or in extracts and whether in hard or electronic copy, may be made only in accordance with the Copyright, Designs and Patents Act 1988 (as amended) and regulations issued under it or, where appropriate, in accordance with licensing agreements which the University has from time to time. This page must form part of any such copies made. The ownership of certain Copyright, patents, designs, trademarks and other intellectual property (the Intellectual Property ) and any reproductions of copyright works in the Thesis, for example graphs and tables ( Reproductions ), which may be described in this Thesis, may not be owned by the author and may be owned by third parties. Such Intellectual Property and Reproductions cannot and must not be made available for use without the prior written permission of the owner(s) of the relevant Intellectual Property and/or Reproductions. Further information on the condition under which disclosure, publication and commercialisation of this Thesis, the Copyright and any Intellectual Property and/or Reproductions described in it may take place is available in the University IP Policy, in any relevant Thesis restriction declarations deposited in the University Library, The University Library's regulations and in The University's policy on Presentation of Theses. 17

18 Acknowledgements Foremost, I would like to express my deepest gratitude to my family for the support throughout this lengthy process of pursuing my research. I would also thank my main supervisor, Dr Lei Ren for his excellent guidance, caring, patience, and encouragement through the ups and downs. The good advice and support of Dr Chris Peach, especially on the clinical aspects of my research, for which I am extremely grateful, has been invaluable for me. His efforts in getting the key industrial partners interested in this project and successfully obtaining the grants for surgical hardware were crucial for the completion of the project. This project owes also to the orthopaedic manufacturers Depuy Synthes, Zimmer Biomet and Exactech who provided the surgical hardware necessary for the experimental works. I am indebted to Dr Zhenmin Zou for his support on the technical works related to the design of test-rigs and mechanical testing. I thank Professor Paulo Bartolo and Dr Marco Domingos for providing me with the 3D scanning facilities and for their lively discussions on research life in general. 18

19 Chapter 1: Introduction 1.1 Proximal Humerus Fractures The humerus is one of the three bones that comprise the glenohumeral joint, with the other two being clavicle and scapula. The glenohumeral joint is a ball-and-socket joint that offers the greatest range of motion out of all joints in the human body, allowing the shoulder to perform movements such as flexion, extension, abduction, adduction, circumduction and medial and lateral rotation. It is this freedom of movement coupled with that of the joints belonging to the forearm and hand which represent the triumph of evolution, allowing humans to perform complex tasks. Fractures of the proximal humerus are the third most common fractures of the human body, representing 5% of all fractures in all age groups [1,2]. They are also the third most common fractures in patients over the age of 65 years [1] and are mostly linked to osteoporosis [3]. For older patients, the population of female patients has been reported to be more than twice of that of male [4]. With an ageing population and increasing prevalence of osteoporosis, the number of these fractures is expected to increase dramatically. There are several studies that highlight this unimodal age distribution of proximal humerus fractures and their bias towards the female population and patients with osteoporosis. Epidemiological studies project up to a 300% increase in the incidence of proximal humerus fractures over the next 30 years [5]. Results from such studies along with other global socioeconomic issues such as the increasing cost of healthcare highlight the importance of selecting optimal treatments for proximal humerus fractures. Taking the USA as an example, the estimated total cost of treatment of musculoskeletal conditions in general and the lost wages were reported to be $849 billion in 2004, making up almost 8% of the gross domestic product (GDP) that year, a burden for both patients and the hospitals [6]. In older patients, proximal humerus fractures mostly occur in isolation from a low-energy trauma such as a fall on an outstretched hand from a standing height, the risk of which is often increased by poor vision, balance, coordination and protective reflex. In younger patients, they are more commonly caused by higher energy injuries and direct blows often as part of a polytrauma. While the younger population, in general, has a higher chance of recovery, the elderly face a high risk of prolonged loss of shoulder function and immobility, preventing them from performing activities of daily living activities. A five-year prospective study by Court- Brown et al. [7] has shown that almost a quarter of the elderly patients (20-25%) who could undertake essential activities such as shopping, dressing, personal hygiene, and housework 19

20 were unable to do so and required domestic assistance after non-operative management of proximal humerus fracture. The history of proximal humerus fractures and their management owes much to the study of classification of the fracture patterns. Fracture classification systems guide the decision of fracture treatment, estimation of prognosis and prediction of risks associated with complications. Much has been written in the scientific literature with regard to devising an ideal classification system. For example, the ideal system ought to be clinically useful and intuitive enough to allow easy categorisation from pre-operative images so that it could serve as a tool for reporting in the scientific literature. It will, therefore, be useful to briefly explore some of the classification systems so that our future discussions are easy to follow. One of the earliest attempts at classifying proximal humerus was by Codman [8] who presented a scheme consisting all possible fracture patterns based on the observation that fracture lines mostly occurred between four fragments: humeral head, greater tuberosity, lesser tuberosity, and humeral shaft. Depending on the number of fragments separated, proximal humerus fractures can be classified as two-part, three-part or four-part fractures. Since then, the most commonly employed classification systems have been the Neer and AO/ASIF (Arbeitsgemeinschaft fuer Osteosynthesefragen/Association for the Study of Internal Fixation) classifications which were first proposed by Charles Neer ( ) in 1970 and Maurice Müller et al. ( ) in 1990 respectively. Neer classification is divided into six groups based on Codman s concept of four fragments and it differentiates fracture patterns according to the number of fragments, direction of dislocation, and the involvement of the articular surface [9]. Fracture parts are considered fragments if they have displaced more than 1 cm or angulated more than 45 o and if however, they have displaced less than 1 cm with less than 45 of angulation, they are considered minimally displaced regardless of the number of fracture lines (Fig. 1). 20

21 Figure 1. Classification of proximal humerus fractures according to Neer classification system [9] According to AO/ASIF, proximal fractures can broadly be classified as Type A (extraarticular unifocal), Type B (extraarticular bifocal) and Type C (intraarticular) and each type is then further divided into groups and subgroups (Fig. 2). Unlike Neer classification, this system has a special emphasis on the likelihood of avascular necrosis of the humeral head and the integrity of the vascular supply. For example, (intra)articular fracture patterns (Type C) have the highest risk of avascular necrosis due to the high severity of the injury (Fig. 2). Its complexity and the fact that it does not describe fragment displacement limit its use in research. 21

22 Figure 2. Classification of proximal humerus fractures according to AO/ASIF classification system [254] It has emerged from several studies that both the Neer and AO/ASIF classification systems are complex and have low inter-observer reliability and reproducibility [10 13]. These are some of the many reasons why the classification of proximal humerus fracture as a problem has earned increasing popularity in the scientific literature and there exists disagreement with regard to the best system [14,15]. Nevertheless, for consistency, the Neer classification will be adopted throughout this thesis. 1.2 Treatments The majority (85%) of proximal humerus fractures are minimally displaced [16] and can generally be treated with good functional outcomes via the non-operative approach [14,15,17 20] such as sling immobilisation (e.g. collar and cuff). These typically take a few weeks with the assistance of light physical therapy after which the sling could be removed when the pain permits. 22

23 Management of severely displaced fractures such as that of greater tuberosity and three- and four-part fractures remains a challenge due to osteoporosis rotator cuff displacement and increased risk of necrosis caused by humeral head devascularisation. The non-operative approach has been shown to be troublesome for these fractures as it could lead to shoulder deformity and stiffness [14,21]. Due to the limited evidence, controversy still lies around whether an operative or a non-operative approach should be followed for these fractures, despite the fact that the treatment of these fractures has a history that spans decades [22,23]. This controversy remains unresolved and is widely discussed in the literature in addition to the problem of finding the best of all operative treatments for a given fracture pattern. Several operative treatment modalities have been developed over the years, including Open Reduction Internal Fixation (ORIF) with proximal humerus plate, Closed Reduction and Percutaneous Pinning (CRPP), intramedullary nail fixation, hemiarthroplasty and reverse total shoulder arthroplasty. The main focus of this thesis will be the former (ORIF) which has risen in popularity in recent years. ORIF first involves open surgery of the shoulder to set the humerus fragments and the reduction of bone. This is followed by the internal fixation which is by the means of plates in this case. Plates are designed to remain attached to the humerus, acting as an anchor for the fracture fragments with good purchase, preventing micro-motion across the fracture lines and allowing sufficient stability to perform at least simple shoulder movements and guiding the humerus healing process. One of the earliest reported uses of a plate in the proximal humerus was by Bosworth in 1940, who instrumented a 120 o blade plate in a young patient with congenital humerus varus deformity to provide angular stability [24]. A myriad of plates has been developed since then with a variety of designs and features, with notable examples being T-plate, clover-leaf plate and semi-tubular plates. Until the advent of the locking plate technology over a decade ago, results from several clinical studies suggested that compared with the non-operative approach, ORIF offered little real benefit to patients [7,25]. This was particularly the case with the elderly, osteoporotic patients, for whom clinical results were fair to poor, showing complications such as loosening, pulling out of screws and sub-acromial impingement [26]. Because of these shortcomings of conventional plates, alternative techniques and devices were developed including pins and bands (tension) [27 29]. Despite the fact that the clinical performance of some of these devices has been reported to be similar to that of locking plates, they demand greater surgical exposure and techniques associated with them are found to be more complex [27]. 23

24 In locking plates, the screws are able to lock into the threaded plate, resulting in what is known as an angle-stable fixation. This construct fixation provides better stability and is reliant on the bone-screw interface instead of the bone-plate interface so the bone-plate friction is reduced and there is less dissection and stripping of soft tissue. The failure mode of locking plates also differs from that of conventional plates such as blade and T-plates. While conventional plates typically fail in series due to the toggling, loosening or the pulling out of the screws, failure of locking plates demands the simultaneous pull-out or failure of all screws. Therefore, the locking plates exhibit better pull out strength and stiffness since these properties are related to the construct in entirety and not just individual screws [30,31]. This proves advantageous for small to moderate loading range but catastrophic under high impact forces. Due to these theoretical advantages, locking plates have been reported in the literature to exhibit superior torsion, bending and axial compression stiffness as compared to non-locking plates [32 34]. In the clinical setting, however, their performance is highly variable as several implant-specific problems have arisen. Clinical studies report complications such as varus collapse, screw penetration into the glenohumeral joint, screw cut-out, plate impingement under the acromion, need for revision surgery, malreduction, avascular necrosis and tuberosity displacement [2,35 40]. In response to this, contemporary locking plates, employ several solutions, which can be classified as either being technical or relating to the implant design (design-based). The latter will be the focus of this project but a brief discussion of the former will help us understand the clinical scenario and the design process. Technical solutions relate to the techniques and procedures adopted by the surgeons in the clinical settings in order to yield successful implantation of plates into patients. These include factors such as the augmentation of sutures, importance of medial support and the route for surgical intervention (e.g. deltopectoral and deltoid-splitting approach), all of which have been found to affect the outcomes [40 42]. For example, unlike the much simpler, two-part fractures dealing with anatomical or surgical neck, three and four-part fractures demand careful reduction and fixation of both greater and lesser tuberosities. This is because the tuberosities have rotator cuff attachments, a group of muscles and tendons crucial to shoulder mobility and stability. Contemporary locking plates, therefore, have suture holes in addition to screw holes to allow the sutures to be passed through the insertional fibres of the rotator cuff tendons and be secured to the locking plate. This prevents the over-stressing of fractured tuberosities by allowing the pull of the rotator cuff to directly load the implant. The importance of these 24

25 sutures has been highlighted in clinical studies such as those of Micic et al. [40] and Jung et al. [43], despite the lack of scientific and biomechanical studies examining them. Being a good example of form driven by function, suture augmentation highlights the inter-dependence of technical and implant design solutions. Design-based solutions of proximal humerus plates mostly address the design of two main components: plate and screws. To illustrate the importance of these factors, attention should be drawn to varus collapse, which is one of the most frequently reported modes of failure in patients instrumented with locking plates, often due to the presence of medial comminution or poor bone quality [44]. Here, cortical comminution at the surgical neck occurs, causing the head to first displace and then settle in varus. This collapse is often due to the contraction of deltoid and supraspinatus during shoulder abduction making it difficult to raise the shoulder, thus mechanical support at the medio-inferior region of the humerus has been found to help resist the varus stress in cases where the medial hinge was missing [41]. To restore stability, most popular solutions include the use of supporting screws, strut grafts and structural fibular allografts [41,45,46]. For three- and four-part fractures, clinical results have indicated that medial support screws play a vital role in the maintenance of the fracture alignment and reduce the incidence of varus collapse [47]. Similar situations were reproduced in vitro by Ponce et al. [48] who instrumented three-part fractures with locking plates with calcar fixation (by the use of medial supporting screws) and achieved improved stability when loaded in compression to induce varus collapse. A case like this shows the synergistic relationship between biomechanical and clinical studies that is needed for implant design. In addition to the importance of inferomedial support screw, Erhardt et al. [49] investigated the effect of their placement and number on the penetration. The effect of the direction and arrangement of the screws' direction on fracture stability have also been studied [50]. Early locking plate designs involved screws placed in the same direction but were found to retain complications such as the necrosis and collapse of humeral head [51]. Polyaxial locking systems were therefore developed and later tested by the likes of Zettl et al. [50]. In similar fashion, the effects of screw length, and their geometry and arrangement have been explored in the literature. Biomechanical studies have also investigated the designs of telescoping screws as well as smooth pegs [52,53]. Among the most popular proximal humerus plates in the clinic is the Proximal Humerus Internal Locking System (PHILOS; Synthes, PA, USA) that offers insertion of fixed-angle locking screws along multiple directions to enhance structural stability (Fig. 3). This plate was 25

26 studied extensively throughout this thesis as a representative of conventional locking plates and compared against two more contemporary locking plates: Spatial Subchondral Support plate (S3; Zimmer Biomet, IN, USA) and Equinoxe Fx hybrid blade plate (Exactech, FL, USA). Figure 3. Three main plates investigated in this thesis: spatial subchondral support plate, Proximal Humerus Internal Locking System (PHILOS) plate and Fx plate, represented in their correct relative scale Spatial Subchondral Support Plate The spatial subchondral support (S3) plate was designed with these complications of conventional locking plates in mind. One of the problems with locking screws is that the screw purchase cannot be felt. In combination with the insertion of sharp-threaded screws, this poses the risk of screw penetration into glenohumeral joint. This is particularly accentuated in vivo with varus collapse of the humeral head. Thus, in order to avoid the high stress concentration in the surrounding bone, the S3 plate offers the insertion of smooth pegs. Due to their hemispherical tips, these pegs can be implanted further along the humeral head and closer to the high-density subchondral bone, with the aim of enhancing bone-screw interface. The under-surface of the S3 plate is designed to achieve a more distal placement on the humerus than conventional plates such as the PHILOS plate, in order to avoid sub-acromial impingement. As a result, the angle between its screw and the shaft is 135 o which is closer to 26

27 the natural neck-shaft angle of a normal humerus. In contrast, the proximal-most screws of the PHILOS plate form an angle of 90 o with the humeral shaft. S3 plate s distal placement also allows it to have a thicker cross-section, making it theoretically stiffer in the varus and valgus loading direction. Varus stability, in particular, is considered to be vital for the overall stability of the humeral head. The aforementioned design features of the S3 plate both directly and indirectly influence the bone-screw interface, an interface of critical importance with the use of locking plates. Their theoretical biomechanical advantages are yet to be systematically evaluated. In general, although very few clinical studies have performed an in vivo comparison of the S3 plate with another plate, the short-term studies reveal promising results with their use [54,55]. While complication rate is usually more than 20% for PHILOS plate, Stoddard et al. were able to reduce this to only 3.7% with the use of S3 plate [55 57]. In particular, the rate of screw cutout was reduced. In vitro biomechanical studies on S3 plates are scarce in the literature whereas the PHILOS plate is widely studied Equinoxe Fx Hybrid Blade Plate The general category of blade plates dates back to the Codman study [8], much before the development of locking plates. These plates allow insertion of a blade towards the centre of the humeral head. The fundamental reason behind using a blade in place of the screw is that the former provides the construct with a larger surface area than the latter. Theoretically, this is beneficial not only for support in fracture fixation but also for avoiding cut-out, a common complication associated with screw-based fixations. In general, blade plates have become less popular as they were unable to counter the large coronal plane bending moment. When a blade plate is used, it appears to be associated with poor clinical outcomes [58]. Results from the clinical and biomechanical studies of blade plates are also found to vary considerably, making it difficult to derive a generalised conclusion [34,59 61]. Merging the two types of plates (blade and locking), a concept of hybrid plates has recently emerged, offering implantation of both blades and locking screws to a single plate with the aim of reaping the benefits of the locking and blade plates and compensate for each other s disadvantages. Representing this design philosophy is the Equinoxe Fx hybrid fixed angle blade plate which is studied extensively in this project. No studies could be found on the biomechanical comparison of the Fx plate with other proximal humerus plates. 27

28 1.3 Project Aims and Objectives The overall aim of this project is to create a computer-aided design framework for proximal humerus plates using a validated subject-specific humerus-plate finite element (FE) model. The fulfilment of this aim demands the achievement of the following objectives which have been divided into two phases (Fig. 4). Phase I: Finite Element Model Development and Validation 1. To design and conduct in vitro biomechanical tests of leading proximal humerus plates, based on a systematic review of the in vitro biomechanical studies in the literature. 2. To develop a subject-specific FE model of the humerus-plate construct, simulating the in vitro tests conducted and validate it against in vitro test results. Phase II: Computer-Aided Design Using Finite Element Model 1. To conduct an optimisation study of the screw orientation on the S3 plate, using the FE model developed and validated in phase I. 2. To investigate the biomechanical effects of screw surface profile on the S3 plate, using the FE model developed and validated in phase I. Figure 4. Overview of the key stages (and chapters in parentheses) of the two phases in this thesis 28

29 1.4 Thesis Overview Fig. 4 enumerates each stage of the project to the corresponding chapter in this thesis. With the project aims and objectives defined, the second chapter presents the literature review. This review explored the several aspects (e.g. loading conditions and performance parameters) of the in vitro biomechanics studies found in the literature that tested proximal humerus fractures treated with plates. Results from the literature review guided the decisionmaking required for mechanical testing in the third chapter. Chapter 3 is divided into three parts; the first two describe the process of devising the final clinically-relevant experimental protocol for in vitro biomechanical testing. This included the description of the test rig, mechanical testing component designs and the pilot studies. Where possible, a rationale has been provided for the decisions made throughout this process. Chapter 4 begins with the presentation of the results from the mechanical tests described in chapter 3. A discussion is then provided on the mechanical and clinical significance of the obtained results. This is achieved by discussing the design parameters investigated in the mechanical tests. Framing the discussion in this manner aided the design of the FE based studies in chapter 5-7. Chapter 5 describes the computational framework in details, including the different stages of the development and validation of the FE model of the in vitro elastic varus bending tests of a plate-humerus construct. As an implementation of the framework, the FE model developed in the previous chapter was used to perform FE based parametric optimisation of the proximal humerus plate design in chapter 6. Inferomedial screw orientation was optimised to yield minimum change in the fracture gap change along with the loads and von Mises stresses in both the humerus and the implant. The issue of the biomechanical effects of using smooth pegs over threaded screws (and the vice versa) is controversial in the clinical and biomechanical literature. To investigate this in silico, a study was performed to test the effects of different percentages of screw threading on the bone-plate construct stiffness. Chapter 7 describes how the FE model developed in chapter 5 was improved in order to perform this study. Results arising from this study are also presented and discussed in chapter 7. 29

30 Chapter 8 provides an overview of the key findings and the original contributions of the work conducted in this thesis as well as a discussion on the various avenues for future expansion of this work. 30

31 Chapter 2: Literature Review Fractures of the proximal humerus account for 4-5% of all fractures, making them a common upper extremity injury [1]. In the over-65 patient population, this figure is reportedly much higher, at 10%, often related to factors such as osteoporosis [62]. Approximately 85% of the cases can be treated with a non-operative approach while the remaining complicated fractures require surgical treatment [16,63]. The latter cases have been addressed with varying success using a variety of techniques such as K-wire fixation [64,65], intramedullary nailing [66,67] and open reduction internal fixation using proximal humerus plates (PHPs) [68,69]. With the development of locking technology, PHPs have risen in popularity due to their improved in vitro mechanical performance. In clinical studies, however, several reports describe the high incidence of complications such as screw penetration of the articular surface [37,70 72] and sub-acromial impingement of plate [64,73,74]. In elderly patients, these complications are made worse and stable fixation is even harder to achieve due to the poor anchorage of screws to the osteoporotic bone. During the last two decades, a series of new PHPs has been developed, based on different design philosophies. Several in vitro biomechanical comparisons of PHPs have been conducted with the aim of not only comparing the in vitro properties of the plates but also the technologies and techniques associated with them. One of the most popular approaches to enhance the in vivo functionality of an implant is to optimise its design. When studying an implant for this approach, not only the in vivo, but also the in vitro studies should be evaluated. This is because the design processes derived from this approach often involve the in vitro testing of the proposed designs before the in vivo trials. To be specific, evaluation of the in vitro studies should include the implant s performance and the experimental protocols used. It should be noted that the term protocol here includes many aspects, noteworthy of which are four: loading conditions, methods of applying the loads, criteria set to define implant s failure (failure criteria), and the parameters determined to indicate the implant s performance. An ideal protocol would be both standardised and reproducible, consisting of loading conditions, methods, and failure criteria that all fully depict the in vivo scenario. Also, the parameters determined in an ideal protocol would be strong, quantitative indicators of the implant s in vivo functionality. 31

32 Majority of the literature on PHPs consists of in vivo clinical studies (e.g. clinical trials, observational studies, and case studies) so most literature reviews are also limited to them. Comprehensive reviews of the in vitro biomechanical studies are noticeably scarce. The few that do exist have put more emphasis on studies results instead of the protocols used. Also, they often reviewed biomechanical studies not as the primary aim but as a part of a broad review of all types of studies (including clinical studies). Furthermore, the inclusion criteria that they set are strict, allowing only a specific group of studies with certain types of PHPs and fracture patterns, making it difficult to draw generalised conclusions. Currently, in vitro biomechanical studies of PHPs lack standardisation. For example, a census is required on the choice of clinically important parameters but their sheer number makes comparisons of the plates performances very challenging. To address the aforementioned shortcomings with the current literature, a literature review was conducted with a twofold aim: firstly, to systematically categorise and review the protocols used in the in vitro biomechanical studies of PHPs, and secondly to discuss the results arising from these studies. To achieve the latter, studies were categorised thematically according to the technologies and techniques investigated in them before comparison so that a census could be achieved for each category where possible. It is hoped that the thorough review of the protocols will assist the design of future studies which are closer to the ideal and provide better insight into the issue of standardisation. It is also hoped that the seemingly challenging task of comparing the in vitro and in vivo functionality of plates will be simplified if this review is studied alongside reviews of the clinical literature. 32

33 2.1 Survey Methodology A systematic electronic search of Web of Science and Scopus was performed from the earliest available date up to June 2016, using a defined search strategy: ("proximal humer*" OR "shoulder") AND ("fracture*") AND ("fixation" OR "php" OR "angle stable" OR "lock* plate" OR "blade plate") AND ("*mechani*") Inclusion criteria were set to be open enough to ensure that studies involving most PHPs and fracture patterns are included. To be included, studies had to include in vitro biomechanical testing of PHPs, be written in English and published in a peer-reviewed journal. Studies conducting tests of PHPs along with other proximal humerus fixation devices (e.g. intramedullary nails) were also included. Literature reviews, clinical trials, observational and case studies were excluded. Disagreements were resolved based on a common consent after a discussion. A combined search of Web of Science and Scopus yielded 1078 hits (Fig. 5). Titles and abstracts of the obtained studies were examined to determine their eligibility. After removing duplicates and applying the inclusion criteria, only 70 were found to be relevant. For twelve of these 70 studies, full-text was inaccessible and the abstracts alone did not provide sufficient information to allow for adequate reviewing. The remaining 58 studies were therefore included in the review. Figure 5. Literature search profile From the nature of the literature, one could categorise the included studies on a variety of bases such the type of plates tested, types of parameters determined and even chronologically. 33

34 Here, since our focus was on biomechanical testing, categorisation was according to the type of loading performed. Most studies (n=48) employed relatively simple forms of mechanical testing: axial loading, torsion or bending moment, applied directly on the humerus. They formed the humerus-only testing category as they involved humeral specimens with tendons and musculature removed. Accordingly, they were further divided into four subcategories: axial loading (LT1), torsion (LT2), bending (LT3) and combined bending and axial loading (LT4) as illustrated in Fig

35 Figure 6. Four types of loading performed in humerus-only testing studies 35

36 Most humerus-only studies (23) involved only one type of loading, but in others, combinations of two or three were performed (Table 1). This further complicated the comparison of their results because very often, the same specimen within a single study underwent several loading types, making it difficult to isolate the effects of each loading type. To address this, each study s order of loading was carefully studied and rationale was provided, where possible, for their inclusion or exclusion into the corresponding subcategory. Table 1. Loading types for humerus-only testing and the number of studies in which they were performed Loading Type/s (LT) Description Number of studies Reference 1 Axial Loading 6 [32,50,52,75 77] 2 Torsion 2 [34,78] 3 Bending 3 [79 81] 4 Combined Bending and Axial Loading 12 [48,49,67,82 90] [53,91 93] [33,66,68,94 100] [101,102] [103] [104,105] [ ] [69] 36

37 Other than the humerus-only studies, the remaining 10 studies performed more indirect loading of the humerus, with the use of tendons to achieve complex movements. Thus, they were collectively named humerus-tendon testing. Further division of these studies was made based on the type of tendons used for the loading: cadaveric or synthetic tendons (Fig. 7). Figure 7. Overall categorisation of studies included in the literature review Details of the experimental protocol of these 58 studies are presented in Table 2. 37

38 Table 2. Details of the experimental protocols of the 58 in vitro biomechanical studies included in the literature review Study Implant Bone Specimen Loading Conditions Measurements TYPE 1 LOADING ONLY Seide et al., 2007 [32] TIFIX internal fixator Human cadaver humeri Fracture: Two-part fracture, 10 mm subcapital osteotomy Static Test: Recorded force, plate (LITOS) displacement, determined linear Static Test: Axial compression at 5 mm/min until failure elastic stiffness and load at failure Dynamic Test: Cyclic axial compression with 10 N preload and 80 N maximal load at 5 Hz until 1,000,000 cycles Dynamic Test: Recorded forcedisplacement data, determined Failure: For static test, clear deviation from linearity on load-displacement number of cycles to failure and curve; for dynamic test, automatic stopping of testing machine due to control maximum plastic deformation instability Zettl et al., 2011 [50] PHILOS plate Human cadaver humeri Fracture: Two-part AO-A3 fracture, 10 mm medial wedge osteotomy Cyclic Test: Recorded force, (Synthes) displacement, determined Cyclic Test: 90, 180 and 450 N cyclic axial compression at 1 Hz for 200 maximum plastic deformation Non-contact bridging cycles, then N at 1 Hz for 2000 cycles plate (Zimmer) Static Test: Recorded force, Static Test: Axial compression at 1 N/s until failure displacement, determined force at failure Failure: >30% pressure drop, 30 mm deformation or observation of screw cut-out Bae et al., 2011 [75] PHILOS plate Human cadaver humeri Fracture: First osteotomy in line with humeral anatomical neck and second 1 Dynamic Test: Recorded force (Synthes) cm distal to the inferomedial aspect of articular cartilage of humeral head and displacement, determined construct displacement after 38

39 Dynamic Test: N cyclic axial compression at 5 Hz for 1,000,000 1,000,000 cycles (difference in cycles actuator position before and after cyclical loading) Static Test: Axial compression at 5 mm/min until failure Static Test: Recorded force and Failure: Clear deviation from linearity on load-displacement curve displacement, determined linear elastic stiffness and failure load Wallace et al., 2012 [77] Proximal humerus Human cadaver humeri Fracture: Two-part comminuted fracture, standard 15 mm resection Recorded force and locking plate osteotomy at humeral surgical neck displacement, determined load (Synthes) at failure, stiffness, maximum Static Test: Axial compression at 5 mm/min until failure load, cortical thickness Failure: Screw pull-out or complete closure of fracture gap Gradl et al., 2012 [52] Humerus Tele Human cadaver humeri Fracture: Three-part (Neer IV/3) fracture, 5 mm osteotomy gap simulating Recorded force, displacement Screw plate metaphyseal comminution and number of cycles, (M.O.R.E. Medical determined stiffness and load at Solutions) Cyclic Test: N sinusoidal axial compression at 1 Hz for 50 cycles, failure then increased maximum load by 50 N up to 3000 N every 50 cycles PHILOS plate (Clinical House) Failure: 20 mm humeral head displacement Instrum et al., 1998 [76] Semitubular blade Human cadaver humeri Fracture: Two-part fracture, transverse osteotomy created 10 mm beneath Recorded force and plate the inferior screw of the head fragment displacement, determined yield load and load to failure AO T-plate Static Test 1: Axial tension at 20 mm/min until failure (Synthes) Cyclic Test: N sinusoidal axial tension at 1/6 Hz for 1500 cycles 39

40 Static Test 2: Axial tension at 20 mm/min to failure Failure: N/A TYPE 2 LOADING ONLY Weinstein et al., 2006 Proximal humerus Human cadaver humeri Fracture: Three-part fracture, first osteotomy across surgical neck and Static Test: Recorded torque, [34] locking compression second osteotomy, from lateral aspect of bicipital groove to inferior to rotational displacement, plate (Synthes) tuberosity flare (separating greater tuberosity from humeral head) determined torsional stiffness Angle blade plate Static Test: 0-5 Nm axial external rotation torque at 0.5 /s Cyclic Test: Recorded number (Synthes) of cycles and rotation data, Cyclic Test: 0-5 Nm axial cyclic external rotation torque at 0.5 Hz until determined maximum rotational failure deformation Failure: 30 o humeral head rotation with respect to shaft or 10,000 load cycles Foruria et al., 2010 [78] PHILOS plate Human cadaver humeri Fracture: Two-part fracture, 0.5 cm osteotomy, 1 cm distal to inferomedial Cyclic Test: Recorded torque, (Synthes) aspect of head articular cartilage and perpendicular to shaft interfragmentary angular displacement, determined Proximal humerus Cyclic Test: Sinusoidal cyclic rotational torque between 0.25 Nm internal maximum angular displacement nail with spiral blade rotation and 1112 Nm external rotation at 1 Hz for 10,000 cycles between proximal and distal (Synthes) fragments Static Test: External rotation at 30 o /min loading rate until failure Static Test: Recorded torque Failure: For cyclic test, implant and bone dissociation, appearance of new and interfragmentary angular fracture line, >45 o angular displacement displacement, determined torsional stiffness, torque at failure and energy absorbed until 40

41 For static test, dissociation between implant and bone, appearance of new fracture line, or implant fracture failure (area under torque vs. displacement plot) TYPE 3 LOADING ONLY Mathison et al., 2010 [80] Proximal humerus Human cadaver humeri Fracture: Two-part fracture, 10 mm wedge-shaped osteotomy at level of Recorded force, relative locking plate surgical neck movement between the fracture (Synthes) surfaces (using digital image Static Test: Shaft displaced in medial direction (varus bending) at 4 mm/min correlation), determined until failure stiffness and failure load Failure: Humeral head breaking apart or shaft snapping Chow et al., 2012 [79] Locking Human cadaver humeri Fracture: Two-part fracture, 1 cm wedge-shaped osteotomy at the level of Recorded number of cycles, compression plate surgical neck, proximal osteotomy cut made transversely at level of initial load position, position at (Synthes) inferomedial margin of the articular surface peak load application (110 N) for every cycle and displacement at Cyclic Test: N varus load at a rate sufficient to create 600 mm/min peak load until failure Determined average deformation Failure: Varus collapse or load cycles per cycle, number of cycles needed to cause 1 mm 41

42 deformation, number of cycles to failure Weeks et al., 2013 [81] Locking Same as above (Chow et Same as above (Chow et al., 2012 [79]) Same as above (Chow et al., compression plate al., 2012 [79]) 2012 [79]) (Synthes) TYPE 4 LOADING ONLY Koval et al., 1996 [67] AO T-plate Human cadaver humeri Fracture: Two-part fracture, osteotomy created at greater tuberosity base at Recorded force and (Synthes) 10 o oblique angle in medial-inferior direction displacement, determined ultimate load, stiffness Tension banding Static Test: Humeral shaft oriented at 20 o abduction and vertically displaced fixations at superior aspect of humeral head 1 cm medial from its lateral edge at 10 cm/min until failure Kirschner wires fixations Failure: Marked decrease or discontinuity in force-displacement curve AO Schanz pins fixations Ender intramedullary nails (Richards) Chudik et al., 2003 [89] ASIF T-plate Human cadaver humeri Fracture: Two-part fracture, first osteotomy non-comminuted 10 oblique at Axial Preloading: Recorded (Synthes) surgical neck fracture (type 11-A3) and second, a 10 at oblique at surgical force and displacement, neck fracture with removal of a 1 cm medially based wedge of cortex determined stiffness Experimental plate 42

43 Axial Preloading (Humeri-only): Axial compression loading up to 750 N Static Test: Recorded force and displacement, determined Static Test (Humeri with implant): Humeral shaft oriented at 20 o abduction, stiffness, stiffness ratio (static loaded at 10 cm/min until failure test stiffness/preloading stiffness), displacement after 0.3 Failure: marked decrease/discontinuity in load-displacement curve or >1 cm and 0.6 kn (physiological loads) displacement and displacement, ultimate load and energy at failure Ponce et al., 2013 [48] PHILOS plate Human cadaver humeri Fracture: Three-part fracture involving the surgical neck and greater Recorded force and (Synthes) tuberosity displacement, determined stiffness and displacement, load Static Test: Constructs fixed at 20 o from vertical, vertical load at superior and energy to failure aspect of humeral head at 10 cm/min until failure Failure: For comminuted specimen, medial cortical defect closure Gradl et al., 2013 [88] AxSOS locking plate Human cadaver humeri Fracture: Two-part fracture featuring metaphyseal comminution, 5 mm Recorded force and (Stryker) wedge osteotomy at surgical neck level displacement, determined axial stiffness and load to failure Cyclic Test 1: Construct fixed at 20 o of lateral angulation N sinusoidal vertical compressive load at 0.2 Hz for 50 cycles Cyclic Test 2: Construct fixed at 20 o of lateral angulation and sinusoidal vertical compressive displacement (6 mm/min rate, 0.2 mm maximum deflection, 0.2 Hz frequency) applied until failure. Preload kept fixed at 50 N and peak load increased 100 N every 50 cycles until failure or until 5000 N load. 43

44 Failure: Complete osteotomy gap closure, 15 o angular displacement in unloading condition, sudden decrease in recorded force Roderer et al., 2013 [83] PHILOS plate Human cadaver humeri Fracture: Three-part fracture, 10 mm horizontal gap below anatomical neck Recorded plate and humeral (Synthes) and greater tuberosity osteotomy head relative motion using ultrasound-based 3D motion Cyclic Test: Cyclic loading at 0.25 Hz at N with upper load increasing analysis system, determined at N/cycle until failure number of cycles to failure Failure: >0.5 o increase of varus angular tilting within 100 load cycles at lower magnitude (constant 15 N) Roderer et al., 2013 [82] PHILOS plate Human cadaver humeri Fracture: Two-part fracture (OTA type 11-A2), osteotomy with medially Static Test: Recorded axial (Synthes) ascending 10 mm fracture gap force and axial displacement, determined axial stiffness Static Test: Construct fixed at 25 o of lateral angulation. Axial compression loading increased at 0.02 mm/s until 200 N Cyclic Test: Determined number of cycles to failure and Cyclic Test: Construct fixed at 25 o of lateral angulation. Cyclic compression humeral head migration (using loading at 1 Hz with preload fixed at 50 N and peak load increasing at 0.05 fluoroscopic assessment) after N/cycle. Loaded until failure or >15 mm actuator displacement set 1000, 2000 and 3000 cycles Failure: 2 mm head migration Yoon et al., 2014 [85] 3.5 mm locking fixed Human cadaver humeri Fracture: Two-part fracture (AO/OTA 11-A3), first osteotomy at surgical Recorded force and angle plate neck at a 10 o oblique angle directed in a medial-inferior direction and second displacement, determined a 1 cm medially-based wedge of cortex at fracture site stiffness and ultimate load to 4.5 mm locking fixed failure angle plate Static Test: Construct fixed at 20 o of abduction and vertical compressive displacement applied at 10 cm/min until failure 44

45 Intramedullary Nail with locking screw Failure: marked decrease/discontinuity in load-displacement curve Intramedullary Nails with fixed angle blade Gillespie et al., 2009 [86] Locking Human cadaver humeri Fracture: Three-part fracture (AO 11.B2), osteotomies of surgical neck (10, Recorded force and vertical compression plate oblique) and greater tuberosity (5 mm wide) displacement (using (Synthes) extensometer), determined Cyclic Test 1: Construct fixed at 20 o of abduction and N vertical linear elastic stiffness for each 90 blade plate compression applied at 1 Hz for 200 cycles cycle (Zimmer) Cyclic Test 2: Construct fixed at 20 o of abduction and N vertical Periarticular compression at 1 Hz for 50 cycles or until failure standard proximal humerus plate Failure: Fixation loss between some aspect of implant and proximal (Zimmer) humerus Hymes et al., 2013 [87] PERI-LOC plate Human cadaver humeri Fracture: Two-part fracture (OTA 11-A3) created by removing a 1 cm Recorded displacement, number (Smith & Nephew) segment of bone below the surgical neck of cycles, number of microcracks (using acoustic emission), Cyclic Test: 500±100 N cyclic compressive loads applied 30 o posteromedial determined number of cycles to to the anteroposterior (in plane of rotator cuff pull) at 2 Hz for 15,000 cycles failure (or until failure) Failure: 20 mm or rapid (due to sudden bone fracture) actuator displacement 45

46 Erhardt et al., 2012 [49] Non-contact bridging Human cadaver humeri Fracture: Two-part fracture (OTA/AO type C) at anatomic neck, second Recorded force and plate (Zimmer) parallel cut 8 mm proximal to first displacement, determined linear elastic stiffness, number of Cyclic Test: Construct fixed at 30 o flexion and 30 o abduction and N cycles to failure and load to compressive load for 200 cycles with load increasing 100 N every 200 cycles failure until failure Failure: Visible screw perforation of at least one screw Schliemann et al., 2015 [84] CFR-PEEK DiPhos- H plate (Lima Human cadaver humeri Fracture: Three-part fracture, 10 mm gap osteotomy 10 mm below anatomical neck and second at greater tuberosity Recorded relative motion between lesser tuberosity and Corporate) Cyclic Test: Construct fixed at 25 o lateral angulation and vertical compressive load applied at 1 Hz for 10,000 cycles with load increase of N/cycle up to 400 N maximum load after 10,000 cycles shaft and proximal and distal plate aspects (3D motion capture system), determined stiffness and load to failure Static Test: Construct fixed at 25 o lateral angulation and vertical compressive displacement applied at 5 mm/min until failure Failure: Fracture gap closure, plate breakage, head fragment dislocation (assessed macroscopically) or >0.5 o increase in angular tilting within 100 load cycles at 15 N lower load Burke et al., 2014 [90] PHILOS plate Synthetic humeri Fracture: Three-part fracture Recorded movement of greater (Synthes) (Sawbones 4 th generation, composite) Cyclic Test: N compressive load applied at 1 Hz for 1000 cycles and then increased until failure tuberosity, articular surface and humeral diaphysis relative to the Failure: N/A humeral shaft at 250, 500, 750, and 1000 cycles (using optoelectronic camera system), determined load to failure TYPE 1 & 2 LOADING 46

47 Hessmann et al., 2005 PHILOS 130 o plate Human cadaver humeri Fracture: Two-part fracture (AO 11-A3), wedge osteotomy with 8 mm medial Recorded force, torque and [92] (Stratec) defect at level of the surgical neck and transverse osteotomy started displacement, determined medially, inferior to articular cartilage stiffness, load to failure, plastic PHILOS 95 o plate deformation assessed under (Stratec) Cyclic Test 1: N sinusoidal axial compression at 0.1 Hz for four preload conditions after 20, 100, cycles, then torsion at Nm for four cycles and 200 cycles AO T-plate (Synthes) Cyclic Test 2: 200 cycles of sinusoidal axial compression and torsion under same loading conditions as cyclic test 1 Intramedullary nail with locking spiral Static Test: Axial load for 45s to 500 N until failure blade (Synthes) Failure: Irreversible osteotomy gap closure in unloaded condition, 15 o irreversible angular displacement or sudden deviation from linearity on loaddisplacement curve Schumer et al., 2010 [53] S3 plate (DePuy) Human cadaver humeri Fracture: Two-part fracture (OTA type 11-A3.3), 1 cm surgical neck gap osteotomy Cyclic Test 1: ±2 Nm cyclic torsion at shaft distal end at 1 Hz with N simultaneous compressive axial load for 3000 cycles Cyclic Test 2: Cyclic test 1 repeated for torsion at ±5 Nm for additional 3000 cycles Static Test: Axial compression in axial compression at 20 N/s until failure Failure: Major drop in applied load or closed head and shaft gap with plate plastic deformation Cyclic Test 1: Recorded angular rotation, axial displacement, load and torque every 100 cycles, determined mean rotational displacement for each 3000-cycle testing period Cyclic Test 2: Determined number of cycles to failure for specimens failing before completing 3000 cycles 47

48 Static Test: Determined load at failure Dietz et al., 2012 [91] PHILOS plate Human cadaver humeri Fracture: Two-part fracture (AO 11-A3), wedge osteotomy with 10 mm Cyclic Test 1: Recorded force (Synthes) defect at humeral neck and base of resected triangle at medial side and displacement, determined stiffness (axial and torsional Retron nail (Tantum) Cyclic Test 1: 10 N axial compressive loading applied on humeri only (with stiffness) no implantation) for 15s with Nm sinusoidal torsion torque applied simultaneously for 8 cycles, then, torque set constant at 0.5 Nm with 120 N Cyclic Test 2 & 3: Determined sinusoidal axial compressive load applied simultaneously for 8 cycles the axial and torsional stiffness and percentage stiffness loss Cyclic Test 2: Same loading conditions as cyclic test 1 applied on implanted (compared to initial stiffness) humeri for 4 cycles Cyclic Test 3: Same loading conditions as cyclic test 1 applied on implanted humeri for 1000 cycles Failure: Irreversible osteotomy gap closure, peri-implant fracture, sudden deviation from linearity on load displacement curve 48

49 Maldonado et al., 2003 Locking Human cadaver humeri Fracture: Two-part fracture, 5 mm osteotomy at level of surgical neck Recorded fragment relative [93] compression plate movements (using optical (Mathys) Static Test: 0.5 mm axial compressive displacement at 5 mm/min and 4 o measurement system), torsional torque with 25 N axial compressive preload at 50 o /min determined axial and torsional stiffness Failure: N/A TYPE 2 & 3 LOADING Fuchtmeier et al., 2007 Sirus nail (Zimmer) Human cadaver humeri Fracture: Two-part (AO 11 A3), osteotomy below humeral head at level of Recorded force and [66] surgical neck to create a 5 mm defect displacement between Proximal humerus osteotomy centre and distal nail with spiral blade Cyclic Bending Test: Load of zero to a limit load (load required to create point of load transmission, 7.5 Nm bending moment at osteotomy) applied at 1 mm/min for 5 cycles determined deformation and AO T-plate angular deflection after a set (Synthes) Cyclic Torsional Test: N (8.3 Nm moment at osteotomy) applied at 1 number of cycles, bending and mm/min for 5 cycles torsional stiffness determined PHILOS plate and loss of bending and (Synthes) Failure: N/A torsional stiffness Yamamoto et al., 2013 Locking Human cadaver humeri Fracture: Two-part fracture, 10 mm wedge surgical neck gap osteotomy Recorded distal fragment vertical [99] compression plate displacement and angular (Synthes) Cyclic Bending Test: Cyclic distal shaft load with cantilever at 5 mm/s for rotations and number of cycles, 10,000 cycles, producing Nm bending moment determined maximum S3 (Depuy) displacement Cyclic Torsion Test: ±2 Nm cyclic distal shaft torsion at 20 o /s for 10,000 cycles Failure: N/A 49

50 Edwards et al., 2006 [94] Proximal humerus Human cadaver humeri Fracture: Two-part fracture of surgical neck created at greater tuberosity Cyclic Bending Test: Recorded nail (Synthes) base at 10 oblique angle, directed in a medial-inferior direction with excision force and displacement, of 10 mm bone wedge to simulate comminution determined bending stiffness, Locking mean and maximum compression plate Cyclic Bending Test: Cyclic cantilever load to produce Nm varus displacement for each (Synthes) bending moment at fracture site for 5000 cycles cycle interval and number of cycles and the maximum Cyclic Torsion Test: ±2 Nm of cyclic torsional torque to distal fragment for displacement until failure 5000 cycles Cyclic Torsion Test: Recorded Bending Test 2: 125 mm bending displacement torque and angular rotation, Torsion Test 2: Torsional load until failure determined torsional stiffness, mean and maximum angular Failure: N/A rotation for each 1000-cycle interval and number of cycles and the maximum angular rotation until failure Bending Test 2: Determined displacement and torque at maximum displacement Torsion Test: Determined torque and angle at failure 50

51 Siffri et al., 2006 [33] Locking Human cadaver humeri Fracture: Two-part fracture (OTA type 11 A3), transverse osteotomy at Bending Test: Recorded force, compression plate (Synthes) and synthetic humeri (Sawbones 2 nd generation) surgical neck Bending Test: Cyclic distal shaft load to produce Nm bending moment displacement and number of cycles Fixed angle blade for 10,000 cycles Torsion Test: Recorded number plate (Synthes) Torsion Test: Cyclic shaft load to produce ±2 Nm torsional torque for 5000 of cycles and angular rotation of cycles distal fragment with respect to fixed humeral head Failure: N/A Huff et al., 2013 [95] Locking Human cadaver humeri Fracture: Two-part fracture, transverse cut at surgical neck 5 cm distal from Cyclic Bending Test: Recorded compression plate (Synthes) and synthetic humeri (Sawbones model 1028 polyuerathane foam) highest point of humeral head with excision of 10 mm bone block Cyclic Bending Test: ±5 mm distal shaft displacement at 1 mm/s for 100 force and displacement to determine varus, valgus, flexion and extension peak load at first S3 plate (DePuy) cycles, in sagittal plane for flexion/extension bending and frontal plane for and last cycle varus/valgus bending Static Bending Test: Static Bending Test: Varus loading at 1 mm/s until failure Determined varus bending stiffness Cyclic Torsion Test: ±8 o shaft rotation at 1 o /s for 100 cycles Cyclic Torsion Test: Recorded Static Torsion Test: External loading (positive direction) at 1 o /s until failure angular position and torque, determined peak torques of first Failure: N/A and last cycles in internal and external rotation Static Torsion Test: Determined torsional stiffness 51

52 Unger et al., 2012 [98] PHILOS plate Human cadaver humeri Fracture: Three-part fracture, 10 mm horizontal gap osteotomy beneath Cyclic Bending Test: Recorded (Synthes) anatomical neck and greater tubercle osteotomy relative motion between plate and humeral head using 3D Cyclic Bending Test: N sinusoidal varus load at 0.25 Hz with upper motion analysis system, load increasing by N/cycle until screw bone fixation failure determined elastic and plastic deformation in varus angular Cyclic Torsion Test: ±0.5 Nm sinusoidal torsional torque at 0.25 Hz upper tilting load increasing by N/cycle in both directions until screw bone fixation failure, with 20 N constant axial compressive load Cyclic Torsion Test: Determined maximum humeral Failure: For bending test, >0.5 o increase of varus angular tilting within 100 head rotation for a set interval of load cycles at the lower load magnitude (constant 15 N) cycles and number of cycles and For torsion tests, >4 o axial rotation during one load cycle torque to failure Ruch et al., 2000 [97] Tension band wire Human cadaver humeri Fracture: Three-part fracture, involving humeral neck and greater tuberosity Static Bending Test: Recorded fixation with Enders with 2 mm metaphyseal bone resected during osteotomy to simulate bone force and displacement, intramedullary nail loss determined bending stiffness Modified cloverleaf Static Bending Test: Flexion, extension, varus, and valgus loading up to 5 Static Torsion Test: Recorded plate mm at 1 mm/s with 200g preload and each specimen deflected twice in each torque and rotation, determined direction torsional stiffness Intramedullary device with proximal Static Torsion Test: Internal shaft rotation at 0.5 o /s until 5 Nm torque or 30 o interlocking screws of rotation, repeated for opposite direction (Polarus) Failure: N/A Kitson et al., 2007 [68] Intramedullary nail Human cadaver humeri Fracture: Three-part fracture, creating greater and lesser tuberosity and Static Bending Test: (Austofix) head and shaft fragments Determined stiffness 52

53 Humeral locking Static Bending Test: Valgus, varus, extension and flexion loading at 1 mm/s Static Torsion Test: Recorded plate (Mathys with 2 N (200 g force) preload up to 5 mm torque and rotation, determined Medical) torsional stiffness Static Torsion Test: Torsion at 0.5 /s with 0.2 Nm preload up to 6.5 Nm (within elastic region) Static Valgus Bending to Failure Test: Determined load Static Valgus Bending to Failure Test: Valgus load until failure at failure Failure: N/A Roderer et al., 2011 [96] Non-contact bridging Human cadaver humeri Fracture: Three-part fracture (AO/ASIF 11-B1), first, osteotomy at the Recorded interfragmentary plate (Zimmer) surgical neck and second from lateral aspect of bicipital groove to just inferior motion between humeral head to the flare of the tuberosity, separating greater tuberosity from humeral head and shaft after 0, 10, 20, 30, 50, 70 and 100 cycles, determined Cyclic Test: Pure moments around three axes (x = varus/valgus, y = number of load cycles until flexion/extension, z = torsion) at 2, 3.5, 5 and 7.5 Nm at 0.5 /s for 100 cycles failure Failure: >30 o angular displacement 53

54 Kralinger et al., 2008 Humerus locking Human cadaver humeri Fracture: 5 mm slice resected at surgical neck Recorded relative movement of [111] compression plate fracture segments using (Synthes) Static Varus Bending Test: Cantilever bending with torque of 1.5 Nm to ultrasound based 3D motion PHP and HB and 0.75 Nm to IMC (simulating supraspinatus pull) analysis system Humerus Block (Synthes) Static Torsion Test: Torque along humeral shaft axis of 1.5, 0.7 and 0.45 Static Varus Bending Test: Nm for PHP, HB and IMC respectively Determined bending stiffness Intramedullary claw (ITS) Static Medial Shearing Test: Muscle pull of pectoralis major simulated, Static Torsion Test: distal fragment fixed and 30 N applied against head fragment to simulate Determined torsional stiffness pectoralis major pull Static Medial Shearing Test: Cyclic Test: Varus bending (similar to static varus bending test) at 0.5 Hz for Determined stiffness 100 cycles Failure: N/A Cyclic Test: Determined minimum and maximum force and percentage load reduction per 1000 cycles TYPE 2 & 4 LOADING Carrera et al., 2008 [101] Conventional 90 o Synthetic humeri (wood) Fracture: Two-part fracture, oblique 110 o to the diaphysis longitudinal axis Static Test 1: Recorded force angular blade plate and displacement, determined Static Test 1: Humeral shaft oriented at 20 o. Flexion load at 5 mm/min until bending strength (maximum Modified 90 o angular maximum load tolerated by plates load) and stiffness blade plate Static Test 2: Torsional load at 5 mm/min until maximum torque tolerated by Static Test 2: Determined plates torsional stiffness, maximum torsion and angular dislocation Failure: Drop or marked interruption of load associated with an abrupt slide at the point of maximum torsion of proximal segment of wooden model from its initial position 54

55 Kwon et al., 2002 [102] Cloverleaf plate Human cadaver humeri Fracture: Three-part fracture, surgical neck and greater tuberosity Recorded humeral head, shaft (Synthes) osteotomy, cancellous bone of inferior third of humeral head manually and greater tuberosity impacted in an inferior-to-superior direction with flat tip of a 0.64 cm rod to interfragmentary motion (using Angled blade plate simulate bone loss or comminution optoelectronic camera system) (Synthes) Special testing system used to perform all test. Cyclic Abduction Test: Kirschner wires with Determined head translation cement (Synthes) Cyclic Abduction Test: 30 o -120 o at 0.1 Hz (joint compressive load after cycling body weight) for 250 cycles Torsion Tests: Determined Cyclic Torsion Test: Nm cyclic external torque at 0.5 Hz with a 10 N head rotation after cycling and constant compressive load for 250 cycles torsional stiffness and load to failure Static Rotation Test: Load at 1 /s until failure Failure: For rotation tests, 30 rotational displacement TYPE 3 & 4 LOADING Lever et al., 2008 [103] AO low contact Human cadaver humeri Fracture: Two-part fracture, transverse surgical neck osteotomy 10 mm Recorded force and dynamic distal to inferomedial margin of articular surface of humeral head displacement, determined compression plate stiffness and normalised (Synthes) Initial non-destructive testing of humeri within elastic region to determine stiffness their stiffness Modified AO low contact dynamic 20 o Abduction and Flexion Test: Humeri fixed at 20 o of abduction and compression plate forward flexion, 100 N applied at superior end (Synthes) into a blade plate 55

56 AO dynamic compression blade plate (Synthes) AO, T-plate (Synthes) 90 o Abduction and Flexion Test: Humeri fixed at 90 o of flexion, 100 N applied on lesser tuberosity, in posterior direction, then humeri fixed at 90 o of abduction, 100 N applied on centre of greater tuberosity, in medial direction Failure: N/A Modified AO cloverleaf plate (Synthes) TYPE 1, 2 AND 3 LOADING Lill et al., 2003 [104] Humerus T-plate Human cadaver humeri Fracture: Two-part fracture, transverse osteotomy with 5 mm gap Recorded relative movements at (Synthes) fracture gap using optical Static Axial Compression Test: 0.5 mm axial compression measuring system Locking compression plate Static Torsion Test: 4 o axial rotation Static Tests: Recorded force (Mathys, Bettlach) and displacement, determined Static Varus Bending Test: Latero-medial (varus) bending by applying 4 compressive, torsional and Cross-screw mm compressive displacement bending stiffness osteosynthesis Cyclic Test: Varus bending for 1000 cycles Cyclic Test: Recorded moment Humerus nail with and number of cycles, spiral blade Failure: N/A determined maximum load (Synthes) during first cycle, load at 300 cycles, load reduction (%) due to migration or loosening 56

57 Duda et al., 2007 [105] Synclaw humerus nail ButtonFix PEEK plate (Synthes) throughout the 1000 cycles and the slope of the load-cycles curve between cycles 700 and 1000 Human cadaver humeri Same as above (Lill et al., 2003 [104]) Same as above (Lill et al., 2003 [104]) Humerus Block (Synthes) TYPE 1, 2 AND 4 LOADING Lescheid et al., 2010 AxSOS plate Synthetic humeri Fracture: Two-part fracture, transverse wedge osteotomies in all specimens Static Axial Test: Recorded [109] (Stryker) (Sawbones 3 rd to simulate bone loss about medial calcar region at 0 o, 10 o, and 20 o while force and displacement, generation, composite) second osteotomy created for half of specimens to remove a 10 mm determined axial stiffness segmental defect Static Torsion Test: Recorded Static Axial Test: Axial compression at 5 mm/min with 50 N preload until 0.2 torque and angle, determined mm deflection torsional stiffness Static Torsion Test: Internal rotation at 1 o /s and 0 Nm pre-torque, until 1 o Static Shear Test (LT4): Recorded force and Static Shear Test (LT4): Humeri fixed at 20 o of abduction and same displacement, determined shear procedure as static axial test until 0.75 mm maximum displacement stiffness Static Shear Strength Test (LT4): Vertical force to generate shear Static Shear Strength Test compression until catastrophic failure of the implant or the humerus at 5 (LT4): Determined shear load to mm/min with 50 N preload failure 57

58 Failure: N/A Katthagen et al., 2014 [107] PHILOS plate (Synthes) Human cadaver humeri Fracture: Two-part fracture with comminuted medial cortex and bone loss in medial calcar region, transverse wedge osteotomy with 10 mm gap at surgical neck region Recorded 3D fracture gap displacement using an ultrasound based device Static Torsion Test: Static Torsion Test: Torsional load to humeral head at 0.1 o /s until ±3.5 Nm Determined torsional stiffness Static Axial and Shear Test (LT4): Vertical load on humeral head (axial, 20 abduction, 20 adduction) at 0.1 mm/s until 200 N Static Axial and Shear Test (LT4): Determined axial and Cyclic Axial Test: N sinusoidal axial compression load at 1 Hz for 5000 cycles shear (bending) stiffness Static Load to failure: Vertical load at 0.1 mm/s until failure Failure: Implant failure, fracture gap closure (touching of medial cortices) or fracture around humeral head or shaft Cyclic Axial Test: Determined mean displacement for 5000 cycle testing period and maximum displacement after 5, 50, 100, 2500 and 5000 cycles Static Load to failure: Determined load to failure Katthagen et al., 2016 [108] PEEKPower plate (Arthrex) Human cadaver humeri First half identical to above (Katthagen et al.) but additionally performed screw perforation test Identical to above (Katthagen et al.), additionally determined load until first screw perforation Static Screw Perforation Test: Load applied at 0.05 mm/s, pressing humeral head articular fragment against screw tips until visual observation of screw perforation with exposed screw tip and/or cartilage tear Bai et al., 2014 [106] PHILOS plate (Synthes) Human cadaver humeri Fracture: 5 mm transverse wedge osteotomy on one set of specimen and 20 o humeral head collapse simulated on another set Static Axial and Shear Test (LT4): Vertical load at 5 mm/min with 50 N preload and 500 N maximum load until 5 mm maximum compression Static Torsion Test: ±5 o load at 5 o /min Static Tests: Recorded force and displacement, determined stiffness Cyclic Test: Recorded maximum and minimum displacements for each cycle, 58

59 Cyclic Axial Test: N load for 5000 cycles until 5 mm maximum deflection Failure: N/A determined displacements in 1, 10, 200, 400, 600, 800, 1000, 2000, 3000, 4000, and 5000 cycles Zhang et al., 2014 [110] Humeral locking plate (Double Medical) Synthetic humeri (Orthobone) Fracture: Two-part fracture of surgical neck Static Axial Test: Vertical displacement at 5 mm/min with 50 N preload until 0.5 mm maximum deflection Static Axial and Shear Tests(LT4): Recorded force and displacement, determined stiffness and maximum load Static Shear Test (LT4): Humerus fixed at 20 o of abduction, same test procedure as static axial testing with 1.0 mm maximum displacement Static Torsion Test: Rotation at 12 o /min with 0 Nm preload, until 5 o maximum angle Static Shear Failure Test (LT4): Humerus fixed at 20 o of abduction, loading at 5 mm/min with 50 N preload until failure Static Torsion Test: Recorded force and rotation, determined torsional stiffness and maximum torque Static Shear Failure Test (LT4): Determined shear failure load Failure: Catastrophic implant failure TYPE 2, 3 AND 4 LOADING Sanders et al., 2007 [69] Humerus locking Human cadaver humeri Fracture: Three-part fracture, intertubercular groove and surgical neck 1 cm Static Eccentric (LT4) and plate (Synthes) distal to articular surface, separating head, greater tuberosity, and shaft Force Vector Loading: Recorded force and Polaris Static Eccentric Loading (LT4): 0-80 N load 2.5 cm from shaft axis at 10 displacement, determined intramedullary nail N/s stiffness (Accumed) Force Vector Test (LT4): 80 N force vector 30 posteromedial to Static Torsion Test: Recorded anteroposterior plane at 10 N/s torque and angular displacement, determined Static Torsion Test: 0-8 o rotation at 1 /s rotational stiffness Cyclic Bending Test: 500 N (amplitude ± 100 N) load at 10 Hz for 16,000 cycles 59

60 Final Static Test: Load at 100 N/s until failure Failure: Acute decrease in load indicating loss of construct s mechanical integrity Cyclic Bending Test: Recorded number of cycles, determined peak load Final Static Test: Determined load to failure CADAVERIC TENDON-BASED TESTS Walsh et al., 2006 [112] Locking Human cadaver humeri Fracture: Two-part fracture, 7 mm osteotomy at level of surgical neck Recorded force and compression plate displacement, determined (Synthes) Simulated 30 o abduction of glenohumeral joint using custom-made testing maximum load prior to failure mechanism to pull supraspinatus, subscapularis and infraspinatus at 0.5 Cloverleaf plate mm/s (Synthes) Failure: N/A Voigt et al., 2011 [113] PHILOS plate Human cadaver humeri Fracture: Closed-wedge osteotomy, with first subcapital osteotomy at level Determined deltoid forces (Synthes) of surgical neck 5 cm distal to top of head from medial to lateral, and second necessary to elevate the arm in osteotomy at target deformity angle (20 o or 45 o ) from inferior medial to set positions, supraspinatus superior lateral efficiency and ratio of deltoid force to arm elevation angle in Simulated rotator cuff tension (glenohumeral elevation) using robot-assisted different phases of elevation shoulder simulator (RASS, KUKA Robotics) to apply 66, 26 and 61 N to subscapularis, supraspinatus and infraspinatus/teres minor, respectively Failure: N/A Voigt et al., 2009 [114] PHILOS plate Human cadaver humeri Fracture: Three-part fracture, greater tuberosity osteotomy 5 cm distal to the top of the humeral head and ventral directly lateral to bicipital groove Recorded interfragmentary motion using ultrasound based 3D motion analysis system and 60

61 Humeral suture Simulated rotator cuff tension (glenohumeral elevation) using robot-assisted measured rotator cuff strain plate (Arthrex) shoulder simulator (RASS, KUKA Robotics) with same muscles and pulling using an optical system force as above (Voigt et al., 2011 [113]) Also simulated: N Axial loading at 0 glenohumeral abduction by loading all rotator cuff muscles to their physiological tensions in neutral joint position 2. Axial loading at 60 glenohumeral abduction by loading all rotator cuff muscles to their physiological tensions in neutral joint position 3. Internal rotation at 0 abduction by loading subscapularis at its physiological tension and supraspinatus and infraspinatus/teres minor to 10 N to simulate co-contraction 4. External rotation at 0 abduction by loading infraspinatus/teres minor to their physiological tensions and subscapularis and supraspinatus to 10 N Failure: N/A Rose et al., 2010 [115] S3 plate (Depuy) Human cadaver humeri Fracture: Three-part fracture, 10 mm bone segment removed first and an Recorded relative additional osteotomy through intertubercular groove interfragmentary displacement Proximal humerus between head and shaft and locking compression Custom-made testing frame used with biaxial servohydraulic testing machine between head and greater plate (Synthes) tuberosity fragment using an SSP pull simulated by producing 10 o -60 o cyclic abduction using N optical motion tracking system, sinusoidal load at 0.33 Hz for 5000 cycles or until failure determined calcar and tuberosity displacement location and head Failure: Gross failure or loss of fixation rotation after 5000 cycles Sinatra et al., 2014 [116] Proximal humerus locking plate Human cadaver humeri Fracture: Two-part fracture, first 15 mm resection osteotomy, second osteotomy 50 mm distal to plate s distal end Recorded fracture gap distance using video recorder, determined (Synthes) Used custom-made shoulder testing setup connected to a material testing machine o shoulder abduction simulated by applying N cyclic tensile forces to supraspinatus, infraspinatus, subscapularus, and teres minor tendons at 0.5 Hz and 30 o /s for 400 cycles, while lifting 5 lbs (~2.27 kg) to simulate arm weight intercyclic motion (change in fracture gap within a single loading cycle at both 125 and 61

62 Failure: Ultimate tensile strength of construct 200 N) at 100, 200, 300, and 400 cycles Also determined load to failure and residual fracture gap deformation (difference in fracture distance at beginning and end of 400 cycles) SYNTHETIC TENDON-BASED TESTS Brunner et al., 2012 Humerus Block: Human cadaver humeri Fracture: Two-part fracture, first wedge osteotomy with 0.5 cm lateral Recorded interfragmentry motion [117] New Generation fracture gap 1 cm below medial border of anatomical neck, medial hinge later using 3D motion analysis with locking resected to create a 0.5 cm fracture gap system, degree of abduction telescoping pins using inclinometer and muscle (Synthes) Shoulder test bench used with pneumatic muscles attached to original forces, determined implant insertions of supraspinatus, deltoideus, teres major and pectoralis major migration by performing X-rays muscle before testing and after every 500 cycles 15 o -45 o abduction and adduction simulated by loading for 500 cycles. In each cycle, deltoideus and supraspinatus pneumatic muscle loads increased Determined mean maximal stepwise until 45 o humerus abduction, then teres major and pectoralis major sintering and mean per cycle pneumatic muscle loads increased stepwise until 15 o arm abduction movement of head Failure: N/A Determined mean maximum varus tilt (angulation) and per cycle varus tilt (angulation) of head 62

63 Kathrein et al., 2013 PHILOS plate Human cadaver humeri Fracture: Two-part fractures, wedge osteotomy model 1 cm below Recorded fracture gap [118] (Synthes) anatomical neck with 1 cm wedge at base, gap osteotomy model later movement using ultrasound simulated by removing medial hinge based motion analysis system, determined varus impaction and Custom-made testing setup used with pneumatic muscles simulating per cycle relative motion of supraspinatus, deltoideus, pectoralis major and teres major humeral head and plate Simulated abduction and adduction under each fracture model for 500 cycles Failure: N/A Da Graca et al., 2013 Dynamic Synthetic humeri Fracture: Four-part fracture, with fragments: head dome, lesser tuberosity, Recorded force and [119] compression plate (Nacional Ossos plastic greater tuberosity and diaphysis displacement, determined (Synthes) humeri, aluminium relative bending and torsional scapula) Custom-made testing setup based on universal testing machine used with rigidity Transosseous calf leather straps to simulate infraspinatus, supraspinatus, subscapularis sutures fixations and axillary capsular recess with cortical screws and Kirschner wires Simulated resisted shoulder abduction for 30 s with 5 N preload force, then loaded at 20 mm/min until failure, all with fixed scapula Simulated resisted shoulder internal for 30 s with 30 N preload force, then loaded at 20 mm/min until failure, all with fixed scapula and humeral shaft fixed in rotary accessory Failure: Sudden drop in applied load 63

64 Osterhoff et al., 2011 PHILOS plate Synthetic humeri Fracture: Two-part fracture, simulated metaphyseal comminution zone by Recorded fragment gap [120] (Synthes) (Synbone composite) removing 10 mm section at greater tuberosity base distance, determined intercyclic motion (fragment gap amplitude Custom-made shoulder testing device connected to universal testing within 1 loading cycle) machine using artificial tendons determined at 100, 200, 300, and 400 cycles Simulated 45 o to 60 o abduction movement by applying N to supraspinatus and deltoideus and 25 N to infraspinatus/teres minor and Determined fragment migration subscapularis at 300 mm/min for 400 cycles while lifting 3.75 kg arm weight (change in fragment gap to achieve 5 o /s humeral abduction speed distance during cyclic testing in loaded condition) and residual Failure: N/A plastic deformation (difference in fragment gap distance before and after 400 loading cycles in unloaded condition) Clavert et al., 2016 [121] PHILOS plate (Synthes) Synthetic humeri (Sawbones 4 th Fracture: Four-part fracture, humeral neck osteotomy performed as well as intertuberosity osteotomy 8 mm lateral to the bicipital groove Recorded force, displacement and strain applied on greater Aequalis nail generation) Custom-made testing setup connected to mechanical testing machine used with polyethylene rope glued to superior and lateral greater tuberosity aspects tuberosity, determined stiffness and load to failure Simulated 0 glenohumeral abduction and neutral rotation relative to scapula plane or 90 of abduction in scapula plane by translating proximal humerus inferiorly at 50 mm/s with 10 N preload Failure: N/A 64

65 2.2 Biomechanical Testing of Proximal Humerus Plates Overall, the most common loading type was torsion LT2 (n=26), followed by LT4, LT3 and LT1. Many studies involved testing of PHPs along with other devices, most frequently the intramedullary nail. As for the plates, PHILOS plates along with its variants (mainly manufactured by Synthes) were tested the most. Other plates to be tested include traditional plates such as blade-, cloverleaf- and T-plate, as well as those introduced by leading manufacturers: PERI-LOC plate (Smith and Nephew, London, UK), humeral suture plate (Arthrex, FL, USA), PEEKPower plate (Arthrex, FL, USA), Non-Contact Bridging plate (Zimmer, IN, USA), S3 plate (Zimmer Biomet, IN, USA) and AxSOS locking plate (Stryker, MI, USA). Synthetic humeri [33,90,95,101,109,110, ] were assessed in nine studies while others tested human cadaveric humeri. Only Siffri et al. [33] and Huff et al. [95] experimented on both synthetic and cadaveric humeri [33,95]. Two- and three-part fractures were created in most studies while only two studies involved four-part fractures [119,121]. A total of 10 biomechanical studies conducted testing of humeri constructs with tendons and/or musculature attached. These could be split equally according to the type of tendons used in them: cadaveric [ ] or synthetic [ ]. Loading Type 1: Axial Compression and Tension Loading Conditions The mechanically simple loading of the humerus along its shaft axis, either in tension or compression, was named LT1. Out of the 58 studies, 17 were found to conduct LT1, with the earliest study in 1998, by Instrum et al. [76]. Most studies carried out axial compression but Instrum et al. imposed tension to simulate the longitudinal distraction of humerus caused by the upper limb weight. Six studies employed LT1 in exclusion [32,50,52,75 77] as the remaining 11 studies combined it with other load types, always in combination with torsion (LT2). Chudik et al. [89] did perform axial compression but only on unplated humeri during the preloading stage of their study, while the main focus was LT4. Thus, their study was not included in this category. Only two studies, Lescheid et al. [109] and Zhang et al. [110], loaded synthetic humeri. A significant number of the studies tested a two-part fracture model, with Gradl et al. [52] being notable for testing a three-part fracture model. Static and cyclic axial loading have been exclusively achieved in six [77,93,104,105,109,110] and one [52] studies respectively. The remaining ten studies achieved both static and cyclic loading, with two studies [53,91] performing static and cyclic axial (LT1) and torsional (LT2) loadings simultaneously. 65

66 In most studies, the humeral shaft was fixed and the load was applied to the humeral head that was often potted in a polymer holder. In general, the most frequently employed loading condition for static tests was displacement-control loading at a rate of 5 mm/min which was achieved in seven studies [32,75,77,93,106,109,110]. Other loading conditions included displacement rate of 0.1 mm/s [107,108] and 20 mm/min [76] and load rates of 1 N/s [50] and 20 N/s [53]. Other than the fact that Seide et al. [32] and Bae et al. [75] followed an identical loading procedure, overall, loading conditions of cyclic tests were found to be largely varied with studies often consisting of multiple loading stages at different conditions. For six [50,53,75,76,107,108] of the nine studies that carried out both static and cyclic axial loading, static loading was performed to failure at the end to characterise the constructs load to failure behaviour. Various criteria were set to define failure, most frequent of which were the complete (or irreversible) closure of the fracture gap [53,77,91,92,107,108] and the clear deviation in linearity of the load-displacement curve [32,75,91,92]. While some studies had one failure criterion, multiple criteria were also proposed. Seide et al. [32] devised one for the static step (deviation in linearity of the load-displacement curve) and another for cyclic step (automatic stopping of the testing machine). Further two criteria relying on load-displacement curve plots were also devised. One of these considered failure to be when there was a major drop in the load [53] and this was elaborated by Zettl et al. [50] to be a greater than 30% drop in the pressure. The other described failure as humeral displacement greater than 20 [52] or 30 mm [50] on the load-displacement curve. Measurements and Data Analysis For quantitative analysis, most studies recorded the universal testing machine s actuator loads and displacements. In five studies [93,104,105,107,108], relative movements of the proximal and distal fracture fragments were recorded during tests using optical and ultrasound-based three-dimensional (3D) motion analysis systems. This was often achieved with the use of reflective markers attached on either side of the fracture gap to describe movements in terms of translations and rotations in the x-, y- and z-axes. Load-displacement data allowed determination of several parameters that could be used to compare the performance of bone-plate constructs. The most common of these was the linear 66

67 elastic stiffness of the construct, i.e. the gradient of the linear elastic region of the loaddisplacement curve. Due to the nature of their loading, the stiffness determined by Dietz et al. [91] included the combined torsional and compressional behaviour of the construct. Moreover, at the start of their tests, they loaded humeri under elastic conditions to calculate their initial stiffness. After introducing the fracture and fixating the implant, they tested the same humeri to find their second stiffness. They then reported the difference of these two stiffness values as the loss of stiffness which was represented as a percentage. Load to failure was also found from load-displacement data, often in studies with initial submaximal cyclic loading and final static loading to failure tests. Moreover, displacement at failure [50], maximum load [77] and yield load [76] was also reported in the literature. The latter was defined graphically as the peak of the load-displacement curves and in case of Instrum et al. [76], it was the tensile yield load. Gradl et al. [52] plotted load, displacement and stiffness against the number of cycles. Zettl et al. [50] created a displacement-time plot to calculate the plastic deformation after a certain number of cycles and the maximum plastic deformation. Plastic deformation after a certain number of cycles has been calculated by Hessmann et al. [92] and Bae et al. [75]. Similarly, the final plastic deformation [53] as well as the number of cycles required for constructs to fail [32,53], have also been determined for the cyclic tests. Loading Type 2: Torsion Loading Conditions The most prevalent type of loading in the literature was LT2, by applying a torsional moment on the humerus along the shaft axis. Having been conducted in 26 studies, the earliest study dates back to 2000 [97]. Only Weinstein et al. [34] and Foruria et al. [78] performed torsion tests solely, with most studies subjecting specimens to LT2 along with axial loading (LT1) or bending (LT3). In most studies, loading was imposed on cadaveric specimens, while two imposed solely [101,109,110] on synthetic and two in combination [33,95]. Also, three-part fractures were osteotomised in at least six studies [34,68,96 98,102] as the large majority examined two-part fractures. From the 26 studies, static and cyclic torsion was performed exclusively in 14 [34,68,69,93,97,101, ] and 7 studies [33,53,66,92,96,98,99], respectively, with the remaining five studies [50,91,94,95,102] achieving both. Static and cyclic loading has been achieved mostly in separate steps but in case of two studies [53,91], both static and cyclic loads were simultaneously applied. For Dietz et al. [91], static LT1 and cyclic LT2 were applied 67

68 simultaneously, and vice versa, while for Schumer et al. [53], only static LT1 and cyclic LT2 loads were simultaneously applied. The most popular setup was the direct application of torsion using a material testing machine on a holder (e.g. polymer pot) which held the humerus, with the distal fragment fixed. Three studies [33,94,99] imposed torsion on the distal fragment instead of the humeral head. Indirect loading has also been achieved via the use of cables connected to a holding construct [66,101] and by projecting devices connected parallel [68,97], and perpendicular [111], to the shaft axis. Internal and external rotations have been performed both in separation and union, from which different parameters and criteria were determined to define the behaviour of bone-plate constructs. Unlike LT4 where the work of Koval et al. [67] formed the basis of at least eight later studies, LT2 studies differed significantly from each other in terms of protocol. Perhaps the closest such case was that of the studies by Lill et al. [104] and Duda et al. [105] or the two studies by Katthagen et al. [107,108]. The most common angular displacement rate for static loading (1 o /s) was only followed by four [69,95,102,109] out of the 19 studies performing static loading. In general, for both static and cyclic loading, the studies could be separated according to the ascending order of their angular displacement rates: 5 o /min [106], 0.1 o /s [107,108], 0.5 o /s [34,68,78,96,97] and 20 o /s [99] or the displacement rates: 1 mm/min [66], 5 mm/min [101] and 12 mm/min [110]. Similarly, large varieties were found among the values and ranges of torques, angles and the duration of the tests. In case of Foruria et al. [78], rotational moments created by the subscapularis and infraspinatus muscles during shoulder elevation were simulated, based on a previous biomechanical study [122]. Random assignment of humeri to different fixation implants was commonplace in literature but a few studies involved a randomised sequence of tests. One noteworthy example of this latter case is the study by Roderer et al. [96] where, for every 100 cycles, an equal number of cycles was allocated to apply the moment in torsion, flexion/extension and varus/valgus bending of the bone plate constructs. The order of these rotations, however, was randomised. Although the studies involving torsion tests to failure were common, for most studies, separate failure criterion was not proposed for the torsion tests. From those that did, Unger et al. [98] set it to be a torsion greater than 4 o during one load cycle while for Roderer et al. [96], it was axial displacement greater than 30 o. Similar to Seide et al. [32] for LT1, Foruria et al. [78] had separate failure criteria for static and cyclic rotation. Both of these criteria were mostly qualitative but for static loading, it was an axial displacement greater than 45 o. 68

69 Measurements and Data Analysis In terms of measurements, most studies measured angular displacement from actuator as well as the actuator load but the interfragmentary motion was also recorded by nine studies [93,96,98,102,104,107,108,111] using 3D motion analysis systems. Similar to axial loading, the most frequently presented plot from these studies was the torque against displacement from which torsional stiffness was obtained. As mentioned before, the stiffness values determined by Dietz et al. [91] and Schumer et al. [53] represented the combined axial and torsional behaviour of constructs due to the simultaneous nature of their loading. Fuchtmeier et al. [66] calculated the loss of stiffness from values measured before and after a set number of load cycles. This was similar to Dietz et al. [91] who found the loss of stiffness for torsional loading. Based on the force-displacement data for each load direction, Ruch et al. [97] and Sanders et al. [69] recorded separate stiffness values for internal and external rotation and compared them for analysis. In a similar fashion, Huff et al. [95] computed the peak torque of the first and the last cycles in the internal and external rotation tests. Other parameters to be reported were torque-at-failure, angular displacement-at-failure, maximum torque, angular displacement at maximum torsion and energy at failure (area under the torque-displacement plot) [78]. For cyclic tests, based on the angular rotation and the number of cycles plot, angular displacement after a certain number of cycles and number of cycles to failure were plotted the most. Similar to the former, Hessmann et al. [92] calculated the plastic deformation after a certain number of cycles. Huff et al. [95] compared the peak torque values of the first and last cycle for each construct and calculated the difference between them. Loading Type 3: Bending Loading Conditions Loading type 3 (LT3) was the bending of the humerus, commonly by loads along either of the two axes perpendicular to its shaft axis (Fig. 6). As a result, either an extension/flexion or varus/valgus moment is often produced. In one form or another, LT3 was performed in at least 17 of the studies included in this review, originating from the earliest study by Ruch et al. [97] in With only three studies employing LT3 exclusively, most subjected specimens to LT3 in conjunction with other loading types, frequently torsion (LT2). Four studies [68,80,97,103] only achieved static tests with the remaining 13 carrying out cyclic tests. Two of the studies, Huff et al. [95] and Siffri et al. [33], included testing of synthetic 69

70 humeri as well as cadaveric ones. As for the fracture model, several studies [68,96 98] involved simulation of a three-part fracture, with the majority creating two-part fractures. Overall, there was no general protocol and there were few similarities among the loading protocols. Chow et al. [79] and Weeks et al. [81] followed identical loading protocol where cycles of N were applied at 600 mm/min until failure. Lill et al. [104] and Duda et al. [105] followed a protocol of first carrying out static varus bending by the application of 4 mm compressive displacement and then dynamic loading for 1000 cycles. Experimental setups of Ruch et al. [97] and Kitson et al. [68] were also very similar as they both performed static deflection of 5 mm at a rate of 1 mm/s in varus, valgus, flexion and extension direction but the difference was that Kitson et al. also had an additional step where the bone-plate construct was loaded to failure in valgus bending. From the 17 studies, eight subjected humeral shafts to perpendicular loads (Fig 8A), in a cantilever fashion, with the humeral head fixed [33,66,79 81,94,95,99]. To achieve the required head fixation, either an embedding material such as a resin [79 81,95,99], a low melting point metallic alloy [33] or hard gypsum [66] was used, or, in case of Edwards et al. [94], the head was held by a custom-made bone holder consisting of a tube and spiked screws. All of these studies conducted varus bending by orthogonally loading the shaft along the frontal plane. Huff et al. [95] applied valgus, extension and flexion bending in addition to varus bending. Several rationales were presented for the loading conditions used in these eight studies. In case of Mathison et al. [80], the load was transmitted 70 mm distal to the third most proximal row of plate s screw holes with the aim of replicating rotator cuff s moment during abduction. Most of the other seven studies aimed to load the humeral shaft such as to achieve a bending moment of Nm at the fracture site [33,66,79,81,94,99]. Chow et al. [79] and Weeks et al. [81] performed this on the basis of a biomechanical study by Poppen and Walker [123]. They aimed to replicate the supraspinatus forces on bone-plate constructs during the early stages of healing under shoulder immobilisation support. Mechanically, this loading is comparable to humeral immobilisation followed by a varus force acting directly at the supraspinatus insertion site. 70

71 Figure 8. Common experimental setups used in literature for applying bending loads In other eight studies, the humeral head was loaded and the shaft fixed [68,69,96 98, ,111]. Roderer et al. [96], Lever et al. [103] and Kralinger et al. [111] achieved this by fixing the humeral shaft and directly loading the humeral head in the desired direction (Fig 8B) via a biaxial material testing machine or a 3D spinal loading simulator. Lever et al. [103] loaded the humeral head in the posterior direction for flexion and in a medial direction for abduction. Four studies involved attachment of a circular plate (Fig 8C) and/or a long metal rod projected horizontally (Fig 8D) to the humeral head [69,104,105,111]. Load was applied to the plate or 71

72 the rod, at an offset distance away from the shaft axis, using a vertical machine actuator. This offset point was set along different directions to produce extension, flexion, valgus and varus bending to the constructs. Contrarily, Kitson et al. [68] and Ruch et al. [97] fixed a metal rod that projected vertically (Fig 8E), along the shaft axis, and loaded it perpendicularly at a set height above the tip of the humeral head. Four of these eight studies performed all of the four key humeral bending movements: extension, flexion, varus, and valgus [68,69,96,97]. Roderer et al. tried to replicate the peak resultant moment during several activities of daily living such as combing, setting down a 2 kg weight on a board at head height and holding a 10 kg weight, developing on the findings of a previous biomechanical study by Bergmann et al. [124]. Kralinger et al. [111] applied varus bending to reproduce the pull of the supraspinatus and medial shearing (lateral displacement of the head) to simulate the pull of the pectoralis major. Lill et al. [104] and Duda et al. [105] only performed varus bending as the former aimed to reproduce the in vivo displacement of the fracture which occurs mainly due to the tension of the supraspinatus tendon. The study by Unger et al. [98] was unique in the LT3 category in the sense that it neither involved the application of cantilever loads on the shaft nor was the humeral head loaded. Instead, humerus was loaded on the shaft to produce varus bending, with the humeral head set in a custom holder which was connected to a ball-socket joint. Multiple studies explicitly stated that the tests were carried out to failure but few stated the failure criteria. Unger et al. took failure to be the increase of angular tilting of over 0.5 o in varus within 100 load cycles at the lower load magnitudes. Moreover, failure criteria based on the varus collapse and passage of cycles during the cyclic tests was implemented by Chow et al. and Weeks et al. Measurements and Data Analysis As for the other loading types, the load and displacement of the actuator was the most frequently acquired measurement for LT4, often with the use of a universal testing machine or in case of Edwards et al. [94] and Fuchtmeier et al. [66], a force sensor or a linear variable differential transformer. In five studies, 3D motion analysis systems were used to monitor the relative movements of the fracture fragments [96,98,104,105,111]. Mathison et al. [80] used digital image correlation to not only find the relative movement of fracture surfaces but also the local strain across the surface of the specimen. To achieve this, speckling pattern was applied to the specimen surface before starting the tests, which acted as the reference point. During the course of loading, photographs of the specimen were taken which allowed the computation of the relative displacement of the speckles due to the load translations. 72

73 Force-displacement data was generally used to measure elastic stiffness and failure load of the bone-plate construct for the corresponding type of bending. In case of Fuchtmeier et al. [66] and Lever et al. [103], stiffness values were normalised based on the stiffness of the intact bone specimens. For cyclic tests, in addition to the common plot of displacement over number of cycles, Chow et al. [79] and Weeks et al. [81] worked out the gradient of this plot to find the mean displacement per cycle. Additionally, its inverse, number of cycles required to achieve one millimetre of displacement, was used as a parameter to compare constructs performance. Other parameters determined for cyclic tests included the displacement and number of cycles at a set interval or to failure or, in case of Huff et al. [95], the difference in the peak load of the first and last load cycle. Lill et al. [104] and Duda et al. [105] also calculated the load at 300 load cycles and the gradient of the load versus the number of cycles plot. Loading Type 4: Combined Bending and Axial Loading Loading Conditions A total of 21 studies conducted combined bending and axial loading of humeri, LT4, dating back to the earliest study in 1996 by Koval et al. [67]. LT4 was performed exclusively in 12 studies while the other eight studies bar one [103] employed it with one of the other two loading types, always in conjunction with torsion. Four out of the 21 studies tested synthetic humeri while the rest experimented on cadavers [90,101,109,110]. As for the fracture type, over half of the studies induced the two-part fractures and the remaining employed three-part fractures. With the exception of four studies [69,82,84,88] all studies exclusively performed either static or cyclic loading. In addition to its sheer number, the LT4 category contains biomechanical studies with a large variety of loading conditions owing to the multiple sources and rationales on which they are founded. Thus, for a more holistic view, it is necessary to understand these sources and rationales along with other aspects. All studies loaded the humeral head except Kwon et al. [102] who loaded the shaft instead. Koval et al. fixed the humeral shaft at 20 o of abduction to simulate the primarily shear loading (approximately twice the amount of shear than compression) of the bone-plate construct. This set up acted as the basis for eight biomechanical studies [48,85,86,90,101,103,109,110]. As well as 20 o abduction, Lever et al. [103] mounted the shaft at 20 o of forward flexion in a similar manner to Koval et al. 73

74 Poppen and Walker computed the force vectors at the glenohumeral joint during isometric scapular plane abduction [123]. Inspired by this study, Hymes et al. and Sanders et al. applied vertical loads to the humeral head 30 o posteromedial to the anteroposterior in the plane of rotator cuff pull. This represented the glenohumeral joint force in 0 o of abduction that occurs at the surgical neck due to rotator cuff [69,87]. Similarly, Burke et al. [90], imposed a vertical load of N on the head. This was to simulate the maximum reaction forces in the shoulder of a 72 kg average man at 90ºof isometric scapular plane abduction, adapted from the Poppen and Walker study [123]. Kwon et al. [102] loaded the humeral shaft with the head fixed and the scapulothoracic motion absent such that the rotation of the specimen from 30 to 80 approximately recreated the glenohumeral rotation that occurs through 30 to 120 shoulder abduction. The 20-50% body weight joint compressive load applied during this cyclic abduction simulated in vivo joint compressive forces described by Poppen and Walker [123]. Six studies based their loading conditions on one of the two studies by Bergmann et al. [124,125] to introduce glenohumeral contact forces measured in vivo during activities of daily living [82 84,88,107,108]. Out of these, four studies [82 84,88] fixed humeri in lateral angulation to perform varus movement, where Roderer et al. [82] and Schliemann et al. [84] tilted the shaft at an angle of 25 o while Gradl et al. [88] oriented them at 20 o. The remaining two studies were by Katthagen et al. [107,108] where loads were transmitted vertically to the humeral head with the shaft inclined at 20 in adduction, developed from the studies by Bergmann et al. and Westerhoff et al. [126]. As an attempt to evenly load the specimens, Roderer et al. [82] and Gradl et al. [88] used a polymethyl methacrylate load-cup shaped as negative of the humeral head to represent the glenoid. The former also prevented relative rotation between the cup and the humeral head by applying sandpaper strips on surfaces of the cup that were in contact with the head. Erhardt et al. [49] loaded the humeral head while the humeral shaft was set at 30 o flexion and 30 o abduction to simulate the physiological load vector of a shoulder with an intact rotator cuff during 30 o -90 o abduction. This load vector is perpendicular to the glenoid plane and generates a glenohumeral contact force of 240 N at 30 o of abduction and increases up to 582 N at 90 o abduction, as defined by Konrad et al. [127]. Failure criteria describing the failure to be the marked decrease or discontinuity in the loaddisplacement curves were frequently adapted in LT4 studies. Ponce et al. [48] set separate criterion for the comminuted and non-comminuted specimen. For the former, it was the closure of the medial cortical defect. For latter, shearing and simultaneous angulation of the humeral head was observed at increased loading, thus making it difficult to define failure. Therefore, 74

75 load to failure was taken as simply the maximum load recorded. In the study by Gradl et al. [88] complete closure of the osteotomy gap was also defined as failure as well as specimen angular displacement of 15 o in an unloaded condition. Similarly, Roderer et al. [83] defined failure as an increase of varus angular tilting greater than 0.5 o within 100 load cycles at lower magnitude (constant 15 N), determined from the data from the 3D motion analysis system. In a different study [82], they employed a criterion of humeral head migration greater than 2 mm, using the fluoroscopic assessment. Measurements and Data Analysis In addition to the recording of actuator s load and displacement, 3D motion analysis systems have also been utilised in six studies [83,84,90,102,107,108] to measure the movement of the humerus and in case of Roderer et al. [83], the relative motion between the humerus and the plate. Other direct measurements taken were the number, amplitude and distribution of microcracks formed in humeri during testing, which was made possible via the use of acoustic emission testing by Hymes et al. [87] Fluoroscopic assessment which is often conducted for qualitative analysis was used by Roderer et al. [82] to track the migration of humeral head after certain number of load cycles from the recording of the relative position of radio-opaque reference points with respect to implant. Parameters such as stiffness, ultimate load, as well as load and energy to failure, based on the load-displacement data acquired, have been of principal interest in most studies. Lever et al. [103] and Chudik et al. [89] determined normalised stiffness from the stiffness of the corresponding humeri prior to fracturing. Inspired by the work of Poppen and Walker [123], Chudik et al. also recorded displacement at 0.3 kn and 0.6 kn specifically to represent the forces on the humeral articular surface through the humeral geometrical centre during 30 and 90 arm abduction respectively. Using acoustic emission testing, Hymes et al. [87] located and recorded the number of the microcracks that were either theoretically locatable (type I) or not (type II). By combining the location information of these microcracks and the X-ray data, damage propagation was visualised in real time. From this, they plotted the number of each crack type against number of cycles. 75

76 Complex Loading Using Cadaveric Tendons Loading Conditions Unlike the previous four types of loading, this category involves performance of a wide range of movements, that are both complex and physiologically more accurate. Both the two- and three- part fracture models were tested in these studies and in terms of humeral specimen type, all studies only loaded cadaveric humeri. Two studies by Voigt et al. [113,114] involved the use of a RASS (robot-assisted shoulder simulator) along with hydraulic systems to control the pull of supraspinatus, subscapularis and infraspinatus and teres minor via brass wires sutured to the respective muscles. Both studies replicated the rotator cuff tension during glenohumeral elevation while one also recreated the axial loading at 0 o and 60 o of glenohumeral abduction as well as external rotation at 0 o abduction with the load magnitudes taken from previous in vitro biomechanical studies [128,129]. Rose et al. [130] mimicked 10 o - 60 o cyclic abduction by loading the supraspinatus, subscapularis and infraspinatus muscles for 5000 cycles or to failure, with 2.75 kg of mass affixed to distal humerus in order to approximate the mass of the upper extremity. The same three muscles were loaded by Walsh et al. [112] to represent glenohumeral abduction of 30 o. Sinatra et al. [116] used a custommade shoulder testing setup connected to a material testing machine to recreate o single plane shoulder abduction. This was achieved with the application of cyclic tensile forces to supraspinatus, infraspinatus, subscapularis, and teres minor tendons while lifting 5 lbs to simulate arm weight. This loading setup built upon the biomechanical study by Osterhoff et al. [120]. No clear failure criterion was defined in these studies, presumably due to the fact that the loading range of motion was already well-defined in terms of maximum and minimum magnitudes, deeming it unnecessary to set additional criterion. Measurements and Data Analysis All studies recorded the force-displacement data, albeit in different forms, often to calculate the load and displacements at failure or after a given number of cycles. Two studies also used 3D motion analysis systems [114,115] to measure the interfragmentary rotations and translations during the tests. Voigt et al. recorded fracture gap and determined changes in muscle insertion strain using the tracking system. In another study, Voigt et al. [113] recorded the deltoid forces necessary to elevate the arm in set positions, and determined the efficiency of supraspinatus as well as the ratio of deltoid force to arm elevation angle (N/ o ) in different phases of elevation. 76

77 Complex Loading Using Synthetic Tendons Loading Conditions Akin to the cadaveric-tendon studies, all the five studies included in this category involved the use of mechanically complex motions but with the use a variety of custom-made setups and different materials as synthetic tendons. As to the motions conducted, overall, glenohumeral abduction was performed in all the five studies. Of the reviewed literature, only two out of the 58 studies tested the four-part fracture model, both of which belonged to this category of synthetic-tendon testing [119,121]. The remaining three studies in this category experimented on constructs afflicted with two-part fractures. Three studies exclusively studied cadaveric humeri while Osterhoff et al. [120] used composite-based synthetic humeri and da Graca et al. [119] assessed plastic humeri mounted onto aluminium scapulae. Both Brunner et al. [117] and Kathrein et al. [118] used a shoulder joint test bench to perform abduction along the scapular plane and 15 o -45 o adduction. Pneumatic muscles mimicking the supraspinatus and deltoid for abduction and pectoralis major and teres major for adduction were attached to the insertions of the respective muscles using webbing straps. In the case of Brunner et al., the applied muscle forces were comparable to those calculated in an FE study by Terrier et al. [131]. Da Graca et al. simulated infraspinatus, supraspinatus and subscapularis tendons as well as an axillary recess using leather straps. Straps were glued to the insertion points of the corresponding tendons at one end while on the other end, they were drilled onto an aluminium scapula that had holes for the supraspinatus, infraspinatus and subscapular fossae. Using this custom-made setup, abduction and internal rotation to failure was carried out. In a similar fashion, Osterhoff et al. [120] used polyester webbings to represent the pull of muscles and attached them to the corresponding insertions using a cyanoacrylate adhesive for tendon-bone fixation. Pull of supraspinatus and deltoid tendon was replicated for abduction of 45 o to 60 o while lifting a 3.75 kg weight at the distal humerus. Also, to simulate the action of infraspinatus/teres minor and subscapularis, constant loads of 25 N each were applied. Similar to da Graca et al., the loading by Osterhoff et al. was cyclic, albeit lasting only 400 cycles as opposed to until failure. Clavert et al. used a custom-made testing setup connected to a mechanical testing machine and used polyethylene rope glued to superior and lateral greater tuberosity aspects to simulate 0 glenohumeral abduction and neutral rotation, relative to the scapula plane or 90 of abduction in scapular plane. Similar to the cadaveric tendon studies, fracture criteria were not explicitly stated in these studies, possibly due to similar reasons, with 77

78 the exception of da Graca et al. who defined failure as the sudden drop in the load applied by the universal testing machine. Measurements and Data Analysis In terms of measurements, Brunner et al. [117] recorded the maximum resulting forces on the glenoid as well as of the individual muscles. Using a 3D motion analysis system, fracture gap motion along the shaft axis and the maximum varus tilt of the humeral head was recorded for each load cycle. Additionally, mean maximal sintering and per cycle movement of the humeral head along the humeral shaft were plotted against the number of cycles as well as mean maximum varus tilt and per cycle varus tilt of the humeral head. X-ray scans were also performed before testing and after every 500 cycles to determine the changes in the length of each telescoping pin of the Humerus Block implant, as well as the distance between the pins tips and the humeral head cortex. Kathrein et al. [118] reported the median maximum resulting forces on the glenoid and of the individual muscles. With the aid of 3D motion analysis system, the relative motion of the humeral head and the plate and the change at the minimum value of abduction (varus impaction) were recorded for each load cycle. Both parameters were plotted against the number of cycles for data analysis. While da Graca et al. computed both bending and torsional stiffness from the force-displacement data tests, Clavert et al. acquired this data as well as the strain applied to the greater tuberosity and determined not only the stiffness but also the load to failure. Osterhoff et al. utilised inductive sensor system to record fracture gap distance during the tests. Based on this data, they determined the intercyclic motion at a set number of cycles as well as the fragment migration and the change in the fracture gap distance. 78

79 2.3 Comparison of Plate Technologies and Techniques The basis for most studies has been to investigate the technologies and techniques related to plate-based management of proximal humerus fractures. These include the investigation of locking and non-locking screw technology, polyaxial and monoaxial locking screws, rigid and semi-rigid implants, the importance of calcar region and cement augmentation of humerus. As to the design of individual plates, S3 locking plate was introduced to address some of the clinical implications associated with locking plates before it, thus it is essential to review it too. It should be noted that PHPs have been biomechanically compared with several other nonplate treatments, the most common of which is intramedullary nail [66 69,78,85,91,92,94,97,104,111,121]. The focus of this review, however, in the plate-based fixation so the results pertaining to other treatments will not be discussed here. Locking vs. Non-Locking Screws Development of the locking screw technology is one of the major milestones in the management of proximal humerus fractures. Locking screws have threaded heads that lock into the plate s screw holes to create an angular stable fixation. While the conventional nonlocking screws rely on the bone-plate interface for stability, locking screws are reliant on the bone-screw interface instead, resulting in theoretically lower friction and less dissection and stripping of the surrounding soft tissue. The failure mode of locking plates also differs from that of conventional non-locking ones. Non-locking plates typically fail in series due to the toggling, loosening or the pulling out of the screws whereas the failure of locking plates demands simultaneous pull-out or failing of all screws. As a result, locking plates exhibit superior pull-out strength and stiffness as these properties are related to the construct in entirety and not to individual screws [30]. This does prove advantageous for small to moderate loading range but catastrophic under high impact forces. General literature of proximal humerus fractures is laden with the use of locking plates but the most frequently experimented plate employing this technology is the PHILOS plate, which has been tested against plates such as Non-Contact Bridging plate (NCB) [50], humeral suture plate [114], AO T-plate [66,92] and Tele Screw plate [52] as well as proximal humerus nails [66,78,91,92,121]. The theoretical advantages of locking plates are supported by Seide et al. [32] and Walsh et al. [112]. The former demonstrated superior elastic stiffness and better fatigue behaviour for the TIFIX locking plate under axial compression as compared to the non-locking version of the same plate. Similarly, Walsh et al. recorded higher maximum load to failure for constructs treated with Synthes locking plate than those with non-locking cloverleaf plates in cadaveric shoulders during 30 o glenohumeral abduction. 79

80 Traditional blade plates have used non-locking screws and have been tested, often as representative of the non-locking plate category. Weinstein et al. [34] showed that locking plates exhibit significantly larger stiffness than blade plates in cyclic external rotation. Siffri et al. [33] also reported that in cadaveric specimens, in comparison to blade plate constructs, locking plate constructs had significantly greater torsional stability. Statistically similar stability, however, was recorded in cantilever bending for the two construct groups. Kwon et al. [102] loaded humeri that had been treated with either the cloverleaf or the blade plate, both of which were non-locking, under 30 o -120 o cyclic abduction and rotation and reported no significant differences between the performance of the two construct groups. Similarly, Gillespie et al. [86] loaded locking plates, standard non-locking plates and nonlocking blade plate constructs in 20 o of abduction and demonstrated that the blade plate constructs exhibited greater stiffness than locking plate while the locking plate was stiffer than the standard non-locking plate. These differences in the mean stiffness among the three constructs, however, were not statistically significant. Polyaxial vs. Monoaxial Locking Screws Clinical studies for locking plates, in particular the PHILOS plate, report a significant number of complications due to the perforation of screws through the humeral head. One potential solution is to use polyaxial screws in them. This has been named the second generation locking technology as it allows the screws to be directed to the region of interest before locking it, as opposed to the conventional locking systems where screw angles are pre-defined and therefore, monoaxial. One plate employing this strategy is the Non-Contact Bridging plate (NCB) biomechanical performance of which has been tested in three studies [49,50,96]. Zettl et al. [50] demonstrated statistically similar performance between the NCB plate and PHILOS plate under axial compression despite using fewer and thicker screws for NCB plate. However, Erhardt et al. [49] revealed that during simulated 30 o flexion and 30 o abduction, insertion of polyaxial screws instead of monoaxial ones had no significant effect on the perforation of screws. Importance of Calcar Region The importance of recreation and mechanical support of the humeral medial column is emphasised in clinical literature for construct stability [41] and ergo for the reduction of complications such as screw perforation of the articular surface and varus collapse. An in vitro biomechanical study by Lescheid et al. [108] also supports this as they reported higher axial, 80

81 torsional and shear stiffness for constructs with medial cortical contact, as compared to constructs with the cortex removed. One approach to provide this mechanical support is by placing screws across the medial calcar region. In vivo study by Zhang et al. revealed a low incidence of varus collapse for three- and four-part fractures treated with medial support screws that were inserted into the medio-inferior region of the humeral head [47]. However, no significant advantage was achieved in patients afflicted with two-part fractures. Katthagen et al. [107] performed in vitro axial loading, torsion and bending of cadaveric humeri with two-part fractures that had been treated with PHILOS locking plate. No significant difference was detected with the insertion of calcar screws. In contrast, Zhang et al. [110] reported higher axial and shear stiffness for synthetic humeri with two-part fractures treated with medial support screws than those without medial support screws. Also, Erhardt et al. [49] achieved increased resistance to screw perforation during flexion and abduction due to the insertion of the inferomedial support screw. A cadaveric study of three-part fractures by Ponce et al. [48] reports significantly higher mean load to failure and mean energy to failure with the use of calcar screws in PHILOS locking plate during varus collapse tests. Similarly, according to Burke et al. [90], insertion of inferomedial screws in PHILOS plate led to significantly lower mean interfragmentary motion and increased the load to failure in humeri with three-part fracture. With the aim of providing the required mechanical support (compressive strength) to the medial column, Gardner et al. [46] devised the use of fibular allografts to be placed into the intramedullary canal of the proximal humerus. In biomechanical studies, constructs with fibular allograft augmentation (PHP with allograft) exhibited better stability than non-augmented constructs (PHP only) under axial compression [75], varus bending [79,80] and 45 o -60 o simulated glenohumeral abduction [120], with higher stiffness and failure loads. Similarly, Katthagen et al. [107] showed superior performance with the use of femoral head allografts. Rigid vs. Semi-rigid Plates Most of the complications associated with PHPs go back to the issue of poor implant anchorage, particularly in the elderly. A histomorphometric study by Hepp et al. [132] demonstrated that current implants tend to target the central region of the humeral head where bone stock and bone quality are poor and the medial and dorsal aspects of the head should 81

82 be targeted instead. Maldonado et al. [93] showed that in patients with osteoporosis, higher strain forces occur at the implant/bone interface compared to patients with healthy bone. This may lead to early failure of relatively stiff constructs like angular stable plates or nails [93] especially in patients with reduced bone quality. A number of patients admitted with proximal humerus fractures have good bone quality and with this patient population in mind, rigid implants were designed, that prevented micromotion of fracture to provide maximum construct stability. For the geriatric or the osteoporotic patient population, these rigid implants have higher risks of failure due to the poor bone-implant interface. Thus, a new series of implants, named semi-rigid, were designed. This design aims to increase the energy absorption by the implant and reduce the forces acting on the bone-implant interface by allowing some fracture motion. Controversy arises on the matter of defining optimum stiffness of the implant. An excessively rigid implant poses a risk of developing extremely high peak stresses on the humerus which is not only mechanically but also biologically unsafe. On the other hand, implants with too low a stiffness can lead to early failure and head migration due to poor mechanical support [133]. This dilemma is further complicated by the fact that the mechanical role of an implant changes during the fracture healing process and an intricate balance of implant elasticity is required for successful healing [134,135]. Two of the semi-rigid implants for proximal humerus fractures are the Humerus Block and ButtonFix, intended to be minimally invasive fixations as they are accompanied by Kirchner wires and have dimensions smaller than conventional plates. Concerns have been raised regarding their elastic design which may be too elastic for the required healing and the potential risk of K-wires migration [136,137]. As for their in vitro biomechanical performance, Duda et al. [105] reported higher compression, torsion and varus bending stiffness values for humeri treated with ButtonFix system as compared to those treated with Humerus Block system. However, Kralinger et al. [111] evaluated the performance of locking compression plate with Humerus Block in cadaveric humeri in bending and torsion and reported the locking plate construct to be stiffer than the Humerus Block construct in all loading conditions. 82

83 Cement Augmentation Implant-related complications associated with PHPs, such as poor screw purchase, owe mostly to poor bone mineral density. One possible way to enhance implant anchorage in reduced bone stock is to increase the bone-implant interface by augmentation using bone cement. This method has already been established in other fractures of the human body including femur, tibial plateau and distal radius fractures [ ]. Several in vitro biomechanical studies sought to investigate the effect of implant augmentation on the management of proximal humerus fractures [83,84,88,93,98,102,118], all using either a calcium phosphate or a polymethyl methacrylate (PMMA) based cement. Gradl et al. [88] used self-setting calcium phosphate cement into all head screw holes of the AxSOS locking plate. Significantly higher load to failure and stiffness were exhibited by cement-augmented specimens as compared to non-augmented specimens in cadaveric twopart fracture model. Similarly, Kwon et al. [102] demonstrated significantly higher failure torque and torsional stiffness as well as reduced interfragmentary motion for calcium phosphate cement-augmented cloverleaf and blade plate specimens, relative to the non-augmented specimens, using cadaveric three-part fracture model. Kathrein et al. [118] augmented the four proximal screws of PHILOS plate with PMMA cement and demonstrated decreased per cycle motion and varus impaction of the humeral head during simulated cyclic abduction and adduction for 500 cycles in unstable two-part fractures. Roderer et al. [83] performed a mechanical assessment of local bone quality in the screws directions before augmenting two anteriorly directed screws of PHILOS plate with PMMA bone cement to aim at the regions of lowest bone quality. Augmentation was found to significantly increase the number of load cycles to failure under varus bending, using a threepart fracture model. Both Schliemann et al. [84] and Unger et al. [98] augmented the two anteriorly directed head screws with PMMA cement in DiPhis-H and PHILOS plate respectively. Schliemann et al. tested cadaveric humeri, treated for unstable three-part fractures, under varus bending and reported no significant increase in stiffness and failure loads but significant reduction in the bone-implant interface motion with augmentation. Unger et al. achieved a significantly higher number of load cycles until failure for the augmented group than the non-augmented group under cyclic varus bending and torsion in three-part fractures. 83

84 Spatial Subchondral Support Plate As opposed to the PHILOS locking plate which is positioned higher on the greater tuberosity to form a neck angle that is almost a right angle, the S3 locking plate is placed 3 cm distal to the greater tuberosity to achieve a 135 o neck angle. This placement is such as to overcome the potential post-operative complication of sub-acromial impingement, one of the leading post-operative complications with PHILOS plate [142,143]. Huff et al. [95] performed an in vitro biomechanical comparison of two-part fractures treated with an S3 plate and the Synthes locking plate and recorded higher stiffness for the S3 plate in torsion and varus and valgus bending tests but lower stiffness in extension and flexion bending, despite using a longer Synthes plate. Rose et al. [130], however, loaded constructs that been treated for three-part fractures under simulated 10 o -60 o cyclic abduction and reported that the specimens stabilised with the S3 plate showed significantly higher displacement of greater tuberosity fragment and larger rotation of the head fragment than those repaired with conventional locking compression plate. The S3 plate allows the use of smooth pegs rather than threaded screws for subchondral support, which has several theoretical advantages. Smooth pegs offer thicker core diameter for increased strength compared with screws and reduce the risk of articular penetration from humeral head collapse. A biomechanical study by Schumer et al. [53] was focused on this relationship and they detected no significant difference between smooth pegs and threaded screws in S3 plates for humeral head under cyclic compression and torsion in an unstable two-part fracture. Yamamoto et al. [99], however, recorded significantly less distal fragment displacement for cadaveric humeri treated with an S3 plate with smooth pegs than those treated with Synthes locking plate with threaded pegs. However, no significant difference was found for the two fixation methods in torsion tests. Because the authors used two different plates, it was difficult to determine whether this difference was due to the use of smooth pegs or the different screw orientation of the two plates or the more distal placement of the S3 plate. 84

85 2.4 Discussion In Vitro Biomechanical Testing In vitro studies of locking PHPs revealed their superior biomechanical performance over nonlocking ones. Clinical trials, however, show a different picture, one that is laden with cases of post-operative complications. There is thus a strong incentive for devising in vitro biomechanical studies that represent the in vivo situation more accurately, allowing one to foresee and prepare for the potential risk of failure. The popularity of varus bending in LT3 studies helps one to evaluate plates functionality for risk of varus collapse which is a leading complication associated with PHPs. Another common complication is the perforation of plates screws through the humeral articular surface, for which dedicated tests, such as those conduct by Katthagen et al. [108] recently, are promising. Such a problem-based approach ought to be followed to design new tests for other complications. However, due to the complex nature of the fractures, these complications do not occur in isolation but are interlinked. For example, a poor implant-bone interface is commonly reported to be the reason behind most of PHPs post-operative complications. Thus, tests that not only assess the quality of this interface but also quantitatively define it, could potentially serve as a standard for predicting in vivo performance. For accurate recreation of the in vivo situation, it is crucial to use bio-realistic humerus specimens. A vast majority of studies used cadaveric specimens that theoretically present more accurate material properties than the synthetic ones which were used only in at least nine studies. However, in general, biological variability has been known to play a significant role in results, thus making it difficult to develop correlations and draw conclusions. Also, the chronological deterioration of the mechanical properties of cadavers is a well-documented phenomenon and need to be considered. Cartner et al. [144] investigated the effect of the post-freezing delay of fresh-frozen cadaveric femora on the pull-out strength of the implanted screw. Results showed that delaying the test for 50 hours lead to a 9% drop in the pull-out strength relative to the control specimen which was tested after 16 hours. Delaying the tests for 90 minutes resulted in approximately 50% decrease in the screw pull-out strength as compared to the control. It is therefore important to ensure that the results from the cadaveric specimens are not significantly influenced by this phenomenon. Furthermore, in order to characterise the typical elderly patient more accurately, the use of osteoporotic surrogate bone specimens for biomechanical tests, should be considered. 85

86 Despite the popularity of cadaveric humeri, over 80% of all studies involved testing of humerus only, without any musculature or tendons attached, and performed one of the four loading types. The nature of these loading types was, to a large extent, mechanically very basic as they were based on simple axial loading, torsion and bending. Demand for more complex and physiologically accurate loading conditions dates back to pre-2000 studies [67,76] and since then, several steps have been taken to achieve them. First of all, cyclic loading, which is more representative of the in vivo conditions than static loading, was found to be a common mode of loading in the literature. In case of Schumer et al. [53] and Dietz et al. [91], both the static and cyclic axial compression and torsion were imposed simultaneously. Simultaneous loading of the specimen with other loading types such as bending ought to be explored because real-life shoulder movements are a combination of these fundamental loading types. Furthermore, almost half of the studies involved multiple loading types with several using the same humeral specimen for all loading stages, again to make the tests as close to in vivo conditions as possible. It should be noted that despite conducting multiple types of loading, the studies were largely limited to the four basic loading types. Many studies took the step of applying physiological load values and angles that had been determined in previous studies. In particular, a large number of the studies involving combined bending and axial loading followed this approach as they applied load 20 o away from the shaft axis. This was because it had been found in the works by the likes of Inman et al. [145] and Poppen and Walker [123] that loading at this angle emulates the maximal axial and shear load to humerus during a movement similar to early active abduction. On the other hand, there was a noticeable increase in the complexity and variety of loading conditions for the humerus-tendon testing studies despite the fact that there were only 10 of them. Prime examples of this were the studies by Voigt et al. where RASS (robot-assisted shoulder simulator) and hydraulic systems were used to control the pull of several muscles, loading conditions which had been defined in previous studies such as those by Klages et al. [146] and Kedgley et al. [147]. Similar systems have also been used by Walsh et al. [112] and Osterhoff et al. [120]. These studies had the clear advantage of pulling the intact tendons often at the anatomical insertion points, as found in vivo. The manifest advantage of the humeral-tendon studies systems over the humerus-only studies loading is evident but many improvements are required to ensure that the systems are representing physiologically accurate conditions. For example, attempts should first be 86

87 made to fully understand the in vivo loading conditions of the humerus after fracture, not only before it. This includes taking into consideration the post-fracture scapulohumeral rhythm and its effects on the humerus, contribution of individual muscles and the changes in magnitudes and directions of glenohumeral contact forces during everyday movements. As for the glenohumeral contact forces, there is a notable example of their first in vivo measurements which were conducted by Bergmann et al. [124]. Authors implanted telemetered shoulder implants on a patient with arthrosis for the measurements of the postimplantation contact forces during activities of daily living and this formed the basis for several studies that have been included in this literature review. Studies similar to that by Bergmann et al. [124], but on post proximal humerus fracture fixation scenarios, will be highly valuable as they will provide us with the required loading conditions to aim for when designing future test protocols. These studies should be conducted for a larger number of patients than those reported and from various social and medical backgrounds. The number of movements performed should also be increased and varied according to patients lifestyles. Doing so will provide us with data that can be used to simulate and verify different loading scenarios in vitro. It will also allow us to test humeri under pseudo-subject-specific loadings, based on humerus anthropometric data such as its geometry and dimensions, leading to loading conditions that are closer to cadaver donor s own loading conditions. To the best of the author s knowledge, studies basing the loading of humeri on their properties such as geometry and dimension are yet to be found, at least in the literature included in this review. This is mostly because such properties have intentionally been kept constant by selecting similar humeri, to ensure a fair test. After excluding the implants that are modified or are the same product but with different screw types and arrangements, it is found that the vast majority of the studies involved testing of two or three plates. This also, of course, does not include other implants tested alongside plates, which has been the case for many studies. Testing more plates under the same conditions, like Lever et al. [103] would ease the comparison of their performance. As far as the fracture models are concerned, most in vitro studies cover the common two- and three-part fractures, with a few studies even introducing four-part fractures [119,121]. However, the fracture patterns often varied and were according to different classification systems. There were but few studies simulating more than one fracture model, notable examples of which are the works of Brunner et al. [117] and Kathrein et al. [118]. Therefore, future biomechanical studies should include a variety of fracture patterns as well as plates. 87

88 It is evident from the results that the most frequently computed parameter was the stiffness, processed from the load-displacement data. Other than the universal testing machine, the use of a 3D motion analysis system is found to be common in the studies. They were often used to determine the 3D translations and rotations of the fracture fragments. Digital image correlation is a common technique employed in mechanical testing to measure surface strains on the specimens. For the assessment of PHPs, however, it was used in only one study [80]. Additionally, in order to quantify the development of micro-cracks on the specimens, Hymes et al. [87] used acoustic emission testing, a technique popular in civil engineering applications. Current Technologies and Techniques Many questions remain unanswered with regard to technologies and techniques relating to PHPs. For example, results from the biomechanical studies tend to favour the insertion of calcar region support. However, a detailed investigation of the importance of support to the calcar region in comparison with other cephalic regions for construct stability is required. This will also highlight the mechanical importance of different areas and screws of the plates and thus could guide the design process of novel PHPs. Such an investigation, particularly if conducted for multiple complex fracture types is also of high clinical value as it could support clinicians in making pre-operative decisions. The mechanical advantages of employing allografts in open reduction internal fixation are supported in several studies, under a variety of loading conditions. However, these studies on the applications of allografts are limited to two-part fracture models. Thus, their effects and implications under complex fractures such as three-part fractures also ought to be explored. Current testing of blade plates is limited to the traditional non-locking versions. With the recent advent of new hybrid locking blade plates such as the Equinoxe Fx (manufactured by Exactech, FL, USA) that are locking and include the option of blade insertion, a biomechanical comparison of their performance against other leading plates was not found in the literature. In response to the clinical problems reported for locking plates, several plates have been designed, based on new technologies. One such plate is the NCB plate which relies on polyaxial screw systems for support. Biomechanical studies, however, show varying results and thus further investigation is also required on their efficacy in comparison with the traditional monoaxial screw systems. In a similar spirit, the S3 plate was designed with complications such as sub-acromial impingement and screw perforation in mind. So, the plate was designed to be placed more distally than most other locking plates like PHILOS plate but 88

89 also with the option of using smooth pegs. The in vitro results on the latter, in comparison with threaded pegs, remain unclear. Future of Orthopaedic Implant Design The future of orthopaedic implant design is bright, particularly with recent breakthroughs in additive manufacturing and rapid prototyping. 3D printing of orthopaedic implants using common medical-grade materials such as stainless steel (316L), titanium alloy (Ti6Al4V) and cobalt-chrome (CoCr) using techniques such as selective laser melting and electron beam melting is still under development [148]. Coupled with reverse-engineering technologies such as 3D scanning (laser scanners, computerised tomography, magnetic resonance imaging etc.), it will allow one to 3D print custom-based geometries with higher complexities before biomechanically testing them. In silico optimisation of implant design for biomechanical performance is popular in the biomechanics literature, particularly with the recent advances in computing power. A rational and systematic approach is commonly employed to tackle such optimisation problems. After clarifying objectives of the problem, parameters of interest (e.g. implant dimensions such height, width and diameter) are identified. Constraints and requirements are also identified so that the upper and lower bound of these parameters can be set. At this stage 3D geometry of implant is constructed either by reverse engineering with the aid of a 3D scanner or if the exact dimensions are available, they can be constructed from scratch using a computer-aided design software. Based on the nature of the problem, a computer-aided engineering analysis such as FE analysis is then selected for in silico mechanical testing. This testing is non-destructive so it is both cost- and time-effective as compared to physical tests. Ideally, the final solution to the optimisation problem should be reached after performing a computer-aided engineering analysis on all possible designs (i.e. brute force) within the confines of the design constraints and requirements. However, due to the complexity of the problem and the number of parameters involved, this is often impractical. In these situations, one has to resort to numerical optimisation algorithms. The general aim of these is to guide the user with the selection of parameters to analyse, with the hope of converging to a global solution to the problem instead of a local one. This approach is much more efficient than the brute force method. Literature is laden with the use of numerical optimisation algorithms such 89

90 as genetic algorithm, simulated annealing, particle swarm optimisation and Nelder-Mead method for a variety of design problems including orthopaedic implant design [149]. 90

91 2.5 Conclusion In vitro testing of proximal humerus plates is crucial to the design process. Systematic review of 58 biomechanical studies performing these tests has highlighted several of their limitations and advantages. Different aspects of the studies, including the fracture patterns simulated, plates used, nature of loading, loading values, failure criteria, measurement techniques and the parameters to assess performance, were analysed. Forty-eight of these studies could be divided up based on the four relatively simple types of mechanical loading: (1) axial loading, (2) torsion, (3) bending and (4) combined bending and axial loading. All of these loading types have been applied statically and dynamically as well as in isolation and in combined forms. In ten studies, loads were applied to tendons that had been attached to their corresponding insertion regions on the humerus, with the aim of simulating in vivo conditions more accurately. Both synthetic and cadaveric tendons have been used to create the movements such as glenohumeral abduction. Torsion was the most common type of loading. As for the type of fracture model tested, there is a paucity of results for four-part fractures as most studies investigated two- and three-part fractures. Beyond the standard load-displacement data from material testing machines, several unique methods of measurements have been used with the hope of better quantifying implants performance. This included the use of 3D motion analysis systems to track the translations and rotations, digital image correlation for surface strain measurements and acoustic emission testing for measurement of micro-cracks. The PHILOS plate was the most tested plate, often as a representative device for the locking plate category. Several studies were dedicated to the comparison of such locking plates against non-locking ones. In general, the former was found to be mechanically superior to the latter in vitro. However, conflicting results were reported in several studies that compared locking plates with non-locking blade plates. Similarly, results also varied across studies regarding the comparison of polyaxial locking screws against monoaxial locking screws. The importance of the calcar region for mechanical support against screw perforation and varus collapse has been highlighted in several in vivo studies. This support can be provided by means of inserting inferomedial screws, in vitro results of which ranged from statistically significant to insignificant. Alternatively, the calcar region could be supported with the use of allografts. Bone-plate constructs augmented with fibular and femoral head allografts were found to be mechanically superior to non-augmented constructs. 91

92 As for the controversy over the use of rigid and semi-rigid implants, two semi-rigid implants, Humerus Block and ButtonFix, have been tested in the literature. As compared to locking plates, constructs treated with Humerus Block exhibited lower torsional and bending stiffness. Seven studies explored the effect of cement augmentation of the bone-plate constructs mechanical performance. For this, calcium phosphate and PMMA based cement augmentation have been used. In general, the outcome of augmentation was positive, irrespective of the cement type. The S3 locking plate was designed to overcome complications associated with conventional locking plates. For example, to tackle the issue of articular penetration of the screws, the S3 plate employs smooth pegs instead of threaded screws. Two studies have been dedicated to the investigation of the mechanical advantage of these. Because of seemingly conflicting results, further investigation is needed to allow us to draw conclusions. Mixed results were also reported in studies that directly compared S3 locking plate to conventional locking plates. It is hoped that the review will aid in not only selecting the most appropriate experimental protocol for future in vitro and in silico biomechanical tests but also in the development of new ones. It is also hoped that the study of this review alongside the clinical reviews available in the literature would help evaluate the loading conditions, plates and determine how strong of indicators the in vitro parameters are of clinical performance. This ought to guide the design of better plates and studies with more accurate loading conditions and parameters and lead to their standardisation. 92

93 Chapter 3: Mechanical Testing of Proximal Humerus Plates In vitro biomechanical tests were conducted to compare the mechanical performance of three leading proximal humerus plates in current clinical use: PHILOS plate, S3 plate and Fx hybrid blade plate. These plates represent different design philosophies, so testing them allowed us to not only identify the key design parameters but also study how these parameters affect their performance. This information was useful in the validation of the FE model and the design of the FE-based optimisation study. It was decided that the mechanical effects of removing different screws of each plate on the stiffness of the humerus-plate constructs would be investigated. For the purpose of this project, this information would help us to focus our optimisation study on screws that are crucial for performance. From a clinical point of view, the decision of how many and which screws to implant or leave out is a crucial one that is frequently made by clinicians. Very little information is available in the literature with regard to the optimal number of screws for a given fracture. It was hoped that studying the mechanical effects of the presence (or absence) of screws would provide valuable insight into it. With the advantage of having the results of the systematic review available, more informed decisions could be made with regard to the design of the mechanical testing protocol. Sections 3.1 and 3.2 discuss the developmental process of the final protocol described in section Protocol Design and Feasibility Assessment Test Rig Design The literature review revealed several parameters that could be used to quantify mechanical performance. One such is the inter-fragmentary movement of humerus fragments using 3D motion analysis systems. With the availability of a six-camera motion analysis system (Vicon, Oxford, UK), in our research group s gait lab (Pariser building, University of Manchester), it was decided that it would be used for recording the translations and rotations of fracture fragments during the tests and these data would be important for the validation of the FE model, in addition to the load-displacement data from the mechanical testing machine. Since the 3D motion analysis system available was mounted and calibrated to the gait lab, it was not possible to perform the tests outside. This issue, along with the unavailability of small, portable testing machines for long-term use, meant that new test rigs had to be designed and 93

94 manufactured first, in order to proceed. To cater for a variety of loading scenarios in future, it was aimed that these test rigs would allow the performance of axial compression, torsion and cantilever bending. It was also decided that the test rigs would be motor-less and therefore require manual loading, mainly because of the financial restrictions but also to ensure the steady progression of the project. In the meantime, applications for ethical approval and biosafety hazard approval were being discussed with the department s safety advisors to gain the permission for testing cadaveric humeri. To make the best use of the cadaveric humeri and allow easy comparison of the results, it was planned that they would be tested only within the elastic range of behaviour. The rigs were, therefore, designed to perform up to at least 50 kg loading and 10 o clockwise and anticlockwise rotation. Designs of test rigs used not only in the proximal humerus plate literature but also literature relating to biomechanical testing of other parts of body were studied along with mechanical pulley systems. Several designs were proposed and after checking for their mechanical feasibility and practicality, CAD models were developed using SolidWorks (Dassault Systemes Simulia Corp, Providence, RI, USA). Since the main measurement was based on 3D motion analysis, it was planned that reflective markers would be attached to each fracture fragment to record their translations and rotations to determine interfragmentary motion. The final version of the designs proposed two test rigs, one to perform the axial compression and cantilever bending and the other axial rotation. Technical drawings for the compression test rig and the rotation test rig are in Appendix I and II, respectively. The two will now be discussed separately. Test Rig for Compression and Bending The first test rig served a dual purpose, with one mode to perform axial compression (Fig. 9) and the other for cantilever bending (Fig. 10). With dimensions of 175 x 100 x 100 cm, it consisted of a workbench which had the holder through its top to allow fixation of specimen holder in axial compression and slider in cantilever bending. Both the proximal and the distal parts of the specimen were potted and fixed inside a polymer block (Fig. 9 inset). The specimen holder contained cubic locks which could be detached when inserting or removing the specimen in the specimen holder. As required, the distal end of the specimen could be fixed by screwing the bottom of the specimen holder to the workbench base. With the distal 94

95 end fixed, the load could be applied to the humeral head by manually adding weights to the loading platform. The test rig was able to hold up 100 kg of weights. When in the cantilever bending mode, the distal end of the specimen need not be potted in a polymer block (Fig. 10). Here, the polymer block containing the humeral head was attached to a guiding block which could be moved horizontally along the slider using a handle if required (Fig. 10 inset). The load was applied by adding weights onto the loading platform, in a similar manner as in axial compression, except that here a special cantilever tip was used to apply the load onto a smaller area of the specimen surface. 95

96 LOAD Two beams to provide structural support and prevent tilting of the loading platform Track to guide the vertical movement and prevent rotation Loading platform Insertion hole for a key to prevent rotation Polymer blocks containing humeral head Workbench Humeral shaft Detachable cubic locks Specimen holder Specimen holder fixed to prevent distal movement Figure 9. Design of the axial compression test rig that allows loading of humeral head with the distal end fixed (inset) 96

97 LOAD Two beams to provide structural support and prevent tilting of the loading platform Loading platform Cantilever tip Humeral shaft Specimen holder fixed to prevent humeral head movement Workbench Guiding block Handle to allow horizontal adjustment of the specimen position Slider to control horizontal movement Figure 10. Design of the axial compression test rig set in cantilever mode to allow bending of distal humerus with the distal end fixed (inset) 97

98 Test Rig for Axial Rotation Designing a test-rig which allowed manual rotation in both clockwise and anticlockwise directions was more challenging than the first test-rig. The final pulley system design consisted of rope, a rotating wheel (blue) and four fixed pulleys (grey) and one moving pulley (red), as illustrated in Fig. 11. For clockwise rotation, one end of the rope was attached to the uppermost point on the wheel and while the other was on the lowermost. On the right side, the rope passed through two pulleys (grey) before meeting the movable pulley (red). Weights could be attached to the movable pulley to pull it downwards and thus result in a clockwise rotation (Fig. 11A). With the same mechanism but in the opposite direction, the anticlockwise rotation was achieved (Fig. 11B). Switching from clockwise to the anticlockwise mode or the vice versa was relatively easy as it only demanded repositioning of the rope to activate the right pulleys. It was decided that a polymer block containing humeral head such as that used for previous tests could be fixed to the centre of the wheel to allow its rotation. Thus, if the distal end could be fixed, the required testing condition could be achieved. A B LOAD LOAD Figure 11. Illustration showing the working principle of the pulley system designed, to allow axial rotation in both clockwise (A) and anticlockwise (B) directions with the use of a rope, wheel (blue), fixed pulleys (grey) and a moving pulley (red) onto which the weights could be hanged 98

99 Pulleys to guide rope Turning Wheel to rotate specimen and humeral head Fixed distal end Humeral shaft Shaft Axis Fixed rod around which the wheel turns ROTATION Space for hanging weights Specimen holder Rope path (for anticlockwise movement) Supports to prevent towers from falling over Figure 12. Design of the axial rotation test rig set based on the pulley system deciding, allowing axial rotation of humeral head along the shaft axis with the distal end fixed (inset) 99

100 Once the test rig designs were finalised, they were submitted to the experimental officer for approval. Several concerns were raised by the officer, with the conclusion that the manufacturing of the test rigs would take up considerable resources just to create simple movements of axial compression, axial rotation and cantilever bending. It was advised that the non-portable mechanical testing machines should be used in lieu of our proposed test rigs. Thus, the test rigs were not manufactured and the plan for using the 3D motion analysis also had to be abandoned since all other testing machines were not portable. Also, for the steady progress of the project, it was decided that synthetic humeri will be used instead of cadaveric ones since the ethical approval was expected to take much longer than initially expected. With the help of the literature review, a new set of tests for the experimental protocol were decided. Testing facilities were explored to find suitable machines so that the experiments could be designed. Several experimental tests were proposed and verified based on the capabilities and limitations of the testing machines. Appendices III and IV retrospectively present two of the major versions of the protocol proposed. These appendices provide insight into author s learning experience and although the final protocol came to be very different, if need be, these have the potential use as a foundation for future experimental work. Since synthetic humeri were to be used, the results from synthetic humeri had to be first compared against a cadaveric test and so literature was searched for cadaveric studies to compare against. For this reason, the protocols presented in Appendices III and IV both began with comparison studies. In the case of the former, studies by Wallace et al. [77] and Sanders et al. [69] were chosen to be followed and so these involved axial compression and rotation and eccentric loading. The available testing machines were only capable of uniaxial loading and thus rotation was not possible. In the literature, uniaxial machines have been used not only for axial compression and eccentric loading but also for cantilever bending and oriented loading. Therefore, for the later versions of the protocol including the one presented in Appendix IV, literature was searched for studies involving cantilever and oriented loading. Out of all the studies in the literature review, the study by Huff et al. [95] was selected due to several reasons. First, unlike other studies (e.g. Wallace et al. and Sanders et al.), their study involved testing of both cadaveric and synthetic humeri with comparable results. Secondly, it involved testing in a range of directions: extension, flexion, valgus and varus (Fig. 13). Also, the study involved testing of the S3 plate which we had available for testing, allowing for a more direct comparison of results. 100

101 Figure 13. Cantilever bending tests performed by Huff et al. involved loading in the frontal plane to achieve varus and valgus bending (a) and in the sagittal plane for extension and flexion (b). NB. Directions presented here are based on left humerus Cyclic/Static Loading After careful study of the protocol by Huff et al., there arose a key question regarding the relevance of their cyclic loading. Their protocol involved cyclic loading of the specimen at four positions, all within the pre-determined elastic range of the specimen. The issue was not the rationale behind the cyclic loading itself because the physiological loads are, after all, cyclic in nature. It was instead the number of cycles, which was limited to 100 cycles. Had their objective been to determine the fatigue behaviour of the construct, testing of up to hundreds of thousands to millions of cycles would have been more representative of the physiological range. However, achieving such a large number of cycles would come at the cost of a high loading rate to complete the test within an appropriate time frame. A possible explanation for their choice of 100 cycles is that they aimed to prepare a protocol which could be performed on both synthetic and cadaveric specimens. Since the chronological deterioration of the mechanical properties of cadavers is a well-documented phenomenon, they may have used fewer cycles to ensure that the results from the cadaveric humeri were not significantly influenced by this phenomenon and could be compared to the synthetic ones [144]. Based on the loading conditions they provided, the duration of the bending and torsional tests was calculated to be approximately just over one hour for the former and fifty-five minutes for the latter. One could counter this explanation by bringing our attention to one particular 101

102 study from by Seide et al. [32] in which cadaveric humeri were tested up to a million cycles. In fact, despite using a higher frequency of 5 Hz, the total testing time for Seide et al. was just over fifty-five and a half hours, dwarfing the duration of the tests by Huff et al. Cartner et al. [144] investigated the effect of the post-freezing delay of fresh-frozen cadaveric femora on the pull-out strength of the implanted screw. Results showed that delaying the test for fifty hours leads to a 9% drop in the pull-out strength as compared to the control specimen which was tested after sixteen hours. Furthermore, delaying the tests for ninety hours resulted in approximately 30% decrease in the screw pull-out strength as compared to the control. So, in case of Seide et al., did the longer duration itself not affect the mechanical properties? If so, how are the results and the conclusions affected by it? As an attempt to answer this, attention should be drawn to the aim of the said study. The study had been aimed to compare the performance of different fixation methods using cadaveric humeri. Unlike Huff et al., neither did they use synthetic humerus nor was the objective to compare the results to one. Thus, since all their specimens were cadaveric and were all subjected to the deterioration of mechanical properties over time (based on the article), a relative comparison can be made among the specimen, as they did, to explore the trends. Nevertheless, there is no proof that all the specimens deteriorated equally. In this matter, two particular studies, Huff et al. [95] and Siffri et al. [33], are distinguishable in the literature, as they both involve cyclic testing of cadaveric and synthetic humeri. Although Siffri et al. did test cadaveric humeri for 4,000 cycles, the total duration of their cyclic tests could not be calculated based on the limited details of loading conditions provided. Hence, no comments can be made with regard to how much effect the post-freezing time may have had on the mechanical properties of the specimen. At this stage, there were two options. The first was to load the specimen for 100 cycles as Huff et al. This had the advantage of having shorter test duration per specimen and with the assumption that the specimen has not been plastically deformed after 100 cycles, it can be loaded in multiple directions. From the results by Huff et al., one could argue that this assumption is satisfied by the reality, at least for synthetic humeri. This is because, unlike the cadaveric humeri that they tested, they reported that no significant difference was detected between the first and the last cycles peak for synthetic bone. Even then, the results should be dealt with some degree of scepticism, especially if all the tests they performed were in the same sequence e.g. extension followed by flexion, varus and valgus. It needs to be 102

103 investigated how the loading in one direction affect the results obtained for the subsequent loading direction. This issue has been addressed in the final experimental protocol (3.3). Although most of the emphasis in our previous discussions have been on the stability of the construct in the varus position, loading in the four directions would allow us to determine how the optimising of the construct for varus stability affect the stability in the other three directions. Of course, an ideal implant will have improved performance in all four directions but as is evident from the literature review, different implants have directional biases for stability. Thus, keeping the number of cycles small at 100 cycles would allow the testing in multiple directions and investigation of these biases. On the other hand, loading for 100 cycles posed several problems. As discussed before, 100 cycles were not sufficient to predict the fatigue behaviour of the specimen in vivo. Also, since cadaveric specimens were not to be used in this study, the number of cycles needn t be limited to 100 cycles and could be increased without worrying about the mechanical deterioration of specimens over time. The second option, arising from the first one, was to increase the number of cycles. Overall, this would reduce the number of different directions that could be tested for a given number of specimens because once failed, the same specimen could not be used again. As a result, more specimens would have to be bought to test in all the directions. Also, since only one type of test could be performed on each specimen, inter-specimen variability would have to be considered to ensure reliability in the results. However, because synthetic humeri were being tested, this issue was not of most importance. Had cadaveric specimens been used, biological variability would have played a significant role, as is evident from the results by Huff et al. In addition to this, if the protocol was to be used in future for cadaveric specimens, the issue of mechanical deterioration would have to be considered to allow comparison between synthetic and cadaveric specimens. One way to complete the fatigue failure tests in the allocated time was to increase the loading frequency and/or the peak magnitude for each load cycle. Then again, we are bound to the physiological limits of load frequencies and ranges. Thus, increasing either of them too much could be unnatural and far from simulating in vivo scenarios. There were also some specific constraints associated with increasing the number of cycles. For example, if the duration of the experiments was to be increased, it would have been difficult to perform them within the limited bookied time on the testing machine. Due to these restrictions, fatigue failure could not be performed and thus the first option had to be considered. To allow time for all the specimens 103

104 to be tested, it was decided that specimens would be loaded for only 10 cycles, their peak load determined and compared with those reported by Huff et al. To conduct the tests, a uniaxial testing machine (ESH Testing Machine) with both cyclic loading and displacement control capabilities was selected. Also, since the load range was too small to use the available load cells, a new tension/compression load cell (Model 41, RDP Electronics Ltd, West Midlands, UK) with loading range of 8.9 kn (2000 lbs) was ordered. To connect the new load cell to the machine, an adaptor was designed using SolidWorks, matching the screw locations and dimensions of the testing machine on one side and that of the load cell on the other. These drawings were then submitted to the lab workshop to be manufactured. Schematics of the load cell and the adaptor are presented in Appendix V (Fig. A.7 & A.8). To apply cantilever bending, humeri needed to be held at two locations: head and shaft. In the case of Huff et al., the head was fixed while the load was applied at the shaft. Since a uniaxial machine was being used, an arrangement had to be devised to allow the fixation of the humerus in the horizontal direction unlike Huff et al. who used a swivel connector for movement along multiple axes Design of Humeral Shaft Holder For holding the shaft, the diameter of the shaft at the region of load application was measured. The holder was designed (Fig. 14) with a hole with matching diameter to grip the shaft and with a screw on one side to allow its attachment to the load cell. To allow connecting and disconnecting of the specimen, the holder was split into two components which could be screwed on and off (Appendix V, Fig. A.9 for technical drawings of each component). This design also allowed loading of the humerus along anatomical direction pairs (e.g. extension/flexion) in the same plane without requiring the specimen to be unscrewed. Actually, the hole was not designed to have exactly the same diameter as the humeral shaft but instead, it was made slightly larger. This was because the humeral shaft cross-section is not fully circular and making the hole larger allowed making up of the slight differences in diameters along different directions. To ensure complete tightening, a polymeric sheath was wrapped around the shaft before placing it in the holder. 104

105 Top Component to be connected to the load cell Screws to hold the two components together Figure 14. Holder to allow transfer of load from the load testing machine to the humeral shaft There were also two additional holders to allow compression (Appendix V, Fig. A.10). These had been designed earlier on when the study Huff et al. had not been selected yet. Although they were originally intended for the compression tests on the humerus, they proved useful in the performance of earlier cantilever bending pilot studies, especially in the first pilot study where a specimen with a square cross-section was used Design of Humeral Head Holder To fix the humeral head, it was initially planned that bone cement would be used, based on the Huff et al. study and other studies from the literature review. However, this would be very expensive especially for the testing of such a large number of specimens as were planned to be tested in this project. Thus, a more cost-effective alternative had to be found. One possible option was to use epoxy resin. However, after experimenting with several resin combinations on wood pieces, the resin was found to allow a noticeable amount of motion and unable to hold the pieces firmly (Fig. 15). It was immediately obvious that the use of resin to restrain the humeral head motion will not be sufficient and thus a different approach was required. Another option was to simply clamp the humeral head. A range of clamps, particularly ones with circular grip was explored. Pipe- and hose- clamps only gripped at certain points around the humeral head so there was the issue of stress building up near the areas of grip (Fig. 16). Alternatively, rubbered grip or a repair clamp with large grip area, similar to the PVC pipe used by Huff et al. (Fig. 17), could be used. With rubber grip, there was the 105

106 issue of slippage and relative movement between the rubber and the humeral head even when the clamps were tightly screwed. A B Figure 15. Setup for experimenting with different resins, showing the wood specimen (A) and the resin cavities (B) Figure 16. Hose clamp (left) and the pipe clamp (right) [255,256] Figure 17. Pipe (left) and repair (right) clamp with rubbered grip [256,257] One possible solution was to produce a block with the negative impression of the humeral head, which would be clamped in a similar fashion to a pipe clamp. In this case, the gaps between the mould wall and the humerus surface will still have to be accounted for. At this stage, specimen holder designs from the earlier protocol proposals were also studied. Fig. 18 shows one such setup which was designed during the time the proposal in Appendix IV was being designed. It was planned to be used for oriented loading where the humeral shaft 106

107 was glued inside a cylindrical holder at an angle formed by a custom-made wedge. Likewise, wedge bases with desired sloped angles would have to be manufactured. With further research, it was found that the cylindrical holder needn t be manufactured individually. Base plates that are commonly used for scaffolding could be ordered (Fig. 19). Figure 18. Diagram showing the setup for oriented loading from one of the experimental protocols proposed earlier. N.B. to simplify the diagram, only three screws have been shown. Figure 19. A typical base plate used for scaffolding [258] Using a similar plate with a wider diameter to hold the humeral head would cause movement between the head and the plate. To fix it, resin or glue would have to be used. After experimenting with different resins on a wood specimen, their use was avoided. Even if the resin did work well with minimum movements, the sheer amount of it and the base plates for fixating all the specimens from the humeral head would have been costly. Thus, a costeffective alternative had to be found. Cement Block Preparation A different approach was to implant the humeral head into a block of construction cement instead of bone cement. Since synthetic humeri were to be used, there was confidence that cement will provide sufficient levels of reinforcement to prevent movement of the head. Several mixtures of cement and water were tested using trial and error. An ideal block would have sufficient purchase on the humerus to prevent it from moving. The ideal block should 107

108 also be able to be clamped to the testing machine base. For brevity, only the preparation of the final mixture is described here. First, a synthetic humerus was taken and marked at distances of 4, 5, 6, 8 and 21 centimetres from the apex of the humeral head. Using a clamp stand and backdrop clamps, the humeral shaft was clamped vertically, with the humeral head at the bottom and the distal end at the top. To ensure that the humerus was held vertically straight, a spirit level was attached to the shaft and the position of the spirit bubble monitored. At a separate place, a 10x10x10 cm cubic mould was obtained and its screws tightened (Fig. 20). A cloth soaked in oil was used to dampen the inner walls and the base of the mould. This was done to allow easy removal of the cement block later. An advantage of having a cubic cement block instead of a cylindrical one was that not only does it have four side faces, each of which could be made parallel to the anatomical planes (sagittal and frontal planes) but they are also easier to clamp to the testing machine using simple clamps. Figure 20. Mould used for preparation of the cement block The mould was then moved and placed directly below the vertically clamped humerus. The height of the humerus was adjusted such that the section from the head apex to the 4-cm mark was inside the mould. The humerus was then rotated until the four sides of the mould were parallel to sagittal and frontal planes. Final checks were then made to position the humerus at the centre of the mould. The setup was now ready to pour the cement into the mould. 108

109 Experiments showed that the standard, general purpose (Portland limestone) cement although was of high strength but was slow at setting. Thus, it had to be mixed with a rapid mix cement to achieve the desired setting time. A ratio of 4:1:2.5 by volume was finalised for the general-purpose cement, rapid mix cement and water, respectively. 330 ml measuring cup was filled with the general-purpose cement and emptied into a container. This was repeated so that the container had four cups full of general purpose cement and one of rapid mix cement. For homogeneity, the two were mixed with steady turning movements of a stirring rod. Two and a half cups of water were poured into the container in a stepwise fashion, using the stirring rod for mixing. Stirring was continued until a homogenous, thick paste of cement was formed, without any soft lumps in the mixture. On average, this stage took approximately five minutes. Cement was then carefully poured into the mould and the mould filled to the top. A stirring rod was also used here, albeit very gently, to prevent the potential formation of cavities inside the block. Once filled, the cement mixture was left in the mould to dry. One of the distinguishing features of the S3 plate is that unlike most other proximal humerus plates, such as the PHILOS and Fx plates, it is positioned much more distally around the humeral head. Thus, implanting PHILOS and Fx plates was challenging since, according to the aforementioned procedure, the region from the apex of the humeral head to 4 cm distal was to be inside the cement block and these two plates sit at approximately 3 cm distance from it. As a result, it was decided that after roughly thirty minutes of leaving the cement to dry, when the cement had partially solidified, a part of the region where the Fx and PHILOS plates were to be placed would be removed. An attempt was made to remove the minimum amount of cement so that the strength of the block was not affected. This region would be chipped and chiselled later on to create room for the plate (section 3.1.5). At least 48 hours were required for the cement block to be dry enough to be taken out of the mould. If removed too soon, it would break into pieces. This procedure was repeated for the preparation of the specimens in all the subsequent tests including the pilot studies Plate Choice and Implantation One of the key aspects of the mechanical testing performed in this project was the implantation of the plates onto the humeri. The three plates, namely S3, PHILOS, and Fx plates, all had similar procedures for fixation, with minor differences due to their designs. The author (Ali Jabran) underwent training by an orthopaedic surgeon (Dr Chris Peach) on the implantation of each of these devices. A summary of the general implantation procedure, as performed in 109

110 the laboratory, has been provided here. Detailed surgical procedures are available on the implant manufacturers websites [ ]. Once the cement block containing the humeral head was taken out of the mould, it was left for a few hours to for further drying. As it had been found that the cement block also covered areas of the humerus to which the Fx and PHILOS plate would attach, cement around those areas was removed. Even after solidification of the cement block, the regions had to be removed by chiselling, whilst taking care to prevent excessive removal of cement and also to avoid damaging the humerus (Fig. 21). Figure 21. Humerus in the cement block after chiselling A transverse cut was made at a distance 21 cm distal from the humeral head apex, using a junior hack-saw, to remove the shaft from the rest of the humerus. Furthermore, two transverse cuts covering half the depth of the humerus were made 5 and 6 cm from the humeral head apex. It was decided that the process of fixating the plate onto the humerus would be greatly simplified if the plates were to be attached to the bone prior to completing the fracture. The half-cut was carried out to ease the completion of the fracture at a later stage and also to prevent damage caused to the plate during the removal of the fracture piece. These steps were identical in all experiments, irrespective of the plate being implanted. The subsequent steps were slightly different for each plate. The plate was placed on the humerus at the position recommended by the manufacturer. The S3 plate was positioned more distal than the PHILOS and Fx plates. With the plate placed, an outline of the boundary of the plate was marked onto the humerus to maintain the correct position during implantation (Fig. 13). One of the common features among the three plates was that at least one of their shaft screw holes were elongated, allowing the user to make minor adjustments to the positioning of the plate once the screw has been inserted there. Thus, for all the plates, this screw hole was drilled first and the screw inserted and if needed, the position of the plate with respect to 110

111 the humerus was adjusted (Fig. 22). Once the optimal position was found, this screw was then tightened. Figure 22. Positioning of Fx plate on the humerus Next, one of the head screw holes were drilled and screwed to prevent movement of the head while handling. For this, all three plate systems provided an external attachment and guide to aid the drilling and fastening of the screws due to the different angles at which each screw penetrates into the humerus (Fig. 23). As with the lower section of the plate, the drill bit was to be made to penetrate throughout the bone until it came into contact with the cortical bone. Figure 23. Photograph showing the configuration of external attachment and drilling guide Then, the remaining shaft screw holes were drilled and the corresponding screws inserted. After securing the bottom part of the plate, the next process was to drill and insert the remaining head screws, using the same guide as previously described. For the Fx plate, an additional step was required to insert the blade and the large 6.5 mm diameter screw, both of which were inserted at the end of the procedure. The hole for the large screw was drilled using a special drill bit and the screw inserted in a similar fashion to all the other screws. For the blade, two 3.8 mm screw holes were drilled at the side ends of the blade 111

112 hole, with depth match to the length of the blade itself. An initial cavity was then made into the humerus by placing a blade osteotome over the blade slot and hammering it in using a soft faced hammer. This created a slot into which blade could then be pushed in. Once inserted, the blade was held in position by tightening two small screw heads in its shoulders. The final step in the preparation process was to remove the previously marked fracture to simulate the proximal humerus fracture. As a half cut was established towards the beginning of the preparation process, the removal of the 10-mm block of bone was achieved by cutting through the other side of the bone to meet the previous two cuts. The block of bone was gently knocked out to prevent damage to the plate or any of the screws. Following the attachment of the plate onto the humerus, a final inspection was made to ensure that all the screws have been properly fastened. This marked the end of specimen preparation stage. This procedure was followed for all the final experiments including most of the pilot studies. Any subsequent changes to this procedure have been described in the relevant sections. 3.2 Pilot Studies D Scanning Set-up and Feasibility Tests Using Wood Specimen 3D scanners capture images of the specimen surface and transform them into digital 3D geometry. The created mesh can be exported for post-processing and further analysis. Noteworthy applications of this technique include its use for anthropometric measurements, metrological studies and for the development of models to be used in FE analyses. 3D laser scanners, such as the ExaScan (Creaform, Canada) used in this project, are much more convenient for capturing the 3D geometry of specimen during mechanical testing than most traditional techniques such as Computational Tomography (CT). This is due to several reasons. First, the main safety risk associated with 3D laser scanning is the radiation from the laser. However, most lasers scanners are classified as Class 2 according to the IEC standard for the safety of laser product [153]. As a result, it is safe to use but any potential risk can be avoided by not staring directly into the beam. This is much safer than the radiation risks posed by other scanning techniques. Portability and the lightweight of the 3D laser scanners are also advantageous, allowing users to scan regions which are difficult to reach. 112

113 ExaScan (Fig. 24) possesses an auto-positioning system based on multiple retroreflective positioning targets that are randomly placed on the specimen of interest. If the specimen is large, the targets can be attached to its surfaces. However, if the specimen is small or there is a risk of losing surface definition by placing the targets onto the surface, it is best to place the targets onto a board and place the board behind the specimen. LEDs on the ExaScan indicate the scanner s proximity to the object, allowing the user to reposition it to the optimal distance. When the trigger is pressed, a laser crosshair is emitted out from the lower middle end of the scanner and projected onto the object and the positioning targets. Reflected laser beam patterns are captured by the two cameras which are located on the sides of the scanner. The triangulation principle is used by these cameras to determine the relative position of the scanner and the specimen. Real-time rendering of the 3D mesh model provides great convenience in aiming the scanner at the region of interest. Figure 24. Creaform ExaScan in action [259] One of the key limitations of using 3D laser scanners, especially ExaScan is that they are only able to scan the surface geometry and miss out the underlying processes and phenomena that take place in the bulk of the specimen. For this, CT and other techniques have obvious superiority over it. Also, as compared to other techniques, ExaScan has a much lower resolution of 50 µm, limiting its use for scanning of intricate microstructures [154]. In addition to this, it requires the specimen to be stationary when scanning, thus restricting its use for mechanical tests with moving specimen. With the availability of a 3D laser scanner, it was planned that scans of the specimen will be conducted at regular intervals when the specimen would be tested for plastic failure (final 113

114 failure tests in the Huff et al. study). A pilot study was therefore required to determine the feasibility of this and find the best procedure to achieve it. This was also an opportunity to test a specimen and determine whether the experimental protocol itself was feasible. If it was, it was important to know whether the introduction of the 3D scanning steps affected the protocol. Since this pilot study was mainly concerned with determining the feasibility of the protocol and not to analyse the recorded data, it was not necessary to use a synthetic humerus to test. Instead, a wooden column with a square cross section and length similar to humerus was used. Using the procedure described in and 3.1.5, it was fixed in a cement block. To simulate the fracture, a cut was made halfway into it. It was not fully cut through because that would require use of a fixation plate. For this study, a half-cut was sufficient. The cement block was clamped to the machine via a large G-clamp. Since the wood specimen did not have a circular cross section, one of the additional shaft holders (as described in the section and Appendix V, Fig. A.10) was used. In fact, the holder was not long enough to reach the specimen so it had to be made longer by getting an extension piece manufactured to be screwed into it. Also, since this type of holder could only compress the specimen by applying negative displacement and not tension, all the four movements (varus, valgus, extension and flexion) were conducted separately. To produce a large enough deformation, peak displacement was set to 10 mm instead of 5 mm. Displacement rate was set at 5mm/s, same as Huff et al. For 3D scanning, positioning targets were attached both to the specimen and onto a black board which was positioned right below the specimen. To allow scanning of the whole surface of the specimen, including the top and bottom faces, the cement block was raised by placing an additional block between it and the machine s base (Fig. 25). For all 3D scans, before the commencement to the scanning, two parameters of the 3D scanner, laser power and shuttle speed, were configured in order to ensure high scan resolution and effective scanning of the specimen. The auto-adjust option in the 3D scanning software VX Elements 15.0 (Creaform, Levis, QC, Canada) was used to automatically adjust these parameters to their optimum values dependent on the ambient conditions. 114

115 When performing these experiments, several challenges were faced. Firstly, the scanner was unable to scan the bottom face of the specimen because where there was a blackboard (containing positioning targets) in the background when scanning the top face, there was not a similar board when scanning the bottom face. As a result, not enough positioning targets were being detected and thus the bottom surface was missing from the scan. Figure 25. 3D scanning setup, showing the location of the position targets and the specimen One of the ideas was to place an additional blackboard above the specimen which would contain targets that would be detected when the bottom face will be scanned. This was implemented (Fig. 26), however, the problem persisted and even worsened. After placing the additional board above the specimen, the improvement in the scan quality of the bottom face was very small. It was actually counter-productive, as it obstructed the scanning view of the top face. To avoid this, the additional board would have to be placed much higher to clear the view for scanning the top face. However, this would, in turn, make it difficult to scan the bottom face because the additional board would have moved much further from it. So, overall, implementing this idea resulted with reduced the scan quality. 115

116 Figure 26. 3D scanning setup with the black board placed over the specimen After experimenting with several arrangements, a solution was found. It was understood that the inability of the scanner to scan the bottom face was due to the lack of targets when viewing from that angle. Thus, more targets were needed in that position. There was limited space on the wood specimen (or humerus for the subsequent pilot studies) so attaching more targets on it was not the best option. Also, placing the targets on the specimen would cover up the surface underneath them and will result in scans with the reduced surface definition. Thus, the first board, like before, was placed on the machine s base, directly beneath the specimen. The other board was placed vertically (either by reclining or clamping) behind the specimen to allow scanning for the front view (Fig. 27). Then, to scan the back, the same board would be picked up and placed in the front. This way, there would always be markers behind the specimen even when the bottom face was being scanned. 116

117 Figure 27. 3D scanning setup with the black board placed behind the specimen There was, however, additional complexity to this. As described before in the introduction, a limitation of using ExaScan is that when scanning, the physical distance between positioning targets and the specimen must not change. This could either be due to the movement of the specimen itself but could also be due to the movement of the markers. So, if the board containing the targets was moved, the distance between the specimen and the targets would change. The result would be a scan of multiple regions, some according to the previous references and some from the new. To solve this, upon completion of the scan of the front face, the vertical board was moved to the opposite side of the specimen. Then, the captured targets in the vertical plane were deleted from the scan using the software (Fig. 28). This encompassed not only the removal of the vertical wall from the scan but also the deletion of targets in this vertical wall. By doing this, when the scan of the other side commenced, the scanner would rescan the position of the targets, now according to their new positions. The trick here was to only delete the markers which were placed on the vertical board so that the markers on the bottom board and the cement block would not move and could serve as pristine reference points for data alignment 117

118 when switching over. In short, when one side was to be scanned, the second board would be placed on one side and the vice versa for the opposite side. Figure 28. Selection process of scanned vertical wall to be removed after the completion of the scan of one side of the specimen After performing this pilot study, there was confidence that the 3D scanning could be incorporated into the final tests. The procedure and the setup for 3D scanning were much clearer and were ready to be performed for synthetic humeri Validation of Experimental Protocol To build on the first pilot study, it was important to perform tests on a synthetic humerus. So, the aim of this study was to investigate whether results similar to those presented by Huff et al. can be obtained using our protocol. Synthetic humeri were obtained, fixed in a cement block and implanted with S3 plates as before, using the same screw choice as Huff et al. Cement block was clamped to the base of the machine and the load was applied using the shaft holder with a circular hole (Fig. 29). 118

119 Figure 29. Set-up of the specimen implanted with S3 plate and held using the circular hole shaft holder All four bending load modes were applied with +/- 5 mm of displacement per cycle. Since only the peak load for the first few cycles was of interest here, loading for fill 100 cycles was deemed unnecessary. The displacement rate was, however, kept the same at 1 mm/s. Peak loads for each loading positions were determined from the testing machine s outputted raw data and compared. Peak load values for the first cycles showed different trends to that obtained by Huff et al (Fig. 30). Where Huff et al. obtained overall higher values for extension/flexion than varus/valgus overall, the opposite was being found here. Interestingly enough, although the trend was reversed here, the magnitude of peak load for extension and flexion was actually found to be in the same range as what Huff et al. reported. It was the varus and valgus load values which were almost twice as high as them. 119

120 Figure 30. Experimental results obtained for the pilot study (blue) with Huff et al results superimposed approximately (black) The obtained results also seemed to be in contradiction with the theory. Let us assume that the humeral shaft is a perfect cylinder that has not been fractured and thus, it is not implanted with a plate. In this scenario, turning it for any of the four bending directions (extension, flexion, varus and valgus) does not affect its load since the cross section remains circular in all four cases. Now suppose that the cylinder is sliced at a distance (to simulate the fracture) and implanted with a plate. For simplicity, the shape of the plate could be considered to be a cuboid. Due to plate s rectangular cross-section and hence the difference in the area moment of inertia for the vertical and horizontal position, the peak load is theoretically higher in the extension/flexion position where the plate cross-section is in the vertical position than the varus/valgus position, according to the beam-bending theory for cantilever loading. However, the opposite was being reported and so further investigation was required. Subsequently, the identical experiment procedure was conducted but using PHILOS plate (same screw choice as Huff et al.) instead of the S3 plate. Unfortunately, yet, again, almost identical results were obtained. This time, the values for all four positions were even higher. The experiment was repeated five times to but still, the similar trend was found (Fig. 31). 120

121 Figure 31. Experimental results obtained after conducting the pilot study on PHILOS plate To investigate what was causing these results, several ideas were explored. First, it was suggested that possible the theoretical explanation provided for the peak load to be higher in the extension/flexion direction than varus/valgus was overly simplified. To study this, 3D scans of the humerus was conducted and a solid model developed. Cross-sections of the humerus were inspected at set distances starting from the fracture site and ending at the site of load application. The maximum distance in the horizontal and vertical distance was measured and labelled a and b respectively. It was found that the cross section was more elliptical than circular. The cross section was almost always longer in the varus-valgus direction than extension-flexion (Fig. 32). At the distance where the load was applied (12 cm distal to the fracture site) the load, the cross section was approximately 19% longer in the varus/valgus direction. Thus, the second moment of area would be higher for varus/valgus orientation, resulting in higher loads values in these directions. 121

122 Figure 32. Cross-section analysis of 3D humerus model 122

123 To investigate the effect of this difference in cross section, another experiment was later performed. This time, using the same experimental procedure as before, humerus was tested, without any fracture and implant so that the effect of the humerus geometry could be explored. Interestingly, the same trend was found. Peak load values for varus and valgus were still found to be higher than those in extension and flexion (Fig. 33). Figure 33. Peak load values for different positions of humerus specimen on its own, without any fracture and implantation If this trend was in fact due to the differences in the cross section, then the question was whether these differences had a greater effect on the final load values or the fact than the rectangular cross-section of the plate favoured the extension/flexion loading that the plate. At this stage, there were two possible routes. One was to perform an FE analysis using the 3D models of the humerus and the plates to simulating the scenario and obtain more, theoretically accurate values for peak loads. Another was to perform an experiment on a specimen with geometry simpler than humerus and determine whether theoretical trends can still be observed. This planned studied could also be used to recheck the experimental setup. To conclude, the results obtained from this pilot study demanded further investigation. An additional study was required to determine their cause, ideally a study on a simplified model that could be explained in theory needed to be tested in the lab Testing of Cylindrical Wood Specimen Based on the simplified theoretical model described in the previous study, an experiment was designed to find out whether the theoretical trend for a cylindrical specimen was obtained in practice, using the experimental protocol. 123

124 Instead of a humerus, a cylindrical shaft made out of wood was fixed in a cement block. To simulate the fracture, a 10-mm slice was excised off it. Next, a metallic plate, with rectangular cross section was implanted to join the two pieces of wood. Standard DIY screws were used for this purpose, two on the head and four on the shaft (Fig. 34). An identical procedure to that in the last pilot study was used and the load-displacement graph was plotted for each of the four directions. Figure 34. Experimental setup using cylindrical wood specimen and metallic plate Theoretically, the peak loads for extension and flexion ought to be higher than that for varus and valgus. However, the opposite was still observed here (Fig. 35). These results had demonstrated that the used experimental setup was not only problematic but also against the theory. Each stage of the experimental setup and the procedure was inspected very closely to find any possible sources of error. It was found that for the extension/flexion the humeral shaft could be placed into the shaft holder relatively easily whereas, for varus/valgus, the orientation of the shaft was not fully horizontal but rather slanted. As a result, the cement block, which was cubic in shape, also did not sit flat on the machine base in varus/valgus and was lifted off slightly. Even when an attempt was made to clamp it as tightly to the base as possible, there still was a tendency in the cement block to be lifted off. This was not observed for extension and flexion though. Mechanically, this tightening had introduced initial stress into the structure prior to any external loading from the testing machine, causing the peak load to be higher for varus and valgus. The results from this pilot study had put a doubt in the experimental setup. It was important, then, to conduct further studies to evaluate the results using alternative setups. 124

125 Figure 35. Peak load values for different positions after testing circular wood specimen Trial Tests on Humerus With Different Setup With the aim of evaluating alternative setups, this pilot study involved testing the specimen using one of the shaft loader which was designed to not introduce initial stress into the shaft in the varus and valgus position. The proposed loader had the shape of a semi-circular prism and so was only able to apply compression (Fig. 36). Therefore, all the four modes of bending had to be applied separately. In addition to this, to ensure that the grip of the clamp was across the whole top surface of the cement block and not just at a few specific locations, a heavy weight was placed directly on top of the cement block (Fig 37). Later, two large G-clamps were used on both sides of the weights for an even spread of grip. Due to the unavailability of a five-distal hole PHILOS plate, a 3-distal hole one was implanted, using similar screw choice as Huff et al. The experimental procedure was the same as before, using the same peak displacement of 5 mm and applying the four bending modes separately. 125

126 Figure 36. New specimen loader design with a semi-circular prism contact surface Figure 37. Experimental setup using the new specimen loader and placing a heavy weight directly above the cement block The recorded trend in the peak load values was very similar to that reported by Huff et al., with extension/flexion being stiffer than varus/valgus (Fig. 38). The magnitudes themselves were, in general, lower but this was most likely due to the fact that a shorter plate had been used in this case. 126

127 Figure 38. Peak load values for humeri implanted with PHILOS using the new experimental setup As far as the experimental work was concerned, findings from this pilot study were crucial. It was a valuable experience for the author as it highlighted the importance of the experimental setup, something which required paying attention to details. Several months of experimental time had been spent in investigating the results and having finally solved it, was an achievement. To build on this, an attempt was then made to plan out the procedure for the final failure test and incorporate 3D laser scanning into the protocol. In short, as mentioned before, the aim of the 3D scanning steps was to capture 3D scans of the specimens at regular intervals during plastic testing, to be compared later. As with any experimental work, the choices were constrained by equipment limitations and capabilities so decisions had to be made based on a pragmatic approach while keeping the ideals in mind. The first limitation was related to the testing machine (ESH uniaxial testing machine) that had so far been used for all pilot studies. The machine was only capable of cyclic loading and not static. This would be a problem if we aim to follow plastic deformation described by Huff et al. since that did require static loading. 127

128 One possible solution was to increase the duration of each load cycle to achieve an almost static loading at the desired load rate. However, the maximum displacement in each direction (amplitude) of the machine was limited to 12.5 mm and based on the load displacement graph presented by Huff et al., this was too small to determine the plastic behaviour of the specimen. Also, if the displacement was too small, then the use of the 3D scanner would be limited since there wouldn t be a large enough plastic displacement to be compared against. In addition to this, the machine did not have the option to pause and hold at the set points during the test to allow time for 3D scanning. With the hindsight of having practiced the 3D scanning and determined the duration of the 3D scanning step, it was a wise decision to not use this machine since the scanning step required approximately eight minutes. This machine simply didn t have the option to pause and hold it, unless of course if the machine on and off each time which was inconvenient and would still cause the machine s crosshead to be at a slightly different position after each restart. Another major issue with the testing machine was overheating. The machine would suddenly stop due to overheating of the hydraulic pumps, putting the experiment to halt and a wait of several hours and sometimes days was required for it to be useable again. As a consequence, the results of some early trials were incomplete. This issue was found to be more frequent in the trials tests that involved a large number of cycles where the machine had to be left on for longer periods. In fact, as far as the safety and reliability are concerned, even elastic experiments were not exempt from this issue since several elastic pilot study tests too had to be halted due to it. If a large number of experiments are to be conducted, reliability and consistency of the loading have to be the top priority. By performing this pilot study, the cause of the issue with the peak load trends had been found. However, due to several experimental limitations, an alternative method of applying static loading was needed Trial Tests Using Static Loading One alternative was to use the Instron machine (model 4500) which was available. This machine was capable of loading up to large displacements but could only apply static loads, not cyclic. One possible option was to perform the cyclic loading on ESH machine and static on Instron machine but had there not been the issue with overheating with ESH, this option would have been explored further. 128

129 Now that only static loading was to be applied, a strategy had to be devised to convert cyclic loads into static. For each cycle, where Huff et al. loaded along a single plane (e.g. sagittal) to test along a pair of directions (e.g. extension and flexion), this had to be split into two tests. For example, one test would be for extension and another for flexion by turning the specimen after each test. Similarly, varus and valgus tests would be performed. After preparation of the synthetic bone, a 3-distal hole PHILOS plate was implanted using our screw choice (combination 1 in Table 3). The specimen was set in the varus position to perform the 3D scan. Then, the load was applied at a constant displacement rate of 1 mm/s as defined by Huff et al. The plan was to hold the machine at a fixed displacement (by arresting the crosshead) after every 10 mm increments of displacement so that the specimen can be scanned. This would continue until the maximum displacement of 50 mm was reached i.e. basing it on the maximum displacement on the load-displacement plot provided by Huff et al. Following the initial scan when the specimen was tested and then held at 10 mm for a scan, a drop in the load was recorded. However, when the machine was paused again at 20 mm, another drop in the load was observed, this time it was much larger and continued to drop very sharply. With the fear of dropping too low, no scan was taken and with 1 minute of pause, the loading resumed. Similarly, for the subsequent pauses, the machine was paused for 1 minute to be consistent, without performing any scans. Despite the fact that duration of these pauses was the same at 1 minute, the drop was found to increase each time (Fig. 39). Figure 39. Load-displacement graph for PHILOS specimen tested to failure at 1 mm/s displacement rate. Drops in load at 1-minute scan pauses are clearly visible. 129

130 After background study, it was found that these drops in the load were due to stress relaxation. Stress relaxation is the reduction in stress at constant strain. Theoretical models have been developed to explain these phenomena. In case of rubbers, one explanation is that stress relaxation is caused by the rearrangement and untangling of the polymer chains to accommodate the stress. As a result, bonds are broken, leading to the degradation of the polymer network. This causes the breaking and the reformation of secondary bonds between the chains of the polymer network. The breaking of these bonds decreases the stress whereas formation of new bonds does not affect the stress as they are unstressed at the state of formation [155]. As an attempt to minimise this drop in load, an additional test was conducted, this time reducing the displacement rate from 1 mm/s to 0.05 mm/s. Of course, the strain-rate sensitivity of polymers (from which the synthetic plate is made) is well known [156] and so the overall load values were expected to drop in due course. Also, for the sake of consistency in the experimental method, it was important to fix the time-duration of the 3D scanning step. Fo this, the scanning steps were practiced from beginning till end and the duration was timed. Six minutes were found to be adequate to perform one 3D scan, with the scanner capturing 100,000 mesh triangles. To allow for potential errors in the equipment, be it be in the scanner or the Instron machine, additional two minutes were allowed. Thus, eight minutes were decided to be the duration of each scanning step. An experimental procedure, similar to that used earlier in this pilot study was repeated here, with two main changes. First, the displacement rate was lowered to 0.05 mm/s and then an eight-minute pause was introduced into the testing for 3D scanning. In addition to this, it was decided that a total of three scanning steps would be performed at displacements of 0, 15 and 30 mm. 30 mm was also decided to be the failure criteria, at which the experiment was stopped. Results showed that by using this lower displacement rate, load drop (4.78 N) was much smaller for the whole eight-minute scanning step (Fig. 40). As predicted, these smaller drops were achieved at the cost of lower, overall load values. Based on these results, the displacement rate of 0.05 mm/s was selected for the final experiments. Selecting an even lower displacement rate would further increase the duration of each experiment. By performing this pilot study, the stress relaxation behaviour of the specimen was investigated, allowing they author to decided what displacement rate to select for the final protocol. 130

131 Figure 40. Load-displacement graph for PHILOS specimen tested to failure at 0.05 mm/s displacement rate 3.3 Final Protocol With progressing of the pilot studies, the experimental protocol had changed significantly. Some aspects of it have already been described in the discussions in the pilot studies so they will only be briefly mentioned here and where necessary, the changes will be described. Sixty-five synthetic left humeri (model 1028; Pacific Research Laboratories, Vashon, WA, USA) were obtained and divided into thirteen groups of fives. S3 plates (Zimmer Biomet, IN, USA) of length 83 mm were also obtained, along with 90 mm PHILOS plates (Synthes, Paoli, PA, USA) and 80 mm Equinoxe Fx plates (Exactech, Gainesville, FL, USA). PHILOS plates were implanted in five specimen groups while the S3 and Fx plates were each used for the treatment of four specimen groups. Based on their position on the plate, screws and blades of each plate were numbered and then categorised into zones (Fig. 41). For decisions regarding the selection of screws and blades, an orthopaedic surgeon (Dr Chris Peach) was consulted. Using the depth gauges provided by the plate manufacturers, the maximum possible lengths of the screws in the synthetic bone were determined (Table 3). From this information, a screw configuration was devised for each of the three plates to act as control specimen groups (S0, F0, F0). For PHILOS and the S3 plate, the remaining specimen groups had either zone 1 (P1, S1), 2 (P2, S2), 3 (P3, S3) or 4 (P4) screw holes vacant. Similarly, two of the groups implanted with 131

132 Fx plate had either zone 1 (F1) or zone 2 (F2) vacant but the specimens in the F3 specimen group had the blade swapped with two inferomedial locking screws (Fig. 41). In order to minimise the total number of configuration groups, zoning was mostly on the basis of the screw s distance to the fracture. In the case of zone 3 in PHILOS, screws 3 and 4 were paired as they were both were in convergent mode. Zone 1 in PHILOS plate also included three screws, but since screw 7 was left vacant in all PHILOS configurations, the zone's implanted screws were at a constant distance from fracture site. 132

133 Figure 41. Numbering and zoning of screws and blade holes on S3 (A), PHILOS (B) and Fx (C) plates based on their proximity to fracture gap 133

134 Table 3. Length (mm) and descriptions of the screws and blades of S3 plate (green), PHILOS plate (blue) and Fx plate (red) configuration groups where TP stands for threaded peg, SP smooth peg, ND 90o screw, M multidirectional, CA cancellous, CO-L cortical locking, CO-C cortical compression, N/A where no such hole exists and None where the hole does exist but was left unscrewed. Configuration Group Screw Number S0 (Control) 45, TP 45, TP 45, TP 47.5, SP 47.5, SP 55, SP 34, ND 30, M 30, ND 30, ND N/A N/A S1 (No Zone 1) 45, TP 45, TP 45, TP 47.5, SP 47.5, SP None 34, ND 30, M 30, ND 30, ND N/A N/A S2 (No Zone 2) 45, TP 45, TP 45, TP None None 55, SP 34, ND 30, M 30, ND 30, ND N/A N/A S3 (No Zone 3) 45, TP None None 47.5, SP 47.5, SP 55, SP 34, ND 30, M 30, ND 30, ND N/A N/A P0 (Control) None P1 (No Zone 1) None None None P2 (No Zone 2) None None None P3 (No Zone 3) None None None P4 (No Zone 4) None None None F0 (Control) 29, CA 29, CA 44, CA 44, CA 50 45, Blade 26, CO-L 32, CO-C 26, CO-L N/A N/A F1 (No Zone 1) 29, CA 29, CA 44, CA 44, CA 50 None 26, CO-L 32, CO-C 26, CO-L N/A N/A F2 (No Zone 2) 29, CA 29, CA 44, CA 44, CA None 45, Blade 26, CO-L 32, CO-C 26, CO-L N/A N/A F3 (Swap Blade with Screws) 29, CA 29, CA 44, CA 44, CA 50 44, CA 44, CA 26, CO-L 32, CO-C 26, CO-L N/A N/A 134

135 The initial preparation stage was the same for all the specimens where the synthetic humerus was obtained and prepared in the cement block and then implanted, as described in sections and respectively. Following this, each specimen was subjected to biomechanical testing, which could be divided into two stages, elastic and plastic testing Elastic Testing For elastic testing, the specimen was placed in the Instron machine such that the humeral shaft was in a horizontal orientation. The specimen was then held rigidly at its proximal end (cement block) by a custom fixture, which consisted of a rectangular steel plate placed horizontally on top of the cement block, with the major axis of this plate being perpendicular to the humeral shaft of the specimen. Near to the end of each of its two short sides, the plate was fixed in position by the nuts and bolts. The testing machine was installed with a semi-circular prismatic shaft loader designed in section Care was taken in fixing the specimen so that this crosshead loaded the humeral shaft at a distance of 30 mm from the cut distal end. The control panel of the machine was used to load the specimen to a deflection of 5 mm (i.e. downwards) at a rate of 1 mm/s and then unload to 0 mm (original position). Data was recorded by the machine at a rate of 50 points per second. The aforementioned experimental procedures were carried out sequentially in four directions: flexion, extension, valgus and varus (Fig. 42). For each direction, the steps were performed five times. Figure 42. The four loading directions for the elastic tests 135

136 3.3.2 Plastic Testing Upon completion of the elastic deformation tests, preparations were then made for the subsequent plastic deformation test and intermediate 3D scanning of the specimen at displacements of 0 mm, 15 mm and 30 mm. The plastic test was only performed in the varus direction. The control panel of the testing machine was used to load the specimen at a fixed rate of 0.05 mm/s, with 5 points being captured per second. The specimen was loaded to a maximum deflection of 30 mm, with 8-minute pauses in the loading taking place at displacements of 15 mm and 30 mm each to allow for adequate scanning time. The 3D scanning steps were similar to that determined in section To improve the scan quality, the plate, which was very reflective, was paint sprayed to produce a matte white powder coating. To allow automated data alignment in 3D scanning, the cement block and the two black boards were randomly covered with 3D scanning markers, with the first board being placed on the machine surface directly beneath the humeral shaft of the specimen. Adhesive tape was applied to the cement block to allow adequate adherence of the markers to the block. Additionally, a reference block was placed on the first horizontal board at a set position that would remain unaltered during the plastic test. This position was chosen such that the larger side of the reference block was parallel with the sagittal plane and coincident with one of the edges of the bottom machine surface, and placed 10 cm from the edge of the machine surface furthest from cement block. The purpose of placing the referencing block was to allow comparisons to be made between the scans in the subsequent data processing stages. As a general rule, reaching a total triangle count of 100,000 from the scan of two sides at any stage was considered to be adequate. It was noted, however, that the first scan, which contained the reference block, often yielded more than the midway 50,000 triangles counts. This rule ensured that scans of reasonable quality were recorded. After successfully completing the whole experiment according to this protocol, the obtained graphs and scans were collected and organised. For the elastic tests, data for at least 20 trials was recorded (five for each direction). For the plastic tests, testing machine data for 0-15 and mm displacement was obtained. Of course, there were also the three raw 3D scan files, one for each scan step, were exported in the.stl file format to be processed. 136

137 3.3.3 Data Processing and Statistical Analysis Load-displacement data for all trials of elastic and plastic testing were imported into Microsoft Excel 2010 (Microsoft, Redmond, WA, USA). For elastic testing, load at 5 mm displacement (F 5) and the stiffness (K, gradient of the load-displacement plot) were determined for each trial. For the plastic testing, load-displacement data was plotted for the entire duration of the tests to determine the loads at 15 mm, before (F 15a) and after (F 15b) the intermission and at 30-mm displacement (F 30). SPSS 22.0 software (IBM, NY, USA) was used to perform statistical analyses of the experimental data collected. Using a linear mixed model approach, the effects on specimen groups stiffness and load values were analysed, by accounting for intra- and inter-subject variability. The fixed effect in the analysis was the configuration groups (S0-S3, P0-P4, F0-F3) while the specimens and their trials were set as the random effects. Dependent variables in the elastic test data were K and F 5 while F 15a, F 15b and F 30 were set as dependent variables in the plastic tests Pair-wise difference was tested using Fisher s least significant difference (LSD) multiple comparison based on the least-squared means Post-Processing of 3D Scans In order to be used for further analysis, the 3D scans needed to be processed. To achieve this, the reverse engineering software Geomagic Studio (3D Systems, SC, USA) was used. Here, a brief description of the cleaning process has been described, which was performed on every scan files. After importing the 3D scan (.stl format) into the Studio, the cleaning process was started. First, large, unwanted regions of the scan were first removed manually (e.g. walls, floor, loading machine, cement block) by simply selecting and pressing the delete key on the keyboard (Fig. 43). Having removed these large regions, the Studio s Manifold tool was used to delete smaller triangles that were not connected to the main scan (Fig. 44). Next, before repairing any imperfections in the model manually, the Mesh Doctor tool was used, through which the Studio automatically repairs the polygon mesh by first detecting the imperfections (e.g. spikes) and then rearranging the mesh triangles accordingly. This tool was used repetitively throughout the editing process. 137

138 Figure 43. 3D scans of the setup with the unwanted regions highlight to be deleted Figure 44. 3D scan showing the specimen (left) and the reference block (right) after using the manifold tool and manually deleting unwanted regions Mesh Doctor did not always fully repair the mesh so manual repairing was required. Manual repairing involved two steps: deleting noisy surfaces and filling the holes. Just like Mesh Doctor, both of these tools were used repetitively. For deletion, the model was rotated and inspected to ensure that all spikes and overlapping layers were not present. If the required care was not taken here, the surface, which would later be filled, will be uneven (Fig. 45). 138

139 Figure 45. Before filling any holes, the hole was inspected for spikes After deleting the selected triangles, Fill Single option (using the Complete feature) was chosen and the boundary of the hole to be filled was selected and simply deleted to create a flat opening (Fig. 46). Figure 46. A flat opening such as the distal-most end of the humerus was cleaned using the Fill option The humerus part of the scan was rotated and viewed at various angles in order to make the folds and self-intersections visible (Fig. 47). Then, as before, any triangles that created spikes and overlap another surface were highlighted and deleted. In term of the options for filling holes, there were three features that could be applied appropriately, according to the requirement. 139

140 Figure 47. Folds and self-intersection were a common issue with the scan of the humerus First, there was the Complete fill option by which, as the name suggests, the entire opening was filled by simply selecting the boundary of the hole to be filled. The Partial fill option filled only a portion of the hole. For this, two points on the hole boundary were selected to separate the hole into portions, then a third selection was required to specify whether the left or the right portion was to be filled. Thirdly, the Bridge option allowed the building of bridges across the hole to separate it into two separate holes, by selecting the starting and the ending points of the bridge. There were further options which defined the shape of the filling. By using the Curvature option, the new mesh created by the hole fill matched the curvature of the surrounding mesh. The Tangent option would create filling with a mesh similar to the curvature of the surrounding mesh but with more tapering than the Curvature fill method. Finally, the Flat option simply created a hole filling with a relatively flat mesh. Since all of the aforementioned options were used, the capabilities and limitations of each one of them were studied carefully (Fig. 48). Figure 48. Scan of the inner sides of the fracture gaps required filling and bridging 140

141 Just like in reality, the fracture gap surface was flat in the scans. Several regions in the inner sides of the fractured surfaces were often missing since it was difficult for the scanner to reach them. These holes were filled and set as a smooth surface, using the Bridge and the Flat tool. On the other hand, the humeral head side of the fractured did not need to be edited with such detail and so the bumps and overlapping layers did not have to be removed. In case of the Fx plate, the fracture gap was one of the most difficult areas to scan properly. The scan usually only included the general outline of the sides of the blade, and so tools had to be used to reconstruct the geometry. As mentioned before, the tools and the procedures described here were used numerous times in order to achieve a final, clean version. Thus, several versions of the scan files are saved throughout the cleaning process. The cleaning processing was relatively time-consuming, requiring several hours per scan, even for an experienced user. As an example, Fig. 49 presents the post-processed 3D scans of a specimen that was loaded up to 50 mm in varus in a trial experiment. Geomagic Control software (3D Systems, SC, USA) has a 3D deviation analysis option which produces a 3D colour map, showing the local displacements (Fig. 50). It was planned that for each specimen, a 3D deviation analysis will be conducted to compare the local displacement of the bone-plate construct between 0, 15 and 30 mm loading points. The change in the local displacement would then allow calculation of local strains. Then, with the assumption of material homogeneity and isotropy for both plate and humerus, local stresses could be calculated by the product of the local strain and Young's modulus. This stress and strain data would have then been used for the validation of the FE model developed in chapter

142 Figure 49. Post-processed scans of a bone-plate construct before (A) and after (B) 50 mm varus bending Figure 50. Colour map showing the resultant 3D deviation (mm) across a specimen subjected to 50 mm varus bending load, produced using the 3D deviation analysis tool in Geomagic Control software 142

143 Chapter 4: Results and Discussion of Mechanical Tests 4.1 Results No implant failure or cut-out was noted for any of the construct groups in either elastic or plastic tests. Out of all the configuration groups, the S3 plate's control group (S0) had the highest flexion, valgus and varus bending stiffness. In flexion, it was superseded by the F3 configuration group of the Fx plate where the two inferomedial screws were inserted in place of a blade. Elastic and plastic tests have been presented graphically in Fig and tabulated in tables 4-6. Statistical analysis results are presented in table Elastic Testing In the S3 plate, during extension and flexion bending, the mean stiffness of the control group (S0) was higher (10.233/ N/mm) than that for S1 (10.130/ N/mm), followed by the S2 (7.860/8.052 N/mm) and S3 (7.202/7.542 N/mm) configuration groups (Fig. 51). Therefore, the removal of zone 3 had the greatest effect on extension and flexion, leading to 29.6% and 37.8% drop in mean stiffness respectively, as compared to S0. Removal of the zone closest to the fracture site (zone 1) failed to have a statistically significant effect on extension bending stiffness (Table 4). In PHILOS, removal of zone 1 screws had least impact on mean stiffness values in extension and flexion (7.956/6.349 N/mm). For loading along these two directions, removal of zone 2 led to the largest drop (33.4% and 31.1%) in mean stiffness (6.349/6.887), followed by the removal of screws of zone 3 (6.644/7.045 N/mm), zone 4 (6.871/7.377 N/mm) and zone 1 (7.956/8.284 N/mm). In extension and flexion testing of Fx plate, mean stiffness with the swapping of the blade with inferomedial screws in F3 (10.915/ N/mm) was higher than the control group F0, followed by the removal of 6.5 mm screws in F2 (8.122/8.990 N/mm) and blade in F1 (7.734/8.248/ N/mm). Mean valgus and varus bending stiffness for the S0 group was statistically significantly higher (9.803/ N/mm) than S1 (9.045/8.357 N/mm), followed by the S3 (7.638/6.822 N/mm) and S2 (6.580/6.496 N/mm) configuration groups. 143

144 Out of all zones tested for PHILOS plate, removal of zone 1 screws in P1 had the greatest effect on valgus and varus stability, leading to 23.3% and 28.5% drop in mean stiffness (mean stiffness: 4.671/4.726 N/mm) when compared to the control group P0 (Table 5). In the order of decreasing effect on varus stiffness with their removal, zone 1 screws were followed by screws of zone 2 (5.867 N/mm), zone 3 (6.059 N/mm) and zone 4 (6.268 N/mm). Removing the blade in the Fx plate (F1) led to a larger decrease in valgus and varus stiffness (8.2% and 26.0%) than the removal of 6.5 mm screw (F2) when compared to control group F0. Swapping the blade with screws led to a statistically significant increase in extension, flexion, valgus and varus stiffness than the control group (8.245/8.663 vs 6.900/7.590 N/mm). From the pairwise comparison, there was a statistically significant difference between peak load values and stiffness of all S3 plate s configuration pairs, except for the S0 and S1 pair in extension (Table 7). In the PHILOS configuration groups showed that there were statistically significant differences (P values less than 0.05) between stiffness and load values of all configuration group pairs, except two cases (Table 8). These were P3 and P4 in extension, and P2 and P3 in flexion. As for the pairwise comparison of the Fx plate configuration groups, there were statistically significant differences between peak loads and stiffness values of all configuration pairs (Table 9). For all three plates, trends in the elastic bending loads (F 5) were similar to those in the stiffness (K), as shown in Fig Plastic Testing Temporal stress decay was observed during the eight-minute intermission, owing to stress relaxation, a phenomenon commonly exhibited by viscoelastic materials such as polyurethane (a primary constituent of synthetic humeri) when under constant strain. As a result of stress relaxation, up to an approximately 4-5 N drop in load (difference between F 15b and F 15a) was recorded on the load-displacement plot at abscissa of 15 mm (Fig. 53 and 54). With a few exceptions, the load trends recorded for the plastic tests among the different configurations were similar to those recorded for elastic varus tests. There was a statistically significant difference between all configuration pairs with three exceptions. These were the were the F 15b values for the F0 and F2 configuration group pair, F 30 values between S0 and S1 pair and the F 15a, F 15b and F 30 values of the S2 and S3 configuration group. Similarly, although the mean elastic varus bending peak load F 5 of S0 was over 9 N higher than that of F3, in the plastic tests, their difference in the mean final load F 30 was only over 1 N (Table 4 and 5). 144

145 Figure 51. Mean stiffness (S) for S3, PHILOS and Fx plate configuration groups during elastic loading of 5 mm cantilever displacement in extension, flexion, valgus and varus directions 145

146 Figure 52. Mean peak load (F5) for S3, PHILOS and Fx plate configuration groups during elastic loading of 5 mm cantilever displacement in extension, flexion, valgus and varus directions 146

147 Figure 53. Mean peak loads (F) for S3, PHILOS and Fx plate configuration groups during plastic loading at 15 mm displacement before (F15a) and after (F15b) eight-minute intermission and at 30 mm displacement (F30) 147

148 Figure 54: Typical load-displacement curves at load point for S3, Fx and PHILOS plate control groups (S0, F0, P0) constructs during plastic loading. A drop of 4-5 N in load is noted at 15 mm displacement due to the stress relaxation of construct during the eight-minute intermission. 148

149 Table 4. Mean stiffness (K) and load values (F) for all S3 plate configuration groups, along extension, flexion, valgus and varus, with their respective standard deviations (S.D.). K and F5 denote stiffness and peak load values obtained during elastic tests while F15a and F15b are loads at 15 mm before and after eight-minute intermission and F30 is the load at 30 mm during plastic tests. Extension S0 S1 S2 S3 K (N/mm) ± ± ± ± F5 (N) ± ± ± ± Flexion K (N/mm) ± ± ± ± F5 (N) ± ± ± ± Valgus K (N/mm) ± ± ± ± F5 (N) ± ± ± ± Varus K (N/mm) ± ± ± ± F5 (N) ± ± ± ± F15a (N) ± ± ± ± F15b (N) ± ± ± ± F30 (N) ± ± ± ±

150 Table 5. Mean stiffness (K) and load values (F) for all PHILOS plate configuration groups, along extension, flexion, valgus and varus, with their respective standard deviations (S.D.). K and F5 denote stiffness and peak load values obtained during elastic tests while F15a and F15b are loads at 15 mm before and after eight-minute intermission and F30 is the load at 30 mm during plastic tests. Extension P0 P1 P2 P3 P4 K (N/mm) ± ± ± ± ± F5 (N) ± ± ± ± ± Flexion K (N/mm) ± ± ± ± ± F5 (N) ± ± ± ± ± Valgus K (N/mm) ± ± ± ± ± F5 (N) ± ± ± ± ± Varus K (N/mm) ± ± ± ± ± F5 (N) ± ± ± ± ± F15a (N) ± ± ± ± ± F15b (N) ± ± ± ± ± F30 (N) ± ± ± ± ±

151 Table 6. Mean stiffness (K) and load values (F) for all Fx plate configuration groups, along extension, flexion, valgus and varus, with their respective standard deviations (S.D.). K and F5 denote stiffness and peak load values obtained during elastic tests while F15a and F15b are loads at 15 mm before and after eight-minute intermission and F30 is the load at 30 mm during plastic tests. Extension F0 F1 F2 F3 K (N/mm) ± ± ± ± F5 (N) ± ± ± ± Flexion K (N/mm) ± ± ± ± F5 (N) ± ± ± ± Valgus K (N/mm) ± ± ± ± F5 (N) ± ± ± ± Varus K (N/mm) ± ± ± ± F5 (N) ± ± ± ± F15a (N) ± ± ± ± F15b (N) ± ± ± ± F30 (N) ± ± ± ±

152 Table 7. P values for elastic stiffness and peak load values of S3 plate configuration groups, obtained from their pairwise comparison statistical analysis Peak Load (F5) Stiffness (K) Direction/Zone S1 S2 S3 S1 S2 S3 Extension S <0.001 < <0.001 <0.001 S1 <0.001 <0.001 <0.001 <0.001 S2 <0.001 <0.001 Flexion S0 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 S1 <0.001 <0.001 <0.001 <0.001 S2 <0.001 <0.001 Valgus S0 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 S1 <0.001 <0.001 <0.001 <0.001 S2 <0.001 <0.001 Varus S0 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 S1 <0.001 <0.001 <0.001 <0.001 S2 <0.001 <

153 Table 8. P values for elastic stiffness and peak load values of PHILOS plate configuration groups, obtained from their pairwise comparison statistical analysis Peak Load (F5) Stiffness (K) Direction/Zone P1 P2 P3 P4 P1 P2 P3 P4 Extension P0 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 P1 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 P2 <0.01 <0.001 <0.01 <0.001 P Flexion P0 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 P1 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 P < <0.001 P3 <0.001 <0.001 Valgus P0 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 P1 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 P2 <0.01 <0.001 <0.01 <0.001 P3 <0.01 <0.01 Varus P0 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 P1 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 P2 <0.05 <0.01 <0.05 <0.01 P3 <0.05 <

154 Table 9. P values for elastic stiffness and peak load values of Fx plate configuration groups, obtained from their pairwise comparison statistical analysis Peak Load (F5) Stiffness (K) Direction/Zone F1 F2 F3 F1 F2 F3 Extension F0 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 F1 <0.001 <0.001 <0.001 <0.001 F2 <0.001 <0.001 Flexion F0 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 F1 <0.001 <0.001 <0.001 <0.001 F2 <0.001 <0.001 Valgus F0 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 F1 <0.001 <0.001 <0.001 <0.001 F2 <0.001 <0.001 Varus F0 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 F1 <0.001 <0.001 <0.001 <0.001 F2 <0.001 <

155 Table 10. Results of pairwise comparison statistical analysis: P values for plastic loads at 15 mm displacement before (F15a) and after (F15b) eight-minute intermission and at 30 mm displacement (F30), for S3 plate configuration groups. Direction/Zone S1 S2 S3 F15a (N) S0 <0.001 <0.001 <0.001 S1 <0.001 <0.001 S F15b (N) S0 <0.001 <0.001 <0.001 S1 <0.001 <0.001 S F30 (N) S <0.001 <0.001 S1 <0.001 <0.001 S

156 Table 11. Results of pairwise comparison statistical analysis: P values for plastic loads at 15 mm displacement before (F15a) and after (F15b) eight-minute intermission and at 30 mm displacement (F30), for PHILOS plate configuration groups. Direction/Zone P1 P2 P3 P4 F15a (N) P0 <0.001 <0.001 <0.001 <0.001 P1 <0.001 <0.001 <0.001 P2 <0.001 <0.001 P3 <0.05 F15b (N) P0 <0.001 <0.001 <0.001 <0.001 P1 <0.001 <0.001 <0.001 P2 <0.001 <0.001 P3 <0.05 F30 (N) P0 <0.001 <0.001 <0.001 <0.001 P1 <0.001 <0.001 <0.001 P2 <0.001 <0.001 P3 <

157 Table 12. Results of pairwise comparison statistical analysis: P values for plastic loads at 15 mm displacement before (F15a) and after (F15b) eight-minute intermission and at 30 mm displacement (F30), for Fx plate configuration groups. Direction/Zone F1 F2 F3 F15a (N) F0 <0.001 <0.05 <0.001 F1 <0.001 <0.001 F2 <0.001 F15b (N) F0 < <0.001 F1 <0.001 <0.001 F2 <0.001 F30 (N) F0 <0.001 <0.05 <0.05 F1 <0.05 <0.001 F2 <

158 D Scanning Post-processing of the 3D scans, as described in section was relatively time-consuming, and up to 195 scans were conducted (3 scans per specimen x 5 specimens per configuration group x 13 configuration groups), each require between minutes for post processing. More importantly, the scans were often poor, require a lot of manual editing, thus risking its validity for use for further analysis. For many scans, the 3D comparison analysis could not be completed successfully as either the software was unable to find the corresponding cloud points on the two scans or the scans had been manually edited so much that they validity of the comparison was questionable. This problem was found after processing scans of several different configuration groups. Thus, it was not possible to remove the 3D scanning steps from the subsequent tests as this would make it difficult to compare the plastic test results. As a result, for consistency, 3D scanning steps were kept in the plastic tests of all specimens, despite the fact that the scans could not be used for further analysis. 158

159 4.2 Discussion The PHILOS plate is one of the most widely studied implants in the literature regarding in vitro biomechanical testing of proximal humerus fixation systems, introduced typically as the representative of the broad category of locking plates. On the contrary, to our knowledge, no such study exists on biomechanical testing of the Fx plate constructs yet or to compare the performance of this new concept. There is evidence concerning the comparison of blade plates with locking plates. Weinstein et al. [34], for example, compared the torsional performance of a locking plate with an angled blade plate for treatment of three-part fractures on cadaveric humeri. Mean initial torsional stiffness was reported to be significantly higher for the locking plate than a blade plate. Siffri et al. [33] seem to provide ample support to this conclusion. Like Weinstein et al., they also applied cyclic torsional loading and reported significantly less fixation loosening for locking plate than the blade plate in cadaveric humeri. On synthetic humeri, however, both plates performances were statistically similar. With the aim of mimicking humeral forces during common activities of daily living, Gillespie et al. [86] loaded cadaveric humeri, with three-part fractures, at 20 of abduction from vertical. Mean stiffness value for blade plate was found to be 12% higher than locking plate but with no statistically significant difference. Beyond the few studies mentioned, there is insufficient research into the biomechanical comparison of locking plate and blade plates, particularly that of the hybrid plate systems, to draw any firm conclusions and delineate a superior mode of fixation, hence the need for our mechanical tests. Three studies compared the biomechanical performance of S3 plate with other plates, all with varying conclusions. Yamamoto et al. [99] recorded significantly less distal fragment displacement during varus bending for cadaveric humeri treated with an S3 plate with smooth pegs than those treated with Synthes locking plate with threaded pegs. Huff et al. [95] reported higher torsional and varus and valgus bending stiffness for the S3 plate than PHILOS plate, but a lower extension and flexion bending stiffness. However, it is difficult to draw concrete conclusions from this study regarding the superior implant, since the PHILOS plate used was longer than the S3 plate. For the treatment of three-part fractures, Rose et al. [130] found that specimens treated with the S3 plate showed significantly higher displacement of greater tuberosity fragment and larger rotation of the head fragment than those treated with conventional locking compression plate under simulated o cyclic abduction. For varus bending stiffness, our findings more closely support the conclusions of Huff et al. and Yamamoto et al. Our study further showed that S3 control group (S0) was stiffer than both PHILOS and Fx plates control groups (P0 and F0), in all four bending directions. 159

160 Cantilever loading has been used previously in the literature, often to achieve a bending moment of Nm at the fracture site [33,79,81,94,99,157]. Comparable bending loads were applied by Chow et al. [79] and Weeks et al. [81] who performed cantilever bending on the basis of a biomechanical study by Poppen and Walker [123], with the aim of replicating the supraspinatus forces on bone-plate constructs during the early stages of healing under shoulder immobilisation support. Mechanically, this loading is comparable to humeral immobilisation followed by a varus force acting directly at the supraspinatus insertion site. Taking 7.5 Nm as the upper limit, the maximum bending moment at the fracture site achieved in our study during varus elastic tests was Nm (for S0), well within the Nm range. For the plastic loading, however, these bending moments reached up to approximately 17 Nm for Fx and S3 plate constructs and 14 Nm for PHILOS plate constructs. Despite these loadings, no implant failure was recorded. A distinguishing feature of our study is its relatively large size and the testing of multiple zones for each plate. Most in vitro studies in the literature do not divide the plates into zones and those that do, are limited to a comparison of one or two zones, often the inferomedial screw zone. This makes the detailed comparison between our study and them difficult. The choice for different configurations groups in our study was made with the consultation of a shoulder orthopaedic surgeon (Dr Chris Peach), to ensure that each group represented a clinically applicable option that clinicians would use in the theatre. However, the zoning that we defined based on this principle meant that the zones differed from each other not only by their distance from the fracture site but also in other design parameters. So, by testing for the effect of a zone, we were, in fact, testing the combined effects of several design parameters. All the design parameters changed in our study could be classified as either relating to the position, orientation and geometry of plates and screws (Fig. 55). A plate's geometry can be modified by changing its surface profile, dimensions or the number and location of screw holes in it. Likewise, a screw s geometry was changed by changing its surface profile, dimensions or by shaping it into a blade. These design parameters were highly inter-dependent and so ideally further mechanical testing is required to investigate their effects in isolation. 160

161 Figure 55. Overview of the different ways plates' zones differed from each other. The three plates (S3, PHILOS and Fx) were implanted at different positions and orientations (A). Their geometries were also different, in terms of their dimensions (B), surface profile (C) as well as the number (topology) and position of screw holes (D). Zones also differed in terms of the number (D), orientation (E) and the geometry of their screws. The latter differed with a changed in surface profile (F; e.g. using smooth pegs), dimensions (G) or use of a blade in place of the screw (H). 161

162 It is found that the contribution of most of these parameters on the mechanical testing results obtained in section 4.1 can be explained by studying a simple analytical model which is commonly discussed in the literature [ ] (Fig. 56). The model simplifies the humerus and screws as cylinders and the plate as a cuboid and assumes perfect contact at all surface interfaces (bone-screw, bone-plate and plate-screw). A fracture gap is introduced to dividing the humerus into a head fragment which is fixed perfectly at the end and a shaft fragment which is loaded in a cantilever fashion along the four directions. Figure 56. Schematic of a simple analytical model commonly studied in the literature. Cylindrical bone is split into two fragments ('head' and 'shaft') and held together with a cuboid plate and screws. With the head fixed, shaft is loaded in the varus bending direction. Despite these large simplifications of the in vitro loading, the model is able to explain the mechanical effects of most design parameters. One possible reason for the success of this model is the role material properties role in the construct stability. Young s modulus of the stainless steel 316L is several hundred times higher than the synthetic bones used for the tests. Thus, theoretically, plates and screws design dominate the trends in the bending stiffness along the four loading directions, despite the irregularities of the humeral head geometry as noted in the pilot study Regarding this, Gautier et al. [158] noted that the neutral axis of a composite beam (humerus with plate) shifts based on the material properties of the plate. A plate with an elastic modulus of bone leads only to a small shift but that with a steel plate (such as S3, PHILOS and Fx) with high elastic modulus increases the shift of the new common neutral axis (Fig. 57). Due to this, the second moment of area and thus the stiffness of the entire construct is enhanced. It follows from this that it is better to focus on the plate geometry if the construct stiff is to be enhanced. 162

163 Figure 57. Insertion of a plate with same elastic stiffness modulus as bone shifts the neutral axis (A). With the insertion of a steel plate, this shift (arrow) is increased much further, highlighting the domination of the plate in the bending behaviour of the whole construct (Adapted from Fig in [158]). In our case, however, humerus role in the construct stability must not be ignored, as it is humerus geometry and material properties that ultimately dictate the implant design. For example, the stability of locking plates such as the three plates is largely dependent on the bone-screw interface. So, the great efforts to optimise the screw s position, orientation and geometry for maximum purchase will be in vain if the bone quality is poor, as is the case in osteoporotic patients. Similarly, it is the humerus which dictates the physical boundaries within which the screws and plates can be placed, such as to not disturb the surrounding soft tissue. Sections discuss the mechanical test results in terms of the design parameters. Where possible, the analytical model has been used for explanation. To aid these discussions, Fig. 58 presents the elastic bending results of each configuration group as a percentage change in stiffness relative to their control groups. 163

164 Figure 58. Percentage change in the extension, flexion, valgus and varus bending stiffness of all configurations of the S3, PHILOS and Fx plate, as compared to their respective control groups. NB: Cell background colours indicate the magnitude of the change and plates are represented in their correct relative scale. 164

165 4.2.1 Plate Position and Orientation Although difficult to assess separately, effects of plate s position and orientation are more noticeable when screws position and orientation is decided. If done correctly, it can allow insertion of screws that can target the regions of the humerus that were once difficult to reach and hold multiple fracture fragments. Both the PHILOS and Fx plates are positioned 8 mm distal from the greater tuberosity. Although the PHILOS plate is known for its low-profile design, especially of its humeral head section, its impingement under the acromion continues to be a common post-operative complication [142]. In their clinical study, Geiger et al. [143] reported cases of sub-acromial impingement with the use of the PHILOS plate and recommended the plate to be placed more distally. To address this issue, the S3 plate was designed to be implanted 30 mm distal to greater tuberosity. As a result, the S3 plate had a much smaller contact area with the humeral head and was more compact than the PHILOS and Fx plate. All three plates formed a small but noticeable clockwise angle with the humeral shaft axis. This was particularly noticeable in the PHILOS plate. If large enough, this angulation may affect the trends in the bending stiffness in the humerus Plate Dimensions Based on the analytical model, dimensions of the cuboid plate could be changed by changing its length, width and thickness (Fig. 55). An attempt was made to use the plates of similar length in the mechanical tests. However, due to manufacturer s defined lengths, the PHILOS plate was the longest (90 mm), followed by the S3 plate (83 mm) and the Fx plate (80 mm). Longer plates that have screws at the extreme screw holes are found to be theoretically better at resisting bending than shorter plates, due to longer effective length. Increasing the plate length reduces the pull-out force at the bone-screw interface and increased working leverage for each screw [161,162]. For the three proximal humerus plates, long and shaft plates differ mainly by the lengths of their shaft sections. Increasing the plate s length along it s the humeral head section poses the risk of interfering with surrounding muscles, as does the increase in its width. Also, plates with longer shaft section allow insert of additional shaft screws that are useful in the treatment of proximal humerus fractures that extend to the humeral shafts. A cubic relationship is derived between the plate s thickness and the second moment of area during varus and valgus, as opposed to a linear one for its width. This was indeed true for the PHILOS and Fx plates which had a large enough difference between their thickness and the 165

166 width that their varus and valgus bending stiffness was lower than the extension and flexion bending stiffness. Out of the three plates investigated, S3 stands out as it has a noticeably thicker cross-sectional area, especially at the fracture site (Fig. 59). These mechanical advantages of the S3 plate were manifested in the results where the S0 configuration group exhibited 59.5% and 21.7% higher mean bending stiffness in varus bending stiffness than the stiffest configurations of the PHILOS (P0) and Fx plate (F3) groups, respectively. Similar trends were reported for valgus bending stiffness. The S3 plate's superiority over PHILOS and Fx plate in varus and valgus bending stiffness owes partly to its thicker cross-section. Figure 59. Cross section profiles of the S3, PHILOS and the Fx plate at plate sections spanning the fracture site, just before the beginning of their respective zone 1 screw holes. NB: plates are represented in their correct relative scale. 166

167 4.2.3 Plate Surface Profile In addition to the differences in the dimensions, the surface profile (or local geometry) of the three plates varied significantly, affecting the stress and strain distribution of the whole construct. For example, based on their position and orientation on the humerus, all three plates had anatomically accurate under-surfaces. This increased the bone-plate interface area, allowing the plate to be snug as tightly to the humerus as possible and distribute the stresses across the surface. Furthermore, the minimal design of the S3 plate, with relatively small size and high stiffness, poses the risk of high stress concentration, a risk common with rigid implants [163]. A certain degree of plate flexibility of plate movement is advocated, especially for osteoporotic bones in several studies to allow sufficient fragment movements for bone healing [111,117,164]. This raises the question of whether entire plate should have a thicker cross-section area or only the critical regions such as the section spanning the fracture gap. Stresses and strains were not measured directly in our mechanical tests so it is difficult to comment on the exact consequences of the changes in surface profiles. Further mechanical tests or computational simulations are required to measure them and locate stress concentrations. Nevertheless, as for the plates position and orientation, the benefits of a particular surface profile become apparent indirectly. For example, all three plates had a larger width for the head section than the shaft section and the benefit of this became apparent when it allowed the insertion of multiple screws per zone that could be oriented at large diverging and converging angles Screw Number The decision of how many and which screws to implant or leave out is a crucial one that is frequently made by clinicians. There is very little information in the literature on the optimal number of screws to insert for a given fracture. Erhardt et al. [49] and Cohen et al. [165] recommend that five humeral head screws should be inserted including at least one inferomedial screw. Ultimately, this decision is dependent on how many screw holes the plate offers (its topology). The PHILOS and Fx plate employs nine and seven humeral head screws respectively including two inferomedial screws. On the other hand, the S3 plate only allows insertion of up to six screws where also two are inferomedial. The fewer number of S3 plate screw holes may limit its use for the treatment of more complex humerus fracture where contact with a more extreme region of the humerus is vital for construct stability. This is particularly important given that the direction of the screws cannot be changed in fixed-angle locking devices. The PHILOS plate, on the other hand, offers more screw holes, allowing a wider range of directions to serve 167

168 more complex fracture cases. Rose et al. [115] tested three-part fractures of the proximal humerus and found that the S3 plates had significantly greater tuberosity movement than PHILOS. This is possibly because in their study, the S3 plate had only two pegs connecting the greater tuberosity with the rest of the head whereas the PHILOS plate had four. Stability of locking plates has been found to depend on their screw density [166]. This is defined as the quotient formed by the number of screws inserted and the total number of plate holes and is empirically recommended to be less than 0.4 for simple fractures and less than or equal to 0.5 for comminuted fractures [162]. In our study two of the configurations involved removal of single screws. Removal of the Fx plate s 6.5 mm screw had less effect on the mean stiffness in all four directions than the removal of the blade (F0) or insertion of the two inferomedial screws (F3). Similarly, the removing the single screw in zone 1 (S1), reduced the mean stiffness in all four directions less than the removal of zone 2 (S2) and 3 screws (S3). In fact, statistical analysis revealed that its removal had no statistically significant difference in the extension stiffness when compared to the control group (S0). These results suggest that two screws in a zone are more important than one, possibly due to the increase in the bone-screw interface area Screw Position The change in the surface profiles of the head and the shaft axes allowed the head screws of the Fx and S3 plate to be positioned along a new axis (Fig. 60). This axis deviated even further away from the humeral shaft than its shaft screws. For the PHILOS plate, however, both head and shaft screws were positioned along the same axis. Whether this had a significant effect on construct stiffness recorded in our mechanical tests, is not clear. 168

169 Figure 60. Photographs of the S3 (A), PHILOS (B) and the Fx plate, superimposed with the shaft- (red) and head- (green) screws' axes 169

170 A clearer relationship is found between the position of the screws head with respect to the fracture site and the bending stiffness, particularly the varus and valgus bending stiffness. Based on the analytical model, Smith et al. [159] defined the plate working length as the distance between the closest two screws on either side of the fracture gap. Since we kept the number of shaft screws constant for a given fracture, the plate working length could only be changed by the filling or emptying of zones neighbouring the fracture gap. If this length is kept too short, the overall construct stiffness is increased but so is the risk of high stress concentration, straining and eventual failure of the plate at the section around the fracture gap [163]. If the spanning segment is too short, the working length of each screw is decreased, and applied loads will lead to high levels of strain. When applied over this short spanning segment, this bending moment increases the local strain experienced by the implant [162,167]. These negative effects are more profound in the osteopenic bone where the bonemetal interface is weak, leading to screws cut-out, second fracture at plate end and reduction in the micromotion needed for postoperative callus formation [117,168,169]. In response to this, the concept of semi-rigid plates arose, where some flexibility and movement is permissible at the fracture site in order to absorb the load energy and reduce strains at the bone-metal interface [111,164]. However, care must be taken during fracture healing to prevent excessive movements keep the fracture fragments undisturbed and intact enough to avoid early failure [117]. This flexibility could be achieved by keeping the screws holes near the fracture site empty to increase the working length and keep stress and strain low and well distributed. This comes at the expense of a reduction in overall construct stiffness and stability. For example, removing zone 1 screws in the PHILOS plate increased the moment arm which reduced the cantilever load required to produce the same moment, leading to reduced stiffness. Mechanically, this is similar to the case of a cantilever beam with a concentrated load at the free end where maximum deflection of the beam has a cubic relationship with its arm length. In our study, P2, P3 and P4 configuration groups had zone 1 screws in place, so the working length was fixed. This would imply that for a given load direction, the bending stiffness of these configurations would be the same. Instead, the effect of screw removal on overall construct stiffness decreased from zone 2 to 3 and eventually zone 4. This decline at each proximal progression is possibly because of the gap in the plate created by the removal of zone 2 screws. Further, this omission of screws near the fracture site affected the plate stability and construct stiffness. For P3 constructs, zone 1 and 2 screws were already present between zone 1 and the fracture gap. So, theoretically, they would exhibit higher stiffness than 170

171 constructs with P2 configuration. Similarly, P4 constructs had zone 1, 2 and 3 screws in place and were stiffer than P3 constructs. Therefore, as the gap in the plate was made more proximal from the fracture gap, its influence over fracture gap stability and the construct stiffness diminished (Fig. 61). Figure 61. Schematic showing the movement of the gap in the screw insertion at a constant working length Based on these principles acquired from the simple analytical model, one could argue that it is the proximity of a zone to the fracture gap that dictates its importance in construct stability, as demonstrated by Stoffel et al. [163]. In varus and valgus bending of the PHILOS plate, this was indeed true as the zones could be listed in the following order of reducing the effect on construct stiffness: zone 1, 2, 3 and 4 (Fig. 58). Similarly, being further away from the fracture site, removal of the 6.5 mm screw in the Fx plate (F2) had less effect on varus and valgus stiffness than the blade (F1) and inferomedial screws (F3). It needs to be investigated whether this is due to the screw s position or to the fact that it was a single screw being compared with two inferomedial screws. Likewise, in the S3 plate, with the zone 3 being further away from fracture site than zone 2, its removal (S3) led to 22.1% and 35.3% decrease in mean valgus and varus bending stiffness as compared to 32.9% and 38.4% for zone 2 (S2). However, despite being the closest zone to the fracture site, removal of zone 1 (S1) led to 7.7% and 20.7% decrease in mean valgus and varus bending stiffness, much less than the other two zones. This might be because unlike zone 2 and 3, zone 1 has only one screw, so its number may have had a larger effect on the stiffness than its position. 171

172 4.2.5 Screw Orientation It can be argued that the orientation of the screws in S3 plate s zone 2 also played a role in the zone s superiority over zone 1 during varus bending. Like zone 1 in the PHILOS and Fx plate, its screws are directed towards the calcar region of the humerus, a region critical for the humeral head s stability against varus collapse [41,47]. As compared to the other screws in their respective plates, removal of these inferomedial screws led to the largest drop in both varus and valgus bending stiffness. We believe that there are two main motives behind orienting screws towards the calcar region. The first motive is purely mechanical: to counter the rotations experienced by the calcar region during varus bending. The second is biological: to connect regions of the humeral head with low bone quality to those of high bone quality such as the medial region [132,170]. This is a commonly reported reason in the literature. However, since the bone specimen we used was made out polyurethane foam with relatively uniform density, the superior performance of medial support was not attributable to the better bone mineral density. The use of synthetic bone was both an advantage and disadvantage of our study. It was an advantage in the sense that it highlighted that medial support is vital for varus stability, irrespective of the bone mineral density. At the same time, it was a simplification of the in vivo scenario and thus demands future testing on cadaveric specimens. In addition to the inferomedial orientation of the screws, directing them at a neck shaft angle of a healthy humerus may also have mechanical benefits. Measurement of the angle between the humerus anatomical neck and its shaft in the frontal plane is a common radiological method of assessing its varus stability. Several studies have stated that the normal anatomical neck-shaft angle is approximately 130 o -135 o [171,172]. A neck-shaft angle of less than or equal to 100 o has been shown to predict failure [173]. Although the use of medial support, especially in the form of inferomedial screws, is a common way of maintaining varus stability and thus the neck-shaft angle, Yewlett et al. concluded that it alone does not compensate for poor fracture reduction [41,173]. Fixed angle locking plates, in general, do not allow for improvement of reduction after locking screws are inserted. Therefore, it is important to obtain anatomical reduction during the surgery. Amidst this, lies the question of whether the angle created by the plate s screws with the humeral shaft needs to be the same as the neck-shaft angle. We were unable to find any study in the literature directly investigating this question. Perhaps the closest study was that by Fuchs et al. [174] where a 90 o blade plate was straightened to a o and satisfactory clinical outcomes were achieved. Their motive for this change in the angle was to move the blade s entry point to a more distal location so that the sub-acromial impingement can be avoided and also to allow the blade to enter the central, high cancellous bone density region of the humeral head. By placing the S3 plate more distally, the zone 3 screws formed a 135 o angle to the shaft, as opposed to near-90 o made by the 172

173 uppermost screws of the PHILOS plate. Dividing the plate into zones and testing the effects of each zone individually allowed us to indirectly investigate the effects of these angles. In the S3 plate, zone 3 screws are located inferior to the zone 4 screw but their tips extend more proximal. Removal of S3 plate s zone 3 screws led to a larger drop (35.3%) in its varus stiffness from its control group S0 than the removal of any screw zone did from PHILOS plate s control group P0. This suggests the advantage of the 135 o angle over 90 o in varus bending. A more controlled investigation, one free from the influence of differing plate and screw design parameters is required in order to draw more concrete conclusions. According to the analytical model, a zone s contribution to the extension and flexion bending stiffness is dependent on its second moment of area. This can be increased geometrically by increasing the screw diameter as was the case with the 6.5 mm screw in Fx plate or by making the zone s screws bi-directional (convergent/divergent) to place the area further away from the neutral axis. Studies on simple biomechanical fractures and specimen cases reveal that less force is needed for the loosening of parallel screws since bi-directional (convergent/divergent) screws have a high area of resistance and thus better bone anchorage [16,175,176]. Based on this understanding, theoretically, a given zone s second moment of area increases from a single centrally aligned 3.8 mm screw (e.g. S3 zone 1) to a single centrally aligned 6.5 mm screw (e.g. Fx zone 2) to near-parallel screw pair (e.g. PHILOS plate zone 1) to a divergent/convergent screw pair (S3 plate, zone 2), as shown in Fig. 62. Figure 62. Cross-sectional views of the bone-plate construct, showing the changes (arrows) in the furthermost point of the screw from the extension and flexion bending neutral axis, for a single centrally aligned 3.8 mm screw (A; e.g. S3 zone 1), a single centrally aligned 6.5 mm screw (B; e.g. Fx zone 2), a near-parallel screw pair (C; e.g. PHILOS plate zone 1) and a divergent screw pair (D; S3 plate, zone 2) Up to some degree, our results obeyed these principles. For example, removal of zone 3 in the S3 plate led to the largest drop in mean stiffness value in extension and flexion (29.6% and 37.8%) when compared to S0 group, followed by zone 2 screw pair (S2) and then single screw in zone 1 (S1). Further, the near-parallel screws of zone 1 (F3) in the Fx plate had a significantly larger effect on extension and flexion bending stiffness than the centrally aligned 173

174 large 6.5 mm screw in zone 2 (F2). It can be hypothesised that if two small screws both offset from plate midline were used instead of one large 6.5 mm screw, the bending stiffness, particularly in extension and flexion cantilever bending would be improved since the offset would increase the second moment of area. In the PHILOS plate, zone 2 screws diverged by a relatively large angle while zone 3 screws had convergent trajectories. Zone 4 and 1 screws, on the other hand, were almost parallel to each other. There, the bidirectional screws of Z2 and Z3 had a more significant effect ( % decrease) on mean extension and flexion bending stiffness than the near-parallel zone 1 ( % decrease). While the removal of zone 2 s diverging screws had the largest impact on the extension and flexion stiffness, it was only a few percent greater than that of the neighbouring convergent screws in zone 3. Furthermore, there was no statistically significant difference between the removal of the converging zone 3 and the near-parallel zone 4. The S3 plate has only two zones of screw pairs to contribute to extension and flexion stability whereas the PHILOS plate and the Fx plate had four and three such screw pairs. Despite this, the S0 control group exhibited 7.3% and 21.3% higher mean extension and flexion bending stiffness than the PHILOS plate s control group (P0). However, when compared to the Fx plate s F3 construct, flexion stiffness of S0 was 9.0% higher but extension stiffness 6.2% lower. Further biomechanical testing is required to systematically investigate the effects of screw orientation on the bending stability along the four loading directions, particularly in terms of the inferomedial screws, neck-shaft angle and their second moment of area Screw Dimensions Based on the analytical model, screws can be modelled as cylinders, whose dimensions can be described by two design parameters: length and radius. Both parameters are required to be contained inside the humeral head and not collide with the other screws. Theoretically, longer screws have a net positive effect on construct stability because of their increased interface area. Further, screw length can determine the extent to which the second moment of area of a zone affects the stiffness, especially in bidirectional screws. For example, the benefits of diverging screw pair are more noticeable if the screws are long enough to increase the distance from the neutral axis. In our tests, screw purchase was idealised by keeping the screws long enough to achieve subchondral bone abutment. Due to the irregular geometry of the humeral head, screws were of varying lengths. Their possible influence on the mechanical performance of different zones can therefore not be ignored. However, since 174

175 all the other design parameters were also changed simultaneously, it is difficult to draw a definitive conclusion on the effects of longer screws, despite their theoretical benefits. In the clinic, glenohumeral perforation of the screws is one of the leading complications associated with angle stable plates [70,71,177]. Using screws of shorter length may prevent screw perforation but can also led to poor bone anchorage since density of the cancellous bone in the subchondral region is relatively high [178]. As a consequence of this, the construct can lose its stability and collapse in varus. A distinguishing feature of the Fx plate is that it allows insertion of a 6.5 mm diameter screw in addition to 3.8 mm ones. For the S3 plate, zone 1 has only one screw and that too a centrally aligned one, like the Fx plate s 6.5 mm locking screw. However, no statistically significant difference was reported between extension bending stiffness recorded for the removal zone 1 when compared to the S0 control group. On the other hand, removal of the 6.5 mm screw led to a 7.4% and 5.8% drop in extension and flexion bending stiffness as compared to the F0 control group. In addition to the influences of orientation, it may also be due to the Fx screw s bigger core diameter, leading to greater second moment of area and thus allowing it to better resist the bending. One advantage of the current large central hole is that it allows deployment of bone-void filler. Biomechanical benefits of cement augmentation have been demonstrated in several studies [83,84,88,98,102,118]. It may be useful, in future studies, to take the most stable configuration group from this study (F3), use bone cement in place of the 6.5 mm screw and investigate whether its augmentation further improves stability Screw Surface Profile One of the difficulties associated with the use of a locking plate is that it is not possible to know whether the screw is in contact with bone or not. This is because the screw head is locked to the plate and can mislead us into thinking that the fragments are held by the implant when they are potentially at risk of screw cut-out and penetration into the glenohumeral joint. Clinically, this effect is particularly noticeable with the varus subsidence of humeral head (poor neck-shaft angle). As a result, screw penetration is one of the most frequent complications reported for locking plates like PHILOS. [37,70 72] These complications exist for other fixation devices too especially for the blade plates where there is a 25% risk of penetration [179]. The use of threaded screws in the humeral head has been criticised based on the concerns that their threads could cause high stress concentration due to their relatively sharp angle. It is feared that this could facilitate the development of cracks especially near the subchondral bone, causing it to cut through the bone. Thus, it is a common surgical practice to keep the screws shorter such that their tips are five to eight millimetres clear from the articular surface. 175

176 This may reduce the risk of chondral breach but it also deprives the screws of the required purchase with the subchondral bone, a region known for high bone quality [180]. In response to this dilemma, the S3 plate offers the use of smooth pegs with the aim being that they will better distribute the stress at the bone-screw interface. Thus, in theory, the blunt smooth pegs can be better abutted to subchondral bone surfaces with less concern of bonethread penetration. Also, since the pegs have a larger core diameter and so larger surface area, they ought to offer better mechanical support [181]. The use of bone cement has also been shown to be an effective way of enhancing the bone anchorage of the plate [83,88,118]. With the use of cannulated pegs that allow the injection of bone cement, Raphael et al. [182] were able to demonstrate good function. The biomechanical study by Yamamato et al. [99] revealed that the S3 plate instrumented with pegs was stronger than PHILOS plate with threaded screws under cyclic cantilever varus bending. However, it was difficult to isolate the exact effect of the smooth pegs since, in their study, the plates not only differed in geometry and position on the humerus but the smooth pegs were also longer and closer to the subchondral bone. A more controlled study by Schumer et al. [53] involved the testing of construct groups (pegs and threaded) using the same S3 plate with same screw and peg length. No statistically significant difference was reported between the two constructs. In terms of the mechanical benefits of using larger core diameter screws and pegs, these finding partially resonate to our testing of the F2 configuration group where the removal of 6.5 mm diameter screw had statistically less effect on the varus bending stiffness than the removal of zone 1 screws and blade in F1 and F2. In our study threaded pegs were inserted in zones 1 and 2 of the S3 and while smooth pegs were inserted in zone 3. Removal of the smooth pegs in zone 2 led to 38.4% and 32.9% drop in the varus and valgus stiffness respectively. Removing threaded pegs in zone 3 led to a 35.3% drop in varus bending stiffness but 22.1% in valgus stiffness. Whether this is difference in the varus or valgus stability is due to the peg insertion needs to be investigated, ideally on cadaveric specimens due to their dependency on the bone quality Blade Medial support in plates, both in the form of screws and blade, targets the calcar region. The importance of inferomedial screws for minimising humeral head collapse in varus bending is well known, but less so on that of the blade [49,109,110]. In a similar fashion to the PHILOS plate s zone 1 inferomedial screws, the importance of the blade for varus stability in the Fx 176

177 plate was manifest when it was removed (F1), causing a 26.0% drop in varus mean stiffness, more than any of the other three directions. Swapping the Fx plate s blade with locking screws (F3) increased construct stiffness even further, not only in varus but also in the other three directions. This could be due to the larger second moment of area of the screws for the extension and flexion loading. Further, this might also be due to the indirect locking mechanism of the blade. Unlike most conventional blade plates, the Fx plate and its blade exist as two separate parts. In order to connect the two, the blade is placed into its slot on the plate and two grub screws are inserted on the plate, just above its blade s shoulders. In this way, it is held in its place by the interference between its shoulders and the grub screws connected to the plate. There is also a geometrical mismatch between the rectangular cross-section of blade s shoulders and the circular screw holes. Loosening of the grub screws would allow the blade to toggle and slide out. On the contrary, the two inferomedial screws do not rely on any grub screws and directly lock to the plate. In this way, they have longer effective working length and thus less toggling when subjected to bending. This issue of getting the blade, which has non-locking shoulders, to fix more rigidly to the plate should be addressed in future plate design Limitations of In Vitro Mechanical Tests The human shoulder is an extremely complex structure consisting of bones, muscles, joints, ligaments and tendons all of which work in synchrony to execute a range of motions around the glenohumeral joint, a range which is widest of all joint ranges in the human body. Designing and conducting in vitro biomechanical tests that accurately simulate the in vivo conditions of a joint with such a high degree of complexity as the glenohumeral joint is in itself a major challenge in the advancement of proximal humerus fracture treatment. Thus, a major limitation of the mechanical tests was the oversimplification of the loading conditions on the humerus despite the fact that the conditions used in the in vitro testing did have clinical relevance. One particular disadvantage of our experimental procedure is that the in vivo forces acting on the humerus are unlikely to be unidirectional and static. This problem was approached in our tests by loading the specimens along the two main anatomical planes and in the four directions sequentially, instead of simultaneously. To achieve this, the 5-mm displacement limit was set in elastic tests, keeping the specimens well within their elastic region. One possible way to tackle this limitation is to load the humerus in the magnitude and direction of the resultant force of the glenohumeral joint for a given shoulder movement which takes into consideration the muscles forces, bone-to-bone forces and connective tissue forces. The value of this glenohumeral resultant force can be determined from a subject-specific multibody 177

178 musculoskeletal model which enables visualisation of joint motion and a set of tools for calculating this inverse solution (Fig. 58). With the input of kinematic, kinetic and anthropometric data such as that obtained from 3D motion analysis and force plates, the model can calculate joint reaction forces and individual muscle forces using inverse solution. One such model has already been developed in our research group using OpenSim software (Simbios, CA, USA) and it is capable of calculation of glenohumeral resultant forces in different shoulder movements. Since the glenohumeral joint resultant force is a vector force with changing direction and magnitude during the course of shoulder movement, applying it in vitro is a challenge. While researching test rig designs, the author came across a recently developed test rig by Kelly and Bennett which has the capability to test specimens in vitro under coordinated six degree of freedom real-time load control [183]. Test rigs such as this could be used to apply complex vector forces such as glenohumeral joint resultant force. Even then, however, there remains inaccuracy in the loading conditions because the glenohumeral joint forces on their own do not fully depict the in vivo scenario. The humerus hosts insertion points for most shoulder muscles including deltoid, infraspinatus, supraspinatus, subscapularis and pectoralis major all of which pull at a different region of the humerus at different stages of shoulder movements. More accurate simulation of in vivo conditions would be obtained using a cadaveric shoulder complex with muscles attached and then pulling individual muscles to create desired movements. Results from the literature review have shown that this type of testing has already been conducted, notably by Voigt and her colleagues new [113,114,184]. They used a robot-assisted shoulder simulator which allowed differentiated application of defined muscle forces (such as that of the rotator cuff muscles) along with their physiological lines of action and in proportion to their respective physiological cross-sectional areas. Studies such as these involve cadaveric humeri, thus the issue of interspecimen variability has to be taken into consideration. There are also studies which did not involve the entire shoulder complex but instead, a humerus with non-cadaveric tendons attached [ ]. These tendons were made from webbing or leather straps and muscles in case of study by Kathrein et al. [118], pneumatic muscles have been connected to these tendons to pull. In one particular study, a synthetic humerus was used instead of a cadaveric one [119]. These results provide us with the potential to explore alternative means to apply accurate loading conditions. In the case of studies by Voigt et al., input values were from previous studies [128,129,146] however, measurement of the in vivo biomechanical contribution of individual muscle during shoulder movement is challenging and is a hot topic in the literature. For our case, the subject- 178

179 specific multibody musculoskeletal models developed in our research group can be used to calculate estimated values of muscle forces. When testing to failure, specimens were loaded under varus as varus malreduction and displacement has been reported to be one of the most common complications associated with locking plates. [37,185,186]. In the plastic tests, after 30 mm displacement, the load was still highest for the S0 group ( N), although only 1 N more than F3 group ( N). In the elastic varus results, the gap between their F 5 loads was over 9 N. This raises the questions as to what extent the elastic test results reflect the longer-term plastic results. It can be argued that this change difference is due to the 8-minute intermission and the stress relaxation. However, when the F 15a values (load before intermission) of the two groups are compared, it is found that difference had already shrunk to 5.38 N by then. Thus, in order to determine the long-term trends, plastic tests with larger displacements will be required. As for the statistical analysis, a fixed critical P value (0.05) was used to determine statistical significance within the experimental test data. However, since the number of sample size and the comparisons to be performed was large, a Bonferroni correction should have been made by dividing the critical P value by the number of comparisons. This may also be particularly useful if a much larger set of experiments involving multiple comparisons are to be performed in the future. By presenting the plastic results of each configuration group as percentage changes from its plate s control group, an interesting trend was noticed. In the S3 and the Fx, the percentage differences between the configuration groups and their control groups became smaller as we progressed from F 5 to F 30. For example, the removal of S3 plate s zone 2 lead to 38.4% decrease in stiffness during elastic tests but only 16% after 30 mm displacement in plastic tests. In fact, at 30 mm displacement, there was no statistical difference between the S1 group and the control group S0. Thus, in general, the mechanical influence of S3 and Fx plates on construct stability was found to decrease with the progression of the plastic tests. On the contrary, the opposite was found for all configuration groups of the PHILOS plate where the difference between the groups increased in the plastic tests. For example, where, in the elastic tests the largest percentage decrease in load was 28.5% (P1), was now over 39%, largest change ever recorded for any zone. This difference in the elastic and plastic test trends needs to be investigated further to determine why it occurs in S3 and the Fx plate but not PHILOS. Caution should be exercised when extrapolating these experimental laboratory findings to clinical situations. 179

180 Table 13. Percentage change in the elastic varus bending load (F5), F15a, F15b, and F30 of all configurations of the S3, PHILOS and the Fx plate, as compared to their respective control groups. NB: Cell background colours indicate the magnitude of the change. Configuration Varus F5 (%) F15a (%) F15b (%) F30 (%) S S S F F F P P P P One obvious limitation of our in vitro tests is that they involve synthetic humeri which may not be representative of the true mechanical behaviour of living, despite the fact that the synthetic bone had high and low foam density regions to represent the cortical and cancellous structures. In particular, the bone-screw interface may not have been modelled accurately in our tests, especially with regard to the in vivo variations in the local bone mineral density of the humerus. The choice to use synthetic humeri was based on a recent biomechanical study by Huff et al. [95] where the same polyurethane foam humeri as those used in this study were tested along with human cadaveric humeri. In the study, the results for the cadaveric specimen had large variations, due to their inherent biologic variability. Huff et al. reported comparable trends for loads and bending stiffness between cadavers and synthetic bones tested with the S3 and PHILOS plate under extension, flexion, valgus and varus bending. Possible improvements for future would be to perform the tests using cadaveric specimens with not only one but a range of bone qualities so that the performance of plates can be investigated especially for the elderly with osteoporosis (main patient population). This would provide more accurate values for in vitro measurements for validation should the bone of elderly with osteoporosis be required to be simulated. It must be noted that the Young s modulus of the stainless steel plates (193 GPa) used in these tests were several hundred times greater than that of the polyurethane foam ( MPa) from which the synthetic humerus is manufactured [152, ]. The bending stiffness (of the bone-plate construct) obtained during these tests were largely dominated by the flexural properties of the plate, leaving the humerus under-represented. Also, cadaveric humeri have a Young's modulus which is tens of times greater than that of these polyuerathane foams [190,191]. If a cadaveric specimen is used in future studies, the humerus can be expected to have more contribution to the construct bending stiffness. 180

181 In order to further investigate the performance of the proximal humeral plates, the tests should be expanded to more complex fractures, either on two-part fractures of other regions of the humerus or on three or more part fractures. This testing on reproducible, multiplanar fractures is particularly needed to find out whether the low screw number limits the use of the S3 plate in reaching the distant region of the humerus. In this study, plates were mostly zoned based on the screws proximity to the fracture site. As reported throughout sections , this made the interpretation of the results difficult, as testing for a zone meant testing for the effect of several design parameters at a time. One can imagine that this proximity-based zoning can only further complicate the interpretation of results if the tests are conducted in the multiple-part fractures. Then, it will have to be decided whether the screws should be zoned based on their distance from one particular fracture site or from all fracture sites. In our study, we inserted the screws of all plates up to the subchondral bone, to maximise the screw purchase. While this is a common practice for the insertion of S3 plate s pegs, it is advised that 5 mm clearance should be left to prevent the risk of screw penetration, especially in the PHILOS plate. Although we did not measure any screw penetration since the humeral head, fluoroscopic assessment and image intensifiers are used in the clinic to check for any penetration. Although the processing of 3D scanning could not be completed, the inter-fragmentary motion and strain distribution could be measured in future using the 3D motion analysis system and techniques such as digital image correlation. Furthermore, development of the FEA model simulating these in vitro tests could also provide the strain and stress distributions needed to assess the effect of design parameters such as plate and screw surface profile. 4.3 Conclusion By dividing the plate into screws zones and testing them one at a time, we were able to isolate their individual contribution to the stability of the bone-plate construct. However, testing for a zone meant simultaneous testing of several design parameters. This made the interpretation of results difficult. The S3 plate demonstrated superiority over the conventional locking plate (PHILOS) and hybrid blade plate (Fx), especially in elastic varus bending. We attribute this to its thicker crosssection and the 135 o inclination of its screws with respect to the humeral shaft which was similar to the sought neck-shaft angle. For both plates, out of all zones tested, removal of the 181

182 inferomedial support led to the largest decrease in varus bending stiffness. This is in line with the commonly-held view in the literature that medial support is important for varus stability. Furthermore, results showed that the type of medial support also matters. In the Fx plate, the medial support provided by inferomedial screws exhibited significantly superior extension, flexion, valgus and varus bending than that by the blade. The removal of the large 6.5 mm screw had a relatively low effect on the mean stiffness in all four directions than the blade or the inferomedial screws and demands further mechanical investigation. In general, the effect of a zone on varus and valgus stability was found to decrease with increasing distance from the fracture site. Screw pairs placed further proximal to the fracture gap played a significant role in extension and flexion bending stiffness. We attribute this to their non-parallel orientation which increased the second moment of area of the screws belonging to the zone. Hence, in appreciation of this role of these zones, it can be concluded that for general stability all four zones are critical as they have a synergistic relationship and clinical decisions ought to be made depending on the nature of the fracture being treated. It is hoped that findings of the present study will provide valuable information to the clinicians with the decision making involved in selecting the optimal number and configuration for a given fracture case. It is also hoped that the design choices discussed in this chapter, especially with regards to the location, orientation and geometry of the screws and the plate, will assist the design of better proximal humerus plates in the FEA based optimisation. 182

183 Chapter 5: Finite Element Model Creation and Validation 5.1 Introduction to Computational Framework The in vitro approach to investigating the biomechanical performance of proximal humerus plates, as reviewed in chapter 2 and later adopted in chapters 3 and 4, had several limitations. The cost, time and resource limitations very often dictated the experiment design. For example, in our mechanical tests (chapters 3 and 4), due to the limited number of the bone and plate specimens, only the zone-wise comparison could be performed. It is true that the removal or presence of zones represented the real-life decisions made by clinicians in the theatre. However, the removal of a zone also meant changing multiple design parameters simultaneously. This made it difficult to interpret the results and isolate the contribution of each design parameter. It is also known from the discussion of the mechanical tests (chapter 4) that the mechanical performance of proximal humerus plates is a complex interplay of many design parameters. If the design of the plate is to be optimised more systematically, in vitro testing of a large set of plate designs, each with a different combination of design parameters, is required. However, the high time and resource cost required for the manufacturing and testing makes it impractical. To ameliorate such limitations, it is necessary to resort to in silico testing techniques such as FE analysis. Advances in computational power have allowed FE analysis to emerge as a highly valuable tool in the field of orthopaedic biomechanics for analysing stresses and strains within structures such as bones and for implant design optimisation. It has the advantage that individual design parameters can be altered in isolation and tested without the issue of environmental or inter-specimen variations. It can be used to test proposed designs in parallel and help reduce the total number of design to be tested in vitro and in vivo. While FE analysis has been used widely for decades to investigate a wide range of fractures of the human body, its application for plate design optimisation of has experienced a significant increase in the previous decade. Early such work focused on biomechanical comparisons of different plates and screw configurations used in the clinic [192,193]. Cegonino et al. [192], in 2004, investigated the functional performance (displacement and stress distribution) of the femur after distal femur fracture and fixation by three types of implants: the Condyle plate, the less invasive stabilization system plate and the distal femur nail. Such comparative studies are still performed today as evident in a recent study by Zhou et al. [194] which compared the biomechanical stability and stress of a locking compression plate and a newly designed assembly locking compression plate, for middle femoral fracture, under slow walking and torsional loading. In fact, while very few studies have conducted FE analysis of the plate-based 183

184 treatment of proximal humerus fractures, those that do it, follow this same approach of comparing a few clinical configurations. Feerick et al. [195] investigated the effect of material properties and bone cement augmentations on four proximal humerus devices, including a locking plate. Yang et al. [196] and He et al. [197] determined the effects of medial support, in the form of calcar screws, cortical support and a secondary plate. Zhang et al. [198] studied the biomechanical effects of the styles of proximal humerus plate screw holes. An alternative approach to optimising implant design in silico is the parametric one. Here, different features of an implant are converted into design parameters. The choice of features to be parameterised is based on mechanical experience and knowledge, such as to select parameters that will have maximum effect on the performance. Next, a large number of FE models are developed and systematically analysed, each with a different parameter choice. While biomechanical testing at such a large scale in vitro will be impractical in most cases, this approach aims to take maximum advantage of the parallel computational capability of modern computers. A manifestation of this approach is the study by Kayabasi et al. [199] where the geometry of the hip prosthesis was described in terms of ten design parameters. An Approximate Design Optimisation algorithm was used to systemically select the values of these parameters in order to achieve the objective of minimising the maximum stress of the whole prosthesis. Similarly, Er et al. [200] developed an FE model of the ankle which had been treated with four different types of syndesmotic screws. Seven design parameters of the screws were identified and several FE models were developed to find the contribution of each design parameters to the stress in the screws. To help select the combinations of these parameters for FE model systematically, Taguchi s robust design method was used. It ought to be noted that while the number of FE studies on proximal humerus plates is already very few, their parametric optimisation is yet to be found in the literature. Fractures such as that of the proximal humerus are very complex in nature as they often occur at multiple locations and planes. This not only complicates their classification (Section 1.1) but also the design of plates required to treat them. While the option of insertion of screws, type of screws profiles and plate lengths are provided to the clinicians, most design decisions have already been made by the manufacturer prior to their sale. This one-size-fits-all approach may leave the clinicians with sub-optimal implant choices, especially for the treatment of complex fracture cases (Section 4.2.3). For example, the biomechanical study by Rose et al. [115] found superior performance with the PHILOS plate than the S3 plate in three-part fractures of the proximal humerus. They concluded that this is possible because the S3 plate had only two 184

185 pegs connecting the greater tuberosity to the rest of the head whereas the PHILOS plate had four. To address this, there has been a particular focus on the development of implants that are specific to patients' needs. [201]. Subject-specific treatment poses the challenge of keeping the implant cost low and successfully completing the process of patient data acquisition, FE model development and validation and the optimisation in such a short time-frame. Despite these challenges, optimisation studies which highly parametrise the plate design promise a good future [202,203]. This allowed Chen et al. to develop subject specific plates that respected the individual differences in bone morphology and fractures. They used a model of the typical femur to select surface features and parameters for the plate. A computer program was developed which allowed the user to specify the design parameters. Such a program could help accelerate the process of developing plates with complex geometries for the subsequent FE analysis. In light of this, we proposed a computational framework which could be divided into three distinct stages: reverse engineering of bone and plate geometry, creation of initial FE model and FE model automation and optimisation. In the first stage, 3D solid models of the fractured bone and the implant geometry are developed from sources such as patient scan data and 3D scanners, In the next stage, an FE model of the bone-implant construct is developed and validated with in vitro results. Once validated, the FE model is used in the final stage to conduct a parametric optimisation study. In this study, the geometry of the implant is parametrised based on a design parameter of choice. It was hoped that such a holistic framework would be used for the subject-specific treatment of a range of fractures. As an example of its implementation, this framework was used for the optimisation of proximal humerus plates to enhance their mechanical performance (Fig. 63). This chapter describes all the steps performed to develop the initial FE model of the fractured humerus that had been treated with the S3 plate, akin to the S0 configuration group in the mechanical tests. Loading and boundary conditions were similar to these of the in vitro elastic varus bending tests and the results for the in vitro testing of S0 were used for its validation. The aim of the subsequent parametric optimisation study was to use this model to investigate the effects of screw orientation on the varus bending load of the bone-plate construct. This optimisation study is described in in greater detail in the next chapter. 185

186 Figure 63. Computation framework, from the reverse engineering of the implant and bone geometry to the creation of the FE model and the subsequent automation and optimisation. Majority of the work was performed on three software: Mimics (green), Geomagic Wrap (blue) and Abaqus (red). 186

187 5.2 Three-dimensional Geometry Reverse Engineering A synthetic humerus of the same type as those used for in vitro experiments was obtained and scanned using a Computed Tomography (CT) scanner (SOMATOM, Siemens, Munich, Germany). The output image stack was in the Digital Imaging and Communications in Medicine (DICOM) file format. Medical processing software Mimics version 16.0 (Materialise, Leuven, Belgium) was used for segmentation: conversion of CT scan images into threedimensional (3D) models. First, the scan image stack was imported into the software and correct anatomical labels were defined. The image stack consisted of both the humerus and its intramedullary canal, with a noticeable transition between the cortical and cancellous structures in the humeral head. Concretely, segmentation involved highlighting in each image the pixels belonging to the humerus. To help achieve this effectively, Mimics offers tools such as thresholding whereby all pixels within the user specified Hounsfield range are added to the mask. Several standard predefined threshold Hounsfield ranges were available for selection of common scan materials such as cadaveric bone. However, this option was of little use in our case since our bone specimens were made of polyurethane foam and had different densities and grey values. For this reason, the minimum and maximum values for the thresholding tool were determined by trial and error until most of the humerus was highlighted. The remaining pixels were added to the mask manually, one slice image at a time. From experience, it was found that inserting the intramedullary canal at this stage is more likely to cause meshing problems in the FE model later. Using the Measure Distance and Measure Angle tools inside Mimics, the dimensions and locations of the intramedullary canal were measured for later referencing, prior to their filling. To discard the distal region of the humerus, all the highlighted pixels over 210 mm away from the head apex were removed from the bone mask, using the Multiple Slice Edit option. In a similar fashion, masks for the slices that were 50 to 60 mm from the head apex were also removed to simulate the two-part fracture. A duplicate of the mask was created in which the highlighted regions from slices below the fractures were removed. Likewise in the original mask, the highlighted regions from slices above the fracture were removed. Appropriately, the original mask was now renamed as humeral shaft. To transform the two masks into 3D models, the Calculate 3D option was selected and mesh quality was set to high, before clicking Calculate. The resulting 3D models were visually inspected to ensure that the anatomical features were being accurately represented. Finally, the surface meshes of the two 3D humeral head and shaft models were exported in the.stl file format (Fig. 64A). 187

188 Figure 64. Reverse engineering process of the humerus, from segmentation of bone geometry in CT scan image (A) to triangular surface mesh s pre-processing and simulation of a two-part fracture (B) The.stl files were imported into Geomagic Wrap 2014 (3D Systems, Rock Hill, SC, USA) for additional processing of the surface mesh. By its very nature, this processing was very timeconsuming as it required a lot of iterations of trial and error of selecting the appropriate processing parameters in order to achieve a geometry that was both anatomically accurate and could be meshed successfully for FE analysis. The.stl file uses triangular elements to describe the complex surface geometry. For computeraided analysis, especially that by FE analysis, the properties of these triangles, such as their aspect ratio plays a critical role in the analysis results. For example, elongated triangles with sharp edges angles can result in extreme differences in stress values. For this reason, equilateral triangles are preferred. However, often times, the quest of a more uniform triangle mesh demands an increase in the triangle count which can, in turn, increase the analysis time. Geomagic Wrap offers the Remesh command which re-triangulates the surface mesh to achieve a more uniform tessellation while maintaining the triangle edges at user-specified target lengths. Mesh was smoothed to remove the sharp edges that may act as unwanted stress risers during FE analysis. Geomagic Wrap also provides several smoothing options two of which were mainly utilised in this study: Remove Spikes and Sandpaper. The first allows the user to detect and flatten single-point spikes on the entire mesh or a highlighted area. The strength of smoothness was accreted using the Smoothness-Level slider. The subsequent meshing problem of the FE model showed that meshing errors were specific to certain regions of the bone. So, it was no longer necessary to smooth the model as a whole and risk further 188

189 loss of its anatomical accuracy. A more localised tool, Sandpaper, was used instead which is a freehand tool that allows manual smoothing of the region of interest. From the Sandpaper dialog box, the two sanding operation modes were used: Quick Smooth and Relax. In the former mode, the mesh is locally flattened by the removal of the triangles. The use of this mode minimum was kept as low as possible in order to preserve the original geometry. The latter allows relaxation of the triangle while keeping their population constant. Like the Remove Spikes option, the pressure of the sandpaper tool was incrementally increased until the meshing errors of the FE models were resolved. The Mesh Doctor command in Geomagic Wrap was extensively used throughout the processing stage, particularly at the end. The command automatically locates problems on the surface mesh such as spikes, self-intersections, small holes, small tunnels, small components, highly creased angle, as well as non-manifold edges. The Mesh Doctor command offers the option to perform a close inspection of each detected instance of the problem and provides a list of repair operations to treat it. Very often these problems corresponded to errors in the subsequent meshing of the FE model. It was, therefore, important to keep the keep the Mesh Doctor error count as low as possible. After the aforementioned processing of the surface models, a copy of the humeral head and shaft model was separately exported as.stl files to be used for backup and subsequent optimisation steps (Fig. 64B). Surface mesh models of both the humeral and shaft model were to be converted into Non-uniform rational B-spline (NURBS) surfaces. The Autosurface command opened a list of options to achieve this. From the options, geometry type was set as Organic since bone geometry was sculpted by nature, as opposed to the Mechanical geometry types which are most suitable for reverse engineering objects designed by computer-aided design. For the specification of the number of NURBS surface patches, the Autoestimate option was selected. The Surface Detail slider was initially set at level 4 (out of the maximum of level 7). It was aimed to keep the surface detail sufficiently high but keep their number low. So, the slider value was incrementally increased until the desired balance was achieved. The Fit Surfaces option was selected to generate the final NURBS surface model from the patches. Once satisfactory surfacing was achieved, the humeral head and shaft solid models were exported as separate.iges files. An S3 plate, identical to those used for in vitro experiments, was obtained and 3D scanned using a FaroArm laser scanner (Faro Technologies, Lake Mary, FL, USA). The scanner directly outputted the.stl file surface mesh that was imported in Geomagic Wrap for processing. Like the humerus model, the Remesh and Smoothing commands were applied as well as 189

190 the Mesh Doctor to perform routine mesh diagnosis. Since only the plate was scanned, the screw holes were visible. In order to help accurately place the plate on the humerus later, temporary cylinder features were created to indicate the position of each screw. The inner surface of the screw hole was highlighted before selecting the Best-Fit Cylinder option to fill the screw hole with a temporary, non-polygonal cylinder. Length of each cylinder was set according to the in vitro length of the screw it represented (Table 1). At this stage, the plate and the screw cylinders were temporarily grouped. The.stl surface mesh of the humeral head and shaft was imported into the plate-screw model. Using the drag handle in the Object-Mover tool, the plates-screw group was manually translated and rotated to correctly position it on the humerus. Protrusion of the screws out of the humeral head was particularly avoided. With the plate and screws resided correctly on the humerus, each cylinder s starting point and the direction vector of its axis were recorded. After the data acquisition, the duplicate of this.wrp file, containing humeral head and shaft as well as plate and screws, was saved and the screws cylinders were deleted as they were no longer needed. It was decided that the screws geometry ought to be added later and fill the screw holes at this stage. First, the surfaces around the screw holes were highlighted and deleted. While Geomagic Wrap offers several options to fill such holes, only the Single Fill tool with flat base geometry was used. This way, all plate holes were flattened. Care was taken to ensure that the geometry of the surfaces neighbouring the holes was not affected. The final plate mesh had no topological holes in it and the resulting surface mesh was then converted to a solid NURBS model and exported as a.iges file using the same commands and settings as that for the humerus models. 5.3 Finite Element Model Creation For the FE analysis, Abaqus CAE Standard 6.13 (Dassault Systemes Simulia Corp, Providence, RI, USA) was used. Solid.iges models of the humeral head, shaft and the filled plate were imported as 3D deformable parts. In the assembly module, instances of each of the three parts were created and their relative and the absolute positions and orientations were verified using the Distance Query tool. To create screw 1, a 3D-deformable part was created with a circular cross-section and was extruded to form a cylinder according to the screw s dimensions. An instance of this cylinder part was created in the Assembly module where a datum axis was created with the same direction vectors as that of screw 1 determined in Geomagic Wrap. With the use of the rotation 190

191 and translation, the screw instance was correctly aligned to the datum axis and the screws starting point. Likewise, the remaining nine screws were modelled and instanced in the assembly. In order to combine the plate and screws, the Merge/Cut Instances command was used with the Merge Geometry option selected. The final plate model consisted of the plate geometry merged with the ten cylinders representing the screws. Only this final plate model was left active in the assembly and the earlier models of the cylinder and the filled plate were suppressed. Based on the CT images, the intramedullary canal was cylindrical in shape with a diameter of 9 mm with its proximal-most point 3 mm from the humeral head apex, while the distal point passed through the shaft section. To model the canal, a solid cylinder part with these dimensions and positions was created to perform a Boolean cut from the solid models of humeral head and shaft. The plate instance was used to cut out the screw holes from the humeral head and shaft. This way, the final assembly consisted of three parts: plate with the cylindrical screws as well as the humeral head and shaft with the intramedullary canal and holes to accommodate the plate screws. In order to create a surface to apply the displacement to the shaft, the Sketch option was used, to create a surface as a projection, perpendicular to varus direction and in the shape of a square of 10 mm side length, centred 180 mm from the head's apex. These dimensions and position of the surface were similar to the contact surface of the semi-cylindrical actuator adaptor used in the in vitro tests. In the in vitro studies, a section of the humeral head up to 40 mm away from the humeral head apex was inside the cement block, but the surfaces around the plate were cleared. In order to simulate these boundary conditions as accurately as possibly, first, a plane parallel to the frontal (coronal) plane but with a 40-mm offset from the head apex was used to partition the faces on the head. The surfaces near the plate placement were left out to be more carefully selected later. Face partitions were also applied to isolate the sections of the humeral head humeral shaft in contact with the undersurface of the plate. With no additional changes to the geometry necessary for this analysis, the humeral head, shaft and the plate parts were now ready for analysis. When a NURBS solid model is created from the triangle mesh and its geometry modified, it comprises of surfaces that have been parametrically defined. For some geometries, such as those designed by computer-aided design and machining, these surfaces represent their part features such as faces and edges in which case it is necessary to keep them. For the creation of NURBS surfaces of organic geometries like bone or even the S3 plate, the resulting parametric surfaces do not necessarily correspond to a particular feature. So, when meshing 191

192 such an organic part, these surfaces impose undue constraints, forcing the mesh to conform to its boundaries. As a result, the final mesh is often poor, with unnecessarily high mesh density at the surfaces boundaries. Abaqus offers the Virtual Topology Toolset which allows the manipulation of the parametric surfaces to improve meshing. This is different from the direct editing of the geometry in the sense that here the part geometry is not changed and only the unnecessary meshing constraint applied by parametric surfaces are ignored, hence the name virtual. The Virtual Topology-Combine Faces command was used to combine the surface facts of the humeral head, plate and shaft that served a similar function in the mechanics of the analysis. For example, in the humeral head, all the surfaces above the partition line but outside the plate s range were combined to ease the selection of the boundary condition. Small surfaces at the regions of loading regions were merged to form a single surface. Since the boundaries of the screws were important to be preserved for the application of interaction properties, the cylinders faces were not combined with each other or with the plate body. Both the plate and the bone were modelled as a linear elastic isotropic material. According to the manufacturer, the S3 plate is made of stainless steel 316L and so Young s modulus and Poisson s ratio were set as 193 GPa and 0.3 [152,187,188]. For the case of the humerus, it had Young s modulus of MPa, well within the manufacturer s range ( MPa) for 8-30 PCF solid rigid polyurethane foam from which it had been made [189]. Poisson s ratio of 0.3 was selected for the humerus [189]. A tie constraint with the surface-to-surface discretisation method was applied to the surfaces of the screws and to their corresponding holes on the bone, with surfaces of the former set as master and the latter as a slave. In order to allow the application of a uniformly-distributed displacement across the square load surface, a reference point was created directly above the load surface, in the humeral shaft. A coupling constraint was then created with the reference point set as the point of control and the load surface as its dependency. In addition to the default initial step, a load step of static, general type was created in the Step module with the initial minimum and maximum increment size set as 0.1, 10-5, and 1 respectively. In the Field Output Request option, Current Nodal Coordinates (COORD) was in particular selected to output the position of the nodes during the analysis. In the in vitro experiments, the humeral head was rigidly fixed in the cement block and clamped to the base of the testing machine in order to restrict its translation and rotation. To model this, the boundary condition surface created from the virtual topology toolset was selected and 192

193 Encastre boundary condition was assigned to it from the initial step. A 5-mm displacement in varus direction was applied to the central point of the varus load surface (Fig. 65). Figure 65. Assembly of humerus and plate in the FE model and selection of the head boundary condition surface and the shaft surface to apply varus displacement (red arrow) The humeral head, shaft and plate parts were meshed using a 10-node quadratic tetrahedron (C3D10) element shape type. The global seed size was set as 1.5 for all three parts whereas local seed of 1 was assigned to the slave screw holes and the bone surface directly below the S3 plate. This ensured that the mesh density of the slave surfaces was higher than that of the master. The final mesh of the humeral head, shaft and the plate had elements. This choice of the seed size and the final element count was determined from a mesh sensitivity study where analysis on meshes of , , , and was performed. The differences between the values for the bending forces values (F 5) obtained for the FE models with the highest and lowest element was only 0.33% and was considered insignificant for the purpose of our analysis (Fig. 66). The duration for the two extremes, however, varied significantly from 20 minutes to 7 days. Since the model would be run for a large number of times, too long an analysis time would limit the number of angles that could be tested in the optimisation stage. Therefore, the medium mesh density, with a simulation time of approximately of 3 hours, was selected for FE models (Fig. 66, red). To allow calculations of the fracture gaps after the analysis, two node sets were created, one for either side of the fracture gaps on humeral head and shaft. After meshing, a job was created and submitted, utilising 5 processing cores on a 20 GB RAM workstation. 193

194 Figure 66. Plot of the varus bending load values and mesh element number obtained for the five FE models included in the mesh sensitivity study 5.4 Finite Element Model Validation Approximately 3 hours later, the job was successfully completed and the resulting.odb file was opened in the Visualisation module. A maximum bending force (F 5) of N was recorded. Next, the load and displacement values at all of the seven increments of this.odb file were noted. From the load-displacement data for all the S0 configuration group's elastic varus bending trials, the experimental loads at these increment displacements were interpolated. Plotting the mean and standard deviation of these experimental loads together with the simulation loads revealed that the two followed a close trend (Fig. 67). In addition to the force value, a more clinically relevant parameter was needed for comparison of the future constructs mechanical performances. Thus, the difference in the fracture gap before and after loading was calculated. This was implemented by comparing the position of the centroid of the node-set on either side of the fracture gap before and after loading (Fig. 68). Next, the distribution of von Mises stress on the humeral head, humeral shaft and the plate was inspected for qualitative validation (Fig ). As seen in Fig. 72, the plate had the highest value of maximum von Mises stress ( MPa) out of these parts, followed by the 194

195 humeral shaft ( MPa) and the head (5.456 MPa). This may be attributable to the superior Young's modulus of the plate. Stress was unevenly distributed in the humeral head, where the majority of the surface had very low stress but it was high around the insertion point of zone 3 screws and around the outlines of the boundary condition surface (Fig. 69A). Transverse and frontal view revealed further stress concentrations (Fig. 69B and 69C) on the back and top of the head, while the translucency option showed regions of high stress around the screw holes (Fig. 69D) due to the bone-screw interface [204]. Similarly, stresses on the humeral shaft were found to be high at the screw holes, particularly the most distal one (Fig. 70). The high stress concentration at the last screw hole near the end of the plate during bending is a known complication in the literature and it may be due to the differences in stiffness between the plate and the bone and the fact that this screw is the support closest to the loading area [169,205,206]. Because of this, there is a risk of further fracturing at the peripheral bone-screw junction, especially in patients with osteoporotic bone. As for the plate, stress was high at the screw-plate junction. Highest stress was found at the screw 7 head. Also, the sagittal view showed that the stresses were high at the section of the plate spanning the fracture site. This has been previously reported in several studies, particularly during the debate of optimum plate working length and rigid versus semi-rigid plates where the minimising of the stresses and the strains of this plate section have been discussed [162,163,167]. In their FE study, He et al. [197] fixed the humeral head and loaded the shaft in cantilever fashion, and they also achieved relatively high von Mises stress in the section of the plate spanning the fracture site, when the load was applied in a direction similar to the varus direction. While there was no experimental stress data to compare the model to, qualitative inspection of the von Mises stress distribution on the head, shaft and plate showed that the model was in good agreement with the theory and experimental results of other studies. This, with the attainment of peak load value very close to the experimental value, validated the use of this model for future in silico elastic varus bending testing. 195

196 Figure 67. Load at each step increment solved by the FE model compared with the mean in vitro biomechanical test loads Figure 68. Frontal view of the control FE model in loaded (A) and unloaded (B) state with the calculation of change in fracture gap 196

197 Figure 69. von Mises stress (MPa) distribution across humeral head in the standard FE model, shown from the sagittal (A), transverse (B) and frontal view (C and D) 197

198 Figure 70. von Mises stress (MPa) distribution across humeral shaft in the standard FE model (A-C) 198

199 Figure 71. von Mises stress (MPa) distribution across plate in the standard FE model (A-C) 199

200 Figure 72. von Mises stress (MPa) distribution across the bone-plate construct in the standard FE model (A-B) 200

201 5.5 Discussion Although the FE model was considered validated, its limitations must still be noted. For example, the humerus was modelled with isotropic material properties without consideration of regional differences in cortical and cancellous microstructures. Developing a model with more accurate microstructures and bone mineral density will allow us to more accurately determine the stresses and loads in the bone-plate construct. In our case, since the model was to be validated with the in vitro tests, the foam structure of the synthetic humeri also needed to be modelled. The polyurethane foam representing the cortical bone had a closed-cell foam structure while that of the cancellous bone had an open-cell foam structure. Sun et al. [207] performed uniaxial compression of three closed-cell foams (polymer, aluminium and fibre-reinforced) to determine their loading-unloading stress-strain profiles. Initially, the unloading modulus was found to increase linearly with compression in the elastic region but then it decreased due to the formation of more crush bands across the specimen. Beyond a critical strain, the unloading modulus increased again due to densification but it was still lower than that of the constituent material. Concretely, their study showed that the unloading elastic modulus was straindependent. This raises a question on our use of a single elastic modulus value to represent the material properties of the entire foam structure. There is also a lack of information on the polyurethane foam manufacturer's testing methodology and definition of elastic modulus. Sun et al. recommend using a linearly extrapolated curve from strains greater than the collapse initiation strain to zero strain for a more accurate calculation of the initial elastic modulus of a cellular material. In future, in vitro tests similar to Sun et al. can be conducted on cortical and cancellous polyurethane foam blocks to calculate their initial elastic modulus from their stressstrain profile. It should also be noted that the current study mainly dealt with the load-displacement profile and the stiffness values of the bone-plate constructs. The construct s stiffness was mostly dependant on the elastic modulus of the plate. Therefore, the dependency of the foam s unloading modulus on the strain, as reported by Sun et al., may not have had a significant effect on the construct stiffness. On the other hand, this dependency may have affected the local stress fields within the specimens, especially in the bone surrounding the screws. To deal with this uncertainty, further in vitro experiments are required to measure stresses in the specimen during bending. This experimental data ought to be used for quantitative verification and validation of stresses obtained in the simulation. Even then, the accuracy of the model 201

202 may not be sufficient since the bone is still modelled as an isotropic elastic model whereas in reality it is viscoelastic and has local variations in density and material properties. In the FE model, the cement block fixation was modelled simply as a rigid, fully-fixed boundary condition (no rotations and displacements). In reality, however, the block is an elastic body and is subjected to elastic deformation during the in vitro tests. This may have affected each construct's load-displacement profile and the stiffness calculated from it. To reduce this uncertainty, it is suggested that the cement block should be modelled with the elastic material properties of the cement. In such a model, the simulation values for the F 5 are expected to be lower than those currently obtained for the rigid boundary condition since the some of the applied load would also be transferred to the cement block. Due to the nature of the bending tests, when a specimen is being tested, the section of the cement block below the plate is subjected to compression while that above it is under tension. Since the cement has a compressive strength significantly higher than its tensile strength, this may have led to local failures inside the specimen. Further biomechanical tests are required to determine whether these failures significantly affected the construct's stiffness and stress distribution during our tests. Nevertheless, inclusion of the tensile and compressive properties of the cement into the FE model, would increase the accuracy of the simulation results. The elastic modulus value of the bone was adjusted in the parametric study to best agree with the experimental results. This way, the material properties were not independent of the results. Let's suppose that our simplification of the boundary condition as a rigid support was incorrect but we still reached a good agreement between the simulation and the experimental results by the means of varying the material properties. In that case, the boundary condition error would have been transferred to the mechanical error. A more robust approach would have been to obtain the material properties independently, from a different set of experiments. The bone-screw interface interaction for all screws was defined with a tie constraint to model the ideal bone-screw purchase, as described by Zhang et al. [198]. This is in contrast to our in vitro experiments where the proximal-most three head screws were partially threaded and the remaining three head screws were smooth. This can affect the stability of the bone-plate construct and the local stresses in the screws and screw holes. These screws could be better modelled by creating their thread geometry and applying frictional properties or by implementing the pseudo-threading techniques (implicit thread modelling) such as that described by Inzana et al [208]. 202

203 In order to model perfect locking of the screws to the plate, we merged the screws to the plate to form a single part. This was because pull-out of screws was not observed in our in vitro experiments. However, it should be noted that the post-operative screw pull-out has been reported in the clinic and thus, the screw-plate interface may need to be considered if the loading conditions of the model are changed [70]. If such a case arises, the screws may need to be modelled separately to the plate and with frictional properties applied at their surfaces of contact. Varus bending was applied in a cantilever fashion based on previous biomechanical studies to simulate the pull of the supraspinatus on the humerus [79,81,123]. Varus direction was selected for loading and optimisation in particular due to the high complication rate of varus collapse in the clinic [120]. The effects of changing the screw orientation on the construct stability in other directions such as valgus, extension and flexion must also be determined. In order to simulate the complex in vivo loading conditions more accurately, the FE model must include the bones, tendons and musculature surrounding the humerus and perform joint movements such as glenohumeral abduction. The current FE model involved loading of the humerus which was static in nature and within the elastic region of the bone-plate construct. Loadings under cyclic conditions and in the plastic region are required to determine the fatigue behaviour and the long-term in vivo biomechanical effects of the plate. Another limitation of the current study is that it is limited to a two-part fracture model. This did have the advantage of being easily reproducible in vitro and simple calculation of the fracture gap change in silico. However, in the clinic, the optimum treatment of complex, multi-planar proximal humerus fractures remains controversial. Therefore, it is important to not only simulate these fracture types but to also investigate how the change in the screws orientation can affect their reach to the extreme region of the humerus. This is particularly important for the optimisation of the S3 plate which has a low plate footprint and screw count [209]. Despite these limitations, our model was validated for the use in the subsequent optimisation study. 203

204 Chapter 6: Design Optimisation based on Finite Element Analysis 6.1 Introduction to Optimisation Study With the benefit of having conduced the in vitro tests systematically, the selection of design parameters for this optimisation study had been made easier. It is known that the stability of the bone-screw interface is crucial to further improve the biomechanical performance of the locking plates [209]. This can be achieved by changing a variety of design parameters such as the plate's geometry, and the geometry, number, position and orientation of the locking screws. In vitro tests showed that out of these, screw orientation can be used to effectively improve the construct stiffness (section 4.2.5). For example, the screws could be orientated towards the calcar region of the humerus, a region critical for humeral head stability against varus collapse [41,47]. Also, as discussed in section 4.2.5, directing the screws at a neck shaft angle of a healthy humerus may also have mechanical benefits. These advantages make the screw orientation an ideal design parameter to be optimised using iterative algorithms and parallel computation. Amongst the screw zones tested in the in vitro experiments, zone 2 had the largest effect on the varus bending load of the bone-plate construct, making it important to focus on screws 4 and 5 for optimisation. Divergence and height angle (θ h, θ d) of screws 4 and 5 are defined as the angles that their axes of direction make with their midpoint in the sagittal and frontal plane (Fig. 73). The objective of the optimisation study was to find a feasible combination of height and divergence angles for screws 4 and 5 that yields the minimum fracture gap change (ΔG) in the FE analysis. 204

205 Figure 73. Visual representation of the divergence angle, θd (A) and height angle, θh (B) angle of screws 4 and 5, along with the screws divergence (large black dot) and plate s midplane (dashed grey line) For each θ h and θ d combination, a new FE model was to be developed. The dimensions of screws 4 and 5 were fixed. Lengths of these two screws were kept same as those used in the initial FE model and the in vitro tests. The lengths, diameters and directions of the other four head screws were also kept constant. To keep the total number of models low, only integer angle increments were considered. 6.2 Finite Element Model Automation Manual creation of an Abaqus model using its graphical user interface is very time consuming and prone to human error, especially since the creation and the selection of so many surfaces are required for preparation of each model. In our case, we aimed to test FE models with a range of screw orientations, making it necessary to automate the preparation process. A common way of achieving this is by scripting the input files generated by Abaqus, using software such as MATLAB (MathWorks, Natick, MA, USA). This approach is very useful when the FE models differ from each other by changes to certain settings parameters but using the same geometry. In our case, however, the geometry of the plate and humeral head was also to be changed because screw angles were different. This necessitated the use of Abaqus 205

206 Scripting Interface to create Python scripts and gain a more direct control over the Abaqus commands (Appendix VI). The script began with the import of a template.cae model. This model contained all the parts, features and settings that were to be kept the same during the study such as the material properties, coupling interactions, step definition, field output requests and the varus displacement on the humeral shaft load surface. The model also contained five parts: two cylinders of lengths 47.5 mm to represent the screws 4 and 5, the humeral shaft with intramedullary canal and load surface and screw holes and a plate with all screws merged with it except screws 4 and 5. The starting positions of these two screws were hard-coded, as determined from Geomagic Wrap preparation step. From the user-specified angles and the starting position, the direction vectors of the two screws and the corresponding end points were calculated. Screws were instanced in the assembly and translation and rotation commands were used to position them according to their starting and end points. They were merged with the instance of the plate and the resulting plate was used to cut out the screw holes from the humeral head instance. Stainless steel and bone-material properties were assigned to these newly-created parts. The time-consuming task of the manual selection of the screw surface and the corresponding screw hole surfaces was automated by using the getbyboundingcylinder function. Surfaces that were to be merged using Virtual Topology toolbox, were identified for each part using functions such as boundary boxes and Boolean surface selection. Attention was paid here to check for potential special cases. The resulting surfaces were merged and used to apply the appropriate boundary conditions, constraints and interactions. Meshing was performed with both global and localised seed settings to the parts and slave surfaces, respectively. Finally, the script created the job and wrote an input file to be used for subsequent analysis. An additional script was also created to perform the post processing calculation of the fracture gap, using the.odb files produced by the analysis. It started by calculating the centroids of the nodal sets of surfaces on either side of the fracture gap. The difference in the centroids position along the shaft axis before loading was taken as the initial fracture gap. Likewise, the node sets centroids were recalculated after the application of 5 mm displacement and their difference was taken as the final fracture gap. Finally, the difference between the initial and the final fracture gap was calculated and outputted by the script as the change in fracture gap. With the development of the two script files and the template model, the necessary work was now in place to allow automatic creation of the FE model with any user-defined orientation of 206

207 screws 4 and 5. The final script reduced the preparation time from 8 hours to only 30 seconds, starting with the user input of the screw angles and ending with the creation of the job. 6.3 Optimisation Method In order to only perform FE analyses onto clinically relevant height and divergence combinations and save computation time, a Python script was written inside Geomagic Wrap to test for special cases first. As shall be described in section 6.3, the script tested whether the screws were out of the humeral head, collided with the trajectories of the other screws or could be made of sufficient length before contacting the subchondral bone (Fig. 74). Source code has been provided in Appendix VII. All screws were simplified as cylinders. A triangle surface mesh of the humeral head, developed by segmentation of CT scan, was imported into Geomagic Wrap. The screws starting position and direction vector and diameter, determined in the previous reverse engineering stage was inputted. To prevent screws 4 and 5 from coming too close to each other, and other screws, their diameter was increased by 1 mm. Also, the ranges of the divergence and height angle (θ dmin, θ dmax, θ hmin, θ hmax), 0-90 o, were inputted along with the clearance of the screws tip from the subchondral bone (5 mm). The latter was subtracted from the final calculation of maximum screw length since screws are often placed approximately 5 mm short of the subchondral bone in the clinic to reduce the risk of screw penetration into the glenohumeral joint. Two feature arrays were created, with the first consisting of screws 4 and 5 (active screws) and second of the remaining screws (fixed screws). From their starting position and the direction vector, maximum length of the fixed screws was calculated to determine their full trajectory inside the humeral head in case a longer length is selected in future (calc_maxscrewlen()). This calculation was implemented by modelling the screw as a ray directed in the screw direction and finding its intersection point with the triangles of the humeral head mesh. Due to the symmetrical nature of the divergence angle, the triangle mesh was split in half (split_mesh()) in the frontal direction and allocated to each of the two screws accordingly. This allowed faster calculation of the maximum length. To begin calculations, the divergence angle was initially set as the minimum of the input of its range. For the special case of angle θ dmin being greater than the angle θ dmax, the script was ended. However, more commonly, θ dmin 207

208 would be less than θ dmax at this stage and from the divergence angle and the starting position from the input, direction vectors of screws 4 and 5 were calculated. 208

209 Figure 74. Flowchart of the Geomagic Wrap script, highlighting the key inputs, decisions, functions and loops required for the calculation of feasible height and divergence angle combinations within the user-specified angle ranges 209

210 Taking into account the screw diameters and the height angle range, a conic inspection path was cut out for each screw. Concretely, this is the section of the humeral head that the screw of infinite length will come in contact with, at a given divergence angle. Thus, for each divergence angle increment, an inspection path was determined first, to be used for calculation for each increment of height angle (get_inspectpath()). This inspection calculation was to narrow the region of interest for each height angle increment and reduce the overall computation time. For comparison, a ten-fold decrease in computation time was found by using this technique instead of a full search of the entire mesh of the humeral head for each height angle increment. After obtaining each screws inspection path for a given divergence angle, all the height angle increments were looped through from θ hmin to θ hmax. For each height angle increment, the maximum possible lengths of screws 4 and 5 were first calculated and updated (calc_maxscrewlen() & update_screwlen()). The special case of screws being outside the mesh was also tested. If the majority of the screw length (greater than 30 mm based on pre-calculations) was outside the mesh, an out of mesh message was outputted and the script skipped to the next height angle increment. This problem often occurred for cases where very large divergence angles were selected. If the screw was indeed inside the humeral head, then, all six screws were modelled as line segments with the same length, starting point and direction as them. This was so that the shortest distance between the line segments could be calculated in the subsequent collision detection steps, based on the works of Lumelsky et al. [210]. In three dimensions, line segment-line segment collision detection with set distance constraints models the capsule-capsule collision (Fig. 74). Capsules differ from cylinders in that they have hemispherical tips, as opposed to flat ones. This was a safe assumption since all the screws were of their maximum length and their tips already touched the triangular mesh. Capsule-capsule collision detection is both mathematically and computationally simpler than cylinder-cylinder collision [ ]. Simply put, if the shortest distance between the line segments of any two screws was less than or equal to the sum of their radii, they collided. First, the collisions between screws 4 and 5 were tested for, as this was more likely to occur due to their increased diameter. Next, collision between each of the active screws and the fixed screws was tested. If a collision was detected between any screws, the rest of the collision testing was skipped, the results outputted and the scripted proceeded to the next increment of height angle. To account for the subchondral bone clearance, 5 mm was subtracted from all the maximum screw length. Eventually, when all increments of height were completed, the process was repeated for the next increment of divergence angle until the maximum divergence angle was reached. From the program output, contour plots were plotted, showing divergence and height 210

211 angle in abscissa and ordinate and the screws maximum length in colour. Boolean search was used to remove any divergence and height angle combinations that yield maximum screw lengths of less than 50 mm for either of the two screws. This way, only cases where maximum screws length of greater than or equal to 50 mm were left. This value was from experience and the understanding that the screws with long lengths can reach the more distant regions of the humeral head in the treatment of complex fracture cases. The resulting map from the script output consisted of one region of feasible combinations, spanning 2-30 o in divergence angle and o in height angle and consisted of 538 points (Fig. 75). The corresponding divergence and height angles along with the maximum length were exported for the finite element analysis and the optimisation steps. Figure 75. Contour plot of 538 feasible combinations of height and divergence angles (yellow), as determined by Python script in Geomagic Wrap 211

212 A list of all the 538 height and divergence angle combinations was created from the Geomagic Wrap. They were then inserted into the first Abaqus model preparation in groups to create the 538 Abaqus models and their input files. To ensure that the models were correctly created, they were manually inspected and double checked in the Abaqus Graphical User Interface. None of the 538 files was found to be miscreated. The input files were then submitted in batches to the Computational Shared Facility at the University of Manchester. Each job was allocated 5 cores and 20 GB RAM. The F 5 data for the.odb files was outputted and the second, post-processing script was used to calculate their change in fracture gap (ΔG). 6.4 Results Fracture Gap Change and F5 Load To ease the comparisons, the load (F 5, N) and the fracture gap change (ΔG, mm) in the FE model (standard S3 plate) developed in the last chapter acted as the base line to compare the 538 models of the optimisation study with. For each model, the percentage change in ΔG and F 5 from the standard model was calculated and presented as a contour plot (Fig. 76 and 77). Since the objective of this optimisation study was to minimise the fracture gap change, the optimum solution (16 o divergence angle and 33 o height angle) had 4.686% lower fracture gap change (0.156 mm) than the standard FE model. The worst solution (10 o divergence angle and 61 o height angle) had a fracture gap change (0.167 mm) which was 1.926% higher ( N) than the standard model. As for the loads (F 5), the opposite trend was observed where the optimum solution had the highest load values ( N) and percentage change from the standard FE model (5.707% higher) while the worst had the lowest ( N, 1.687%). In the contour plots, a general trend was observed where there was a lower percentage fracture gap change and higher percentage peak load with a reduction in the height angle (Fig. 76 and 77). For the changes in divergence angle, however, no significant difference was observed. In fact, the optimum solution was also the one with the lowest height angle (32 o ). Supporting trends were observed for the F 5 contour plot (Fig. 77). By superimposing the design of the standard and the optimum plate (16 o divergence angle and 33 o height angle), similarity in the orientations of their screws 4 and 5 was visually highlighted (Fig. 78). 212

213 Von Mises Stress Distribution Contour Plots The distribution of the von Mises stress on the optimum was similar to that in the standard model (Fig 79-81). However, in terms of magnitude, it had lower maximum von Mises stress in the humeral head and the plate but higher in the shaft. In fact, out of all the 538 models, this model had the highest value for maximum von Mises stress ( MPa). Humeral Head For each model, plotting the maximum von Mises stress in the head, shaft and plates of all 538 models in form of contour plots, highlighted several trends. With the exception of a few models, the majority of the models had maximum von Mises stress of the humeral head below the MPa value achieved with the standard model (Fig. 82). In this contour plot, there is a visible trend where the maximum von Mises stress increases with the increase in height angle. Not such trend with visible for the divergence angle. Humeral Shaft The trends for the humeral shaft maximum von Mises stress with respect to the height and divergence angle (Fig. 83), were similar to that for the F 5 load (Fig. 77). This strongly correlated the two since the model with the highest F 5 load (16 o divergence angle and 33 o height angle), also had the highest stress in the shaft. The model with the lowest values for maximum von Mises stress (10 o divergence angle and 61 o height angle) of MPa was still higher than that for the standard model ( MPa). Plate When comparing the maximum von Mises stress in the plate, the values ranged from MPa (24 o divergence angle and 48 o height angle) to MPa (21 o divergence angle and 24 o height angle) with both extremes closer to each other in terms of orientation (Fig. 84). However, contour map showed that this disparity was due to the few exceptional cases. When the results were sorted in the order of descending maximum von Mises stress values, it was found that the value for the tenth highest (23 o divergence angle and 45 o height angle) was MPa, much lower than the MPa maximum. Thus, a limit of MPa was imposed on the contour plot to find if a trend is observed between this maximum stress and the height and the divergence angle (Fig. 85). As can be seen in Fig. 85, that was not the case in general. However, a row of high maximum stress values was observed at height angle of 44 o. As compared to maximum plate von Mises stress for the standard model ( MPa), most models had lower stress than it. 213

214 Position of Nodes with Maximum von Mises Stress To further investigate these trends from the contour plots, or rather the lack thereof, the initial position of the maximum von Mises stress nodes for the head, shaft and plate of all 538 models were extracted from the.odb files. This coordinate data was exported as point cloud data which was then imported into Geomagic Wrap to be superimposed on the.stl geometry of the standard model (Fig. 86). Concretely, the points represented the position of the maximum von Mises stress nodes for the 538 models. Presenting the data in this manner made it clear that the position of the maximum von Mises stress nodes changed from model to model, especially for the head humeral head and plate. Humeral Head On the humeral head surface, the maximum stress points were found to be around the outlines of the boundary condition surface, the region closest to the screw 4 tip and also around zone 2 screw holes. As shown in Fig. 82, out of all the models tested, three had noticeably higher maximum humeral head von Mises stress value of MPa (5 o divergence angle, 58 o height angle), MPa (9 o, 60 o ) and MPa (22 o, 56 o ). Inspection of their.odb file revealed that they all had maximum stress point around the entry of the screw hole 5 (Fig. 87A). There, it was clear that the stress was unevenly distributed since most of the humeral head surface had stresses well below this maximum value. A cluster of points was also found inside the humeral head, more specifically the region between the tips of screws 1 and 6 (Fig. 86B), indicating that this is also a potential site for stress concentration. Since these two screws were noticeably close, a plausible explanation is that their proximity crushes the bone in-between. Indeed, when the stress distribution of the model with the lowest maximum von Mises stress (1.670 MPa, 16 o divergence angle and 34 o height angle) was analysed, the region between screw 1 and 6 was found to be under relatively high stress (Fig. 87B). Humeral Shaft For the humeral shaft, all models had the maximum stress at the same location: the edge of the distal-most screw hole (Fig. 87C). This consistency may explain why more clear trends were observed for the humeral shaft contour plot (Fig. 83). 214

215 Plate Like the humeral head, the positions of the nodes with maximum von Mises stress in the plate was varied (Fig. 86D), possibly leading to the variation in the contour plots (Fig. 84 and 85). It is noteworthy that nodes were located at the plate-screw junction of either screws 4, 7 or 8, suggesting the importance of these three screws in varus bending. Comparison of the stress distribution of models with highest and lowest maximum von Mises stress values in the plate showed that while the location of their maximum stress nodes was same (plate-screw junction of screw 4), their stress distribution (Fig. 88). The model with the highest von Mises stress had a very small area with extreme stress while the stress in the majority of the plate was less than 200 MPa. (Fig. 88A). On the contrary, for the model with the lowest von Mises stress, there was a larger distribution of stress, albeit with noticeably high stress on the plate screw junction and the plate section spanning the fracture centre. The top eight models with marginally higher maximum plate von Mises stress value had values ranging from MPa (Fig. 84). All these models had height angle of either 45 o or 46 o with divergence angle ranging between 21 o and 27 o. Their point of maximum stress was located at the junction between screw 5 and plate. Setting the range of the contour plot to MPa (Fig. 85) revealed several other screw angle combinations that also had high maximum von Mises stress values. Noteworthy examples include (7 o, 53 o ), (20 o, 36 o ), (11 o, 59 o ), (20 o, 38 o ) and (3 o, 56 o ). Unlike the top nine, these combinations had their maximum stress points located at the junction of screw 8 and the plate. On the contour plot, they were often surrounded by models with relatively low maximum stress values (Fig. 85). The maximum stress points of these neighbouring models were often located at the junction of the plate and either screw 7 or 8 (Fig. 86D). 215

216 Figure 76. Contour plot showing the fracture gap change of each height and divergence angle combination tested in the optimisation study, as a percentage difference from the standard FE model Figure 77. Contour plot showing the varus bending load of each height and divergence angle combination tested in the optimisation study, as a percentage difference from the standard FE model 216

217 Figure 78. Frontal (A) and sagittal (B) view of the superimposition of the standard (blue) and the optimum Figure 79. von Mises stress (MPa) distribution across humeral head in the optimum FE model, shown from the sagittal (A), transverse (B) and frontal view (C and D) 217

218 Figure 80. von Mises stress (MPa) distribution across humeral shaft in the optimum FE model (A-C) 218

219 Figure 81. von Mises stress (MPa) distribution across plate in the optimum FE model (A-C) 219

220 Figure 82. Contour plot showing the Maximum von Mises stress (MPa) on humeral head of each height and divergence angle combination tested in the optimisation study, as a percentage difference from the control FE model Figure 83. Contour plot showing the Maximum von Mises stress (MPa) on humeral shaft of each height and divergence angle combination tested in the optimisation study, as a percentage difference from the control FE model 220

221 Figure 84. Contour plot showing the Maximum von Mises stress (MPa) on plate of each height (stress range set at MPa) and divergence angle combination tested in the optimisation study, as a percentage difference from the control FE model Figure 85. Contour plot showing the Maximum von Mises stress (MPa) on plate of each height (stress range set at MPa) and divergence angle combination tested in the optimisation study, as a percentage difference from the control FE model 221

222 Figure 86. Locations of the nodes (red) with maximum von Mises stress (MPa) on the humeral head (A and B), shaft (C) and plate (D) from the 538 FE models 222

223 Figure 87. Stress distribution of models with highest (A) and lowest (B and C) peak von Mises stress (MPa) in the humeral head and shaft 223

224 Figure 88. Stress distribution of models with highest (A) and lowest (B and C) peak von Mises stress (MPa) in the plate 224

225 6.5 Discussion Contour plots of peak loads and fracture gap against the height and divergence angles show that both the peak load and the fracture gap changes are more sensitive to height angle than divergence angle (Fig. 76 and 77). We understand this to be due to the fact that, unlike divergence angle, height angle directly corresponds to the length of the screw along the varus loading direction. Neck-shaft angle (or head inclination angle) is defined as the angle made by the anatomicak neck with the humeral shaft in the frontal plane. It is used in the clinic as a method of determining the stability of the humeral head against varus collapse. For healthy humeri, this angle is approximately 135 o whereas an angle less than 100 o indicates potential failure [ ,214]. If we allow for the minor changes in the placement of the S3 plate, this angle is similar in nature to the height angle. In fact, after the addition of 90 o, the height angles for the optimum designs are found to be near the range of healthy neck-shaft angles. While a direct relationship between the two angles is yet to be determined in isolation, this suggests that screws 4 and 5 of the S3 plate should be orientated along the healthy range of neck-shaft angle. In both the standard and the optimum S3 plate design, screws 4 and 5 are directed towards the inferomedial region of the humerus. In vitro and in vivo studies have shown that the mechanical support of this region is critical for preventing varus collapse. This can be attributed to the relatively high bone mineral density in the region, leading to the higher pullout strength of the screws. However, since the bone was modelled as a solid, isotropic material, the improvement in the mechanical performance is due to the screws orientation and not the local bone mineral density. A more accurate modelling of humerus geometry and material properties is required to better investigate the biomechanical effects of orientating the screws towards the inferomedial region. Despite the limitations of the FE model, the current study successfully laid out a workflow for future optimisation studies, starting from CT images of the humerus and ending in the final contour plot. Automation of the pre-processing of each FE model significantly reduced its preparation time. Two stages remain to be time-consuming. First, the preparation and validation of the initial FE analysis require automation. It should be noted here that while several solutions for medical image segmentation have been implemented in the literature and commercial software, preparation and validation of the initial 3D meshable FE model requires significant human intervention [215,216]. Second, the issue of computation time needs to be addressed and decisions have to be made to simplify the model without compromising its 225

226 validity. Overcoming the aforementioned challenges would not only allow fast, patient-specific treatment but also guide clinicians in the clinical decision-making process. Findings of this study could be implemented at various levels. At the most basic level, since the results show the importance of height angle on construct stability, the effect of the height angle could be changed indirectly by adjusting plate s proximal-distal placement of the plate along the humeral shaft. This can be done using the currently available S3 plates. Great care should be exercised in the clinic especially when checking the position of the Kirschner wires and adjusting the plate s height. Next, the results could be implemented at a screw-level. Given the advantages of the locking technology, many locking plates such as the S3 plate remain to be fixed-angle devices. This creates a challenge for a clinician aiming to change screw orientation in order to, for example, better direct the screw towards a fragmented region of the humerus. In proximal humerus plates such as the Proximal Humerus Internal Locking System (PHILOS), screws are organised in both convergent and divergent mode to allow better coverage. However, the orientations of the individual screws are still fixed. In response to this, plates such as Non- Contact Bridging (NCB) plates were developed that have poly-axial locking screws [50]. This allows the clinicians to more accurately direct the screws in the optimum directions and save resources and time since the design of the plate itself remains unchanged. The contour plot can be split horizontally into two regions using the 50 o height angle as an approximate baseline, where the angle combinations below the baseline perform better than the standard. The sheer number of these combinations suggests that further optimisation of the screw orientation is possible. By selecting one of these combinations, an additional optimisation study on the other screw zones of the S3 plates could be performed to achieve plates with better mechanical performance. Also, while the current definition of the divergence angle assumes symmetrical divergences of screws 4 and 5, it can be expanded to also include asymmetrical divergence where the two screws diverge independently. In our model, we only assumed symmetrical mirrored divergences along the plate midline. At a more advanced level, the optimisation workflow developed in this study can be used for optimisation of other design parameters in addition to the screw orientation for better bonescrew interface: plate s geometry and the locking screws geometry, number and position. The time and resource advantage of in silico testing such an FE analysis come at the cost of simplification of the real conditions especially due to the requirement for running the simulation several hundred times. Implementing the results at the screw and the advanced level will require running a significantly larger number of FE models. This is particularly true if we plan 226

227 to change multiple parameters (in addition to height and divergence angle) and perform multiobjective optimisations studies where, for example, we may wish to not only minimise the fracture gap change but also the peak von Mises stress values. In such cases, the brute-force approach applied in the current study may be inefficient as the number of simulations would be too large, assuming that the computation time per simulation remains the same. Neural Networks and optimisation algorithms such as genetic algorithms can be used to achieve a global optimum solution faster [217,218]. For example, as mentioned before, Kayabasi et al. and Er et al. used the Approximate Design Optimisation algorithm and Taguchi s robust design method, respectively, for selection of design parameter values [199]. It should also be noted that these two FE-based optimisation studies were conducted on the hip and ankle, thus the framework proposed in this study also has the potential to be used for the design optimisation of implants of other parts of human body. Custom-made implants for femur, tibia, hip and craniofacial applications have been successfully developed in the literature [ ]. Thus, with the modern advances in the additive manufacturing and rapid prototyping, the 3D printing of the final patient-specific design obtained from this multi-objective optimisation can be manufactured based on the clinical demands. 6.6 Conclusion In conclusion, the framework introduced in the last chapter was successfully implemented for the case of two-part proximal humerus fractures. A parametric optimisation study was able to determine the height and divergence angles that yield the minimum fracture gap. The framework is potentially capable of being implemented for not only the proximal humerus fractures but also that of the other regions of the human body and for performing multiobjective optimisation of several design parameters. However, achieving this may necessitate further improvement to the accuracy of the FE model, more automation of the pre-processing steps of the FE model and also the use of more efficient algorithms to test a large number of models in time. 227

228 Chapter 7: Finite Element Based Study of Screw Surface Profile 7.1 Introduction FE models can be used to investigate clinical problems, in addition to their use for implant design optimisation. The study described in this chapter investigated one of the leading complications associated with proximal humerus plates: perforation of screws into the glenohumeral joint. Clinical literature is laden with reporting of cases of both primary and secondary screw perforation, with up to 23% complication rate [39,44,56,71,72,223,224]. Several factors have been identified in the literature that can increase the risk of primary screw perforation. It is noted that when locking screws are being inserted, they always feel tight because of the locking mechanism and the true purchase of the screw in the bone cannot be felt, creating a deceptive sense of security [159]. Furthermore, insertion of screws, particularly the diverging and converging ones, into the spherical surface of the humeral head can make it difficult to correctly assess the screw tip position on orthogonal views [225,226]. This situation is made worse by the limited tactile feedback of the drill bit when drilling into poor quality, osteoporotic bone [225,226]. The longer screw holes created by this over-drilling can be left unnoticed during the surgery and may lead to secondary screw perforation. Although primary screw perforation is associated with technical errors and considered preventable [44], secondary screw perforation can still occur. This is especially reported with the use of locking plates, as a consequence of humeral head collapse where the locking screws cannot back out and perforate into the joint instead (Fig. 89) [227]. If left untreated, it can result in extensive erosion of the glenoid surface and necessitate hemiarthroplasty or total shoulder arthroplasty as a salvage procedure. One solution may be to use non-locking screws in plates, but then there is the risk of screws disengaging from the plate and affecting the mechanical stability of the entire bone-plate construct [228]. 228

229 Figure 89. Three-months post-surgery varus collapse of humeral head and secondary screw penetration [260] Depending on the situation, head screws of locking plates have two roles. First, they are to provide structural support to the humeral head, preventing its collapse. They achieve this by resisting the cantilever bending moments experienced during the loading of bone. Theoretically, bending strength of the screws has a cubic relationship with their core diameter during their cantilever loading [229,230]. Screws can also provide structural support by improving their purchase with the surrounding bone and thus the pull-out strength, particularly in locking plates where the bone-screw interface is of critical importance [41,47,106,120,231]. Their pull-out resistance is dependent on the volume of bone between their threads so it can be controlled by altering their surface profile (pitch, depth and the shape of the thread) or by simply increasing their length [229,232]. Screw purchase can also be improved by orienting them towards regions of high bone quality such as the subchondral region in case of the humerus [178]. The second role of the head screws is to prevent damage to the articular surface, particularly after humeral head collapse. With the aim of achieving this, it is a common surgical practice and manufacturers recommendation to leave the screws shorter (typically 5 mm) from their maximum possible screw length at the subchondral region [233]. However, this could also compromise the screws ability to provide structural support due to their reduced contact with the high density cancellous bone of the subchondral region. One approach to preventing articular surface damage is to modify the screw surface profile, as was the case with the screws provided with the S3 plate. While screws of most proximal humeral plates have a threaded surface profile along their shafts, the S3 plate offers the option 229

230 to insert locking smooth pegs and threaded pegs (Fig. 90A). It is believed that these blunttipped smooth pegs would cause less damage to the glenohumeral joint than the threaded screws [53,181]. As for the structural support to the humeral head, Badman et al. [234] likened the smooth pegs role to that of rebars in concrete. Due to the lower risk of perforation with their use, smooth pegs with lengths up until subchondral abutment can be used. While the smooth pegs can afford more contact with the subchondral bone this way, their purchase with the surrounding bone is not as direct as the threaded screws. On the other hand, the smooth profile does allow pegs to have a greater core diameter than threaded screws, which can provide them with the bending strength required to prevent head collapse. Another approach is to include the smooth surface and the large diameter of pegs with the direct bone purchase of threads into a single screw, with the aim of reaping the best from both profiles. Such a threaded peg is also available for insertion into the S3 plate and has a peglike surface profile from screw head to midway across the screw shaft length after which it is threaded (Fig. 90B). Figure 90. Smooth- (A) and threaded- (B) pegs fixation options provided by the S3 plate Regarding the dilemma of pegs versus threaded screws, there is no general consensus in the clinical and the in vitro biomechanical testing literature as to which fixation type is superior [53,99, ]. Yet unsolved is the question of the biomechanical effects of threaded pegs, on proximal humerus fractures. It is also not known whether this choice of 50% threading provided by the threaded pegs, is an optimum for all screws entering into the humeral head. As for the in silico mechanical studies, no such study exists to our knowledge that supports a 230

231 preference for smooth pegs, threaded pegs or threaded screws. Due to the lack of this information, choice of screw type is currently left to the clinician s discretion. This study had two aims, both to be realised using the FE model developed in chapter 5. First, it aimed to compare the varus bending stiffness of proximally fractured humeri that have been treated with an S3 plate that had threaded screws in all head screws to that which had a smooth peg. Second, it aimed to investigate the effects of different percentages of threading in each S3 plate head screw on the varus bending stiffness of proximally fractured humeri. To fulfil these, a method of pseudo-threading from a previous FE study was first investigated (Section 7.2). Next, this pseudo-threading method was implemented in the FE model developed in chapter 5 and the resulting model was validated against the in vitro biomechanical test results (Section 7.3). Finally, the validated model was used to carry out the main FE study based on the two aims (Section 7.4). 7.2 Pseudo-threading in Homogenous Cylinder Bone One of the limitations of the FE model developed in chapter 5 was the idealised modelling of the bone-screw interface. Many FE studies in the literature follow this approach, where the bone-screw interfacial surface interactions are either perfectly bonded, which is more suited to modelling an osseo-integrated bone-screw interface than an early interface, or with frictional contacts [196, ]. Further, these interactions are typically applied to the entire screw shaft surface. As the screw s threads interact with bone differently to areas between the threads and thus have different local peak stains, the analyses of the screw modelled in this way is less sensitive than if interactions were applied to specific regions on the screw surface, to reflect their true behaviour. It is vital that modelling of the screw surface is as true a representation of a physical screw surface as possible in order to assess the performance of the bone-plate construct more accurately. Failure to do so is likely to yield different stress patterns and reaction force results that may not reflect what would occur in practice, thereby rendering the results less accurate. Although it would be most realistic to model the geometry of screws with fine threads, the computational cost and difficulty in meshing are far greater than approximating the geometry of a screw as a smooth cylinder. Therefore, it would be preferable to approximate screws as smooth cylinders with threads implicitly represented through interfacial surface interactions, i.e. pseudo-threaded screws, as opposed to a perfectly bonded connection or true screw geometry with friction. This technique was successfully implemented in a recent study by 231

232 Inzana et al. [208]. They first prepared three FE models of the single screw-in-bone system with threads modelled as either a smooth cylinder with tie constraints (similar to that used in chapter 5), a smooth cylinder with pseudo-threading or an explicitly modelled thread geometry with frictional contact. During the pull-out tests, the pseudo-threaded model was found to more accurately approximate the screw displacements, strain distribution and strain magnitudes of the explicitly threaded model than the smooth cylinder with tie constraint. Next, they applied the pseudo-threading technique to FE models of the proximal humerus that had been instrumented with a PHILOS plate and had one of the following three screw configurations: with calcar screws, without calcar screws and without both calcar screws and zone 2 screws (chapter 3, Fig. 41). Varus bending was applied by fixing the humeral head and applying an axial force to the shaft, based on the biomechanical study by Unger et al [98]. Results revealed that similar trends in the performance of three screw configurations could be achieved, irrespective of the type of screw modelling: pseudo-threading or fully tying. The pseudo-threaded models, however, did have consistently higher strains and humeral head rotations than the tied models. One noteworthy limitation of their proximal humerus study was that they did not develop a model with explicit threading to allow us to compare these strain and rotation values with it. Nevertheless, with the success of pseudo-threading in more accurately replicating the strains and displacements of the explicitly threaded model in the single screw-in-bone system than that by tying, it was selected for this study. To achieve this, the pseudo-threading and the tying of the single screw-in-bone system FE models of the Inzana et al. study were replicated and qualitatively validated against their results. A screw was drawn in the CAD software Solid Edge (Siemens PLM Software TX, USA) to represent the geometry of the 3.5 mm locking screw. The screw was drawn as a smooth cylinder with a screw head diameter of 3.5 mm and a core diameter of 2.96 mm. Two helices with a pitch of 0.85 mm were drawn on the surface of the smooth cylinder, with one helix offset 0.35 mm from the other. The ends of the helices were connected to create a helical surface to represent threads (Fig. 91A). The screw was inserted to a depth of 12 mm through the centre axis of a 20-mm diameter hole located in a 15-mm long cylinder bone (Fig. 91B). Both parts (screw and bone) were imported into Abaqus CAE Standard 6.13 (Dassault Systemes Simulia Corp, Providence, RI, USA). There, the screw was modelled as a homogenous material with Young s modulus (105 GPa) and Poisson s ratio (0.3) of a titanium alloy and the bone as an homogenous material with Young s modulus of 600 MPa and Poisson s ratio of

233 Uncoupled cohesive behaviour (K nn = 0, K ss = 10,000 N, K tt = 0) was applied to the pseudothread area on the screw surface. Friction using the penalty method (Friction coefficient = 0.3, maximum elastic slip = 0.005) was applied to the area between the pseudo-threads on the screw surface. A reference point 1.5 mm above the centre of the top surface of the screw was created. The reference point was constrained to the top surface of the screw using a kinematic coupling. A 50 N concentrated force was applied to the reference point. The angle of force was varied between 0 (axial pulling) and 180 in 45 increments (Fig. 91B). Pinned boundary conditions were applied to the sides and bottom surface of the cylinder bone. Both the cylinder bone and screw were meshed using quadrilateral tetrahedral elements (C3D10). The total node count was 50526, similar to that achieved by Inzana et al. (52217). Figure 91. A pseudo-threaded screw (A) inserted into a cylinder bone with load applied in five directions (B) The absolute maximum principal strain was limited between 2e -3 and -2e -3 for loading directions 0 and 180 and between 5e -3 and -5e -3 for loading directions 45, 90, 135 and 180. The absolute principal strain distributions were qualitatively assessed against distributions from Innzana et al. study (Fig. 92). Qualitatively, the distributions from this study showed strains of similar magnitudes and locations on the bone, this was especially true of off-axis loading directions (i.e. 45, 90, 135, 180 ). In this study the number of elements with greater strains, for all loading directions, were qualitatively less than that found in Inzana et al. This discrepancy can be attributed to the differences in mesh sizes and the method for creating screw geometry. Despite this, the distributions were similar enough for confidence in this study s replication of the pseudo-threading method and to proceed with the application of the pseudo-threading method for a bone-plate construct. 233

234 Figure 92. Absolute principal strain distributions for qualitative comparison between this study and the Inzana et al. study in five loading directions 234

235 7.3 Validation of Pseudo-threading in Bone-Plate Finite Element Model With the successful modelling of pseudo-threading in homogenous cylinder bone, the next step was to model it in an FE model of a proximal humerus plate. The model developed in chapter 5, which had tied bone-screw interface and was a representation of the S0 configuration group of the S3 plate, was used as a template. Tie constraints were removed from all head screws and pseudo-threading was applied on extreme half of the shaft surfaces of screws 1-3 to simulate the threaded-pegs of the S0 configuration group (Fig. 93). It was necessary to validate this new pseudo-threaded model with the in vitro mechanical test data for the S0 configuration group. Figure 93. S0 screw configuration of the S3 proximal humerus plate in reality (A) and in the FE model (B) and the numbering of the plate s head screw holes (C) The solid part of the S3 plate from this model was imported into Solid Edge to create pseudothreading surfaces. Two helices with a pitch of 1.0 mm were drawn on the surface of the smooth cylinder, with one helix offset 0.46 mm from the other. The ends of the helices were 235

236 connected to create a helical surface of no thickness to represent threads. To imitate the screw configuration used in the experiments, 50% of the shaft length of screws 1-3 was pseudothreaded like the threaded pegs used, while screws 4-6 were smooth like the smooth pegs used. Although non-locking threaded screws were used for shaft screws in experiments, for modelling purposes, the screws were approximated as smooth (Fig. 90B). As for the model developed in chapter 5, screw lengths were same as those for the in vitro experiments. A parametric study was performed to determine the bone s more accurate Young s modulus ( MPa) in the pseudo-threading model. As with the single screw model, cohesive behaviour with the same properties was applied to the threaded area and friction with the same properties was applied to the area between threads. Screws 4-6 were smooth pegs, so represented with only friction applied to the surface of the screw. Tied constraints were applied to the shaft screws, which were fully threaded screws in experiments. The settings for the other pre-processing steps (boundary conditions, loading conditions, step definitions and meshing) were identical to that described for the development of the tied model in chapter 5. After meshing, the job was created and submitted, utilising 5 processing cores on a 20 GB RAM workstation. The solving time of this model was just over 52 hours, considerably longer than those developed in last two chapters (approximately 3 hours). Once the job was successfully completed, the resulting.odb file was opened. Simulation results showed that the load required to obtain 5 mm varus displacement (F 5) in the pseudo-threaded model was similar to that recorded in the in vitro experiments, with only a 0.178% difference between the two. Next, the load and displacement values at all of the seven increments of the.odb files were noted. From the load-displacement data for all the S0 configuration group's elastic varus bending trials, the experimental loads at these increment displacements were interpolated. Plotting the mean and standard deviation of these experimental loads together with the simulation loads revealed that the two follow a close trend (Fig. 94). Thus, the use of this model for this investigation was validated. 236

237 Figure 94. Load at each step increment solved by the FE model compared with the mean in vitro biomechanical test loads 7.4 Final Study Method After its validation, the model could then be used for the main investigation. Here, a total of 26 FE models were developed. The first model had six head screws of the S3 plate smooth (unthreaded). A second model, with 100% threading in all six head screws was developed. Results of these two models were to be compared, in order to achieve the first aim of this study. To achieve the second aim, a further twenty-four models were developed which had either 25%, 50%, 75% or 100% threading on one of the six screw heads screws of the S3 plate. 50% threading represented the threaded peg option currently provided with the S3 plate but 25% and 75% intermediates are currently not commercially available. 237

238 As previously described, cohesive behaviour with the same properties was applied to the threaded area and friction with the same properties was applied to the area between threads. Friction was applied to smooth screws and tied constraints were applied on screws. The preparation process was identical to that described in section 7.3. As before, jobs were created and submitted to a 20 GB RAM workstation with 5 processing cores. 7.5 Results Simulation results for the first two models showed that as opposed to the smooth plate, the load (F 5) solved for the fully threaded plate was over 18% greater ( N vs N). For the other twenty-four models, there was a positive correlation between the load (F 5) and the percentage threading. The presence of any amount of threading on any screw increased the reaction force and therefore stiffness, compared to a smooth plate (Fig. 95 and Table 14). This increase in reaction force and stiffness was most significant when either screw 4 or 5 is threaded and to a lesser extent, when screw 6 was threaded. The greatest increase in stiffness occurred when screw 4 was fully threaded, where reaction force was 4.55% greater than the reaction force for the smooth plate (Fig. 96 and Table 15). Screws 1-3 experienced much smaller improvement in stiffness, with the best screw (screw 2) being only 0.96% higher than that for the smooth plate. For screws 2-6, the increment from 75% to 100% threading did not increase the reaction force as significantly as the increase from 0% to 100%. Similar trends were observed, even after normalising the screws using their screw length (Fig. 97 and Table 16). Over 90% of the maximum stiffness for screws 4-6 could by achieved with 50% threading while for screw 3, 75% threading was required to achieve this (Table 17). On the contrary, screws 1 and 2 had to be fully threaded to achieve over 90% of their maximum stiffness. 238

239 Figure 95. Load required to apply 5 mm varus displacement (F5) for 25%, 50%, 75% and 100% threading of all six screws in comparison with the model with all screws smooth (dashed line) 239

240 Figure 96. Percentage change in the load required to apply 5 mm varus displacement (F5) for 25%, 50%, 75% and 100% threading of all six screws in comparison with the model with all screws smooth 240

241 Figure 97. Percentage change in the load required to apply 5 mm varus displacement (F5) for 25%, 50%, 75% and 100% threading of all six screws in comparison with the model with all screws smooth and normalised using screws' lengths 241

242 Table 14. Load required to apply 5 mm varus displacement (F5) for 0%, 25%, 50%, 75% and 100% threading of all six screws Threading Screw 1 Screw 2 Screw 3 Screw 4 Screw 5 Screw 6 0% % % % % Table 15. Percentage change in the load required to apply 5 mm varus displacement (F5) for 25%, 50%, 75% and 100% threading of all six screws in comparison with the model with all screws smooth Threading Screw 1 Screw 2 Screw 3 Screw 4 Screw 5 Screw 6 25% % % % Table 16. Percentage change in the load required to apply 5 mm varus displacement (F5) for 25%, 50%, 75% and 100% threading of all six screws in comparison with the model with all screws smooth and normalised for screws' lengths Threading Screw 1 Screw 2 Screw 3 Screw 4 Screw 5 Screw 6 25% % % %

243 Table 17. Cumulated percentage change in the load required to apply 5 mm varus displacement (F5) for each 25% increment in threading of all six screws Screw 1 Screw 2 Screw 3 Screw 4 Screw 5 Screw 6 0% % % % %

244 7.6 Discussion As for the first aim of this study, fully threading the pegs led to an increase in varus bending stiffness for all screws. In particular, when all screws where fully threaded simultaneously, an 18.4% increase in F 5 was reported. It should be noted that, in terms of absolute values, this N increase is close to the N increase in mean varus bending stiffness that was observed with the addition of the S1 screw during the in vitro biomechanical tests. Overall, this highlights the importance of threading for construct stiffness. From the literature review, two in vitro biomechanical studies emerged that compare the pegs and threaded screws for proximal humerus plates. Schumer et al. [53] tested fresh-frozen proximal humerus fractures with locking smooth pegs and threaded pegs in the S3 plate under axial compression and torsion loading; the varus bending test was not performed in their study. No statistically significant difference was reported between the smooth- and the threadedpegs. Yamamoto et al. [99] studied fully threaded screws and smooth pegs and found that in cycling varus bending, constructs threaded with smooth pegs were stiffer, being the only in vitro biomechanical to report superiority of pegs over threaded screws. In cyclic torsion, however, their difference did not reach statistical significance. Interpretation of these results is difficult due to the nature of their experiment design. First, they tested two different plates: locking compression plate with fully threaded screws and the S3 plate with smooth pegs. The smooth pegs were also longer as per manufacturer's recommendation, making it difficult to distinguish whether the higher stiffness of the constructs treated with smooth pegs was indeed due to their surface profile or one of these factors. Due to the lack of information on this subject of pegs versus threaded screws, we have to resort to the study of the literature of other joints, particularly the distal radius fractures where this debate has been studied more extensively. There, in vitro biomechanical studies have demonstrated statistically significant superior biomechanical performance with the use of threaded screws over smooth pegs, under axial compression [237,243] and torsion [243,244]. In contrast, Yao et al. [245] report no significant difference in stiffness and failure load between constructs consisting of all threaded locking pegs or all smooth locking pegs during in vitro axial compression tests. Weninger et al. [244] also report no difference in axial stiffness between smooth and threaded pegs. A clinical study by Boretto et al. [246] found no difference in stability between screws and smooth pegs for distal radius fixation. With the exception of Yamamoto et al., none of these studies in vitro studies achieved stiffer constructs with the use of smooth pegs in place of threaded screws. Despite this, since the 244

245 study by Yamamoto et al. was the only one out of these to perform varus bending testing, further, more systematic in vitro biomechanical studies are required to support the findings of this study. The second aim of this study was to investigate the effects of different percentages of threading. In general, the analysis results showed that for all head screws, stiffness increases with more threading. However, threading did not have an equal mechanical effect on all screws; full threading of inferomedial smooth pegs (screw 4 and 5) lead to up to a 4.55% increase in stiffness, nine times higher than that reported for screw 2 (0.49%). Since the trends in effects of threading remained unaffected, even after normalising for screw length, the influence of different screw lengths on stiffness is judge unlikely. Results revealed a disparity in the effect of the threading as it had a larger effect on the lower three screws (4-6) than the upper three. In fact, the percentage increase in stiffness due the full treading of screws 4 (4.55%), 5 (4.05%) or 6 (2.34%) was more than the combined total of the increase with screws 1-3 (2.18%). This cannot be solely attributable to their proximity to the fracture site. Had that been the case, threading of screw 6 would have had more significant impact on stiffness than inferomedial screws. One plausible explanation is that it is due to the divergent orientation of the inferomedial screws which are aimed towards the calcar region of the humerus, as discussed in chapter 4 (4.5.2). There, we found that the removal of zone 2 (screws 4 and 5) had more effect on the varus bending stiffness of the construct than zone 3 (screws 2 and 3). However, since the in vitro testing was conducted zone-wise, the mechanical contribution of each screw in these two zones (2 and 3) is unknown. Unlike the simulation results, a comparison between the experimental stiffness of zone 1 screw and one of the zone 2 screws could not be made. This can be solved by performing the simulation zone-wise instead of screw-wise in future. It was found that for these lower screws, 90-96% of their maximum improvement in stiffness by threading can be had just by using a 50% threaded peg instead of the smooth peg. Thus, as far as the construct stiffness is concerned, there is not a significant improvement by threading the other half of these screws. In fact, it might be possible to increase the stiffness further by using a bigger diameter in that smooth half instead, as is the case with the threaded pegs provided with the standard S3 plate. In screws 2 and 3, 52.80% and 72.72% of the maximum improvement in stiffness was from the threading of half of their shaft length. If the option of 75% threading is available, that is sufficient for screw 3 to achieve 99.2% of the maximum stiffness since the threading of the 245

246 remaining 25% offered no significant benefit. In this respect, screw 1 stands out from the others since it was the only screw where the largest increase in stiffness was from 75% to 100%. Since only 44.04% of the maximum stiffness is attained by 75% threading, it is advisable to use a fully threaded screw. Thus, these upper three screws ought to be fully threaded. This FE based study allowed us to successfully isolate the effects of threading and can act as a starting point for future in silico investigations. Here, several of its limitations must be considered to help understand the implications on clinical applicability of its findings and the potential for future work. First, the effect of screw diameter, especially that of the unthreaded section, demands further investigation. We kept it constant to ensure that the investigation was as controlled as possible. However, the trends obtained may have been different if this diameter was increased, as it is in reality. Furthermore, lengths of all screws were kept constant irrespective of the percentage threading whereas, in the clinic, smooth pegs can be longer than threaded screws. Testing a model with a more accurate choice of screw diameters and lengths may have better represented the mechanical advantage of the unthreaded region for the construct stiffness. The initial increment of 25% threading was found to be crucial in providing the stiffness of five of the six screws investigated in this study. It is noteworthy that this 25% threading is of the tip section of the screw and it restricts this extreme end of the screw while the other end, the screw head, is held by the locking mechanism in the plate hole. We believe that if the cortical and cancellous structure is modelled with a more accurate bone density distribution, the effects of the 25% threading would be accentuated, given that it makes contact with the high density cancellous bone of the subchondral region [233,237]. As for the practicality of developing such a model, the pseudo-threading technique could be applied to bones of a range of material properties. In their study, Inzana et al. found that the pseudo-threading technique was insensitive to bone s Young's modulus. In such a model, the importance of other regions of the humerus, such as the calcar region would also be highlighted and thus the trends in the threading would be more pronounced, especially in the osteoporotic bone. In fact, if the limitation of automation is overcome, the computational framework developed in last two chapters could be used to develop such models on patient-specific data to help the clinicians with the selection of mechanically optimum threading choice. This study tested for the mechanical stability of the construct under varus stability since varus collapse is one of the main complications associated with proximal humerus plates [120]. In 246

247 order to determine the risk of screw penetration more accurately, the stresses and the strains in bone, especially that surrounding the screws, need to be investigated. As noted in chapter 4, there was a disparity between the elastic and plastic results varus bending results. Thus, cyclic plastic loading will be beneficial to determine not only if there is a potential risk of penetration but also if the trends in the effects of threading hold true in the long-term. In this study, the bone-plate construct was tested in a pre-collapse scenario, one where the implant was in good contact with the fracture fragments. Rightly so, here the role of the screws was to provide the mechanical support necessary to maintain the head structure. Further FE studies could be performed out to find out how well they perform this role under a variety of other, physiologically more accurate loading types such as those introduced in the literature review (chapter 2). However, given the high prevalence of osteoporosis in proximal humerus fractures, often times it is the bone, not the implant, that leads to head collapse, Factors such as poor bone quality, avascular necrosis or varus malunion are reported to end in head collapse [44,223]. Osteonecrosis, for example, can take years to manifest, leading to decreased bone quality, followed by head collapse and screw penetration [44,247,248]. To prepare for such cases, the screws have the secondary role of minimising the damage to the articular surface. However, it is found that the two roles have conflicting aims. For example, Rooyen et al. report the case of an elderly patient who had suffered from varus collapse after treatment with a locking plate, leading to a secondary screw perforation of all but one head locking screw. This one screw did not perforate the glenohumeral joint since it had backed out of the plate, possibly due to inadequate tightening during surgery (Fig. 98). This poses the question of whether the bone-plate constructs must be stiff under all circumstances. In case of Rooyen et al., it is clear that the much-advocated ability of the locking screws to maintain a rigid fixation can be counterproductive during the humeral head collapse. This problem is also reported in the plate-based treatment of other fractures. For example, the use of less-rigid, flexible locking plate is recommended where some interfragmentary motion is necessary for the simulation of secondary bone healing by callus formation [249]. 247

248 Figure 98. Humeral head collapse and secondary perforation of locked screws into the glenohumeral joint with one screw backing out [227] Far-cortical screws and dynamic locking screws have been developed both with the aim to keep the stiffness of locking plate construct low while retaining its strength. Far-cortical screws have additional threading regions to purchase the far cortex, akin to the partially threaded screws tested in this study [250]. The screw shaft, however, has a smooth surface profile and smaller diameter than the threaded region (Fig. 99A). It acts as a cantilever beam with a small diameter, reducing the stiffness of the screw and providing the axial flexibility. An in vitro biomechanical study on diaphyseal femur fractures by Bottlang et al. [251] reports a reduction of axial compression, torsion and bending stiffness by 88%, 58% and 29%, respectively, with the use of far cortical locking screws instead of locking screws. Further, 9-54% increase in torsional strength was reported and 20-21% increase in bending strength with their use. The axial compressive strength, however, was 7-16% lower. Dynamic locking screw has a pin-sleeve design where a fully threaded shaft engages with the bone but it is not directly locked to the plate. Instead, it contains an inner pin which has threads that to lock to the plate (Fig. 99B). It is aimed that this design allows micro-motion of the pin within the outer sleeve and leads to a more homogenous distribution of loads and stress around the screw. While the benefits of dynamic locking screws have been shown in several fractures [252,253], their success in proximal humerus fracture is promising. A clinical study by Freude et al. [204] reports a reduction in screw perforation in proximal humeral fractures with the use of dynamic locking screws in the PHILOS plate. 248

249 Figure 99. Far cortical locking screw (A) and the exterior and cross-sectional views of dynamic locking screws (B) [251,261] With an appreciation of the complexity of screw mechanical roles in the prevention of perforation, it can be said that further studies, with not only multiple design parameters but also multiple objectives, need to be undertaken. For example, as for the parameters, the effects of increasing threading and diameter are likely to conflict. Similarly, the objective of increasing construct stiffness pursued in this study may come at the cost of stress concentrations in the bone and insufficient flexibility after the collapse. Such studies, if conducted in vivo or even in vitro are likely to be very time-consuming and resource-intensive. As shown in this study, FE studies can greatly minimise many such limitations, albeit at the cost of simplifications. 7.7 Conclusion Threading is an effective way of increasing the varus bending stiffness of proximal humerus plates construct. However, threading did not affect all head screws of the S3 plate equally. The inferior three screws of the plate exhibited a larger increase in stiffness. For these, most of the mechanical benefits of threading can be achieved by 50% threading. For the superior three screws, 100% threading is advised. Due to the simplifications of the FE model and the study design, the effects of several design parameters (e.g. screw diameter and length) and objectives (e.g. minimising bone stress) must be accounted for in order to devise a superior mode of treatment for screw perforation. 249

250 Chapter 8: Research Overview 8.1 Thesis Overview The overall aim of this project was to create a computer-aided design framework for proximal humerus plates using a validated subject-specific humerus-plate FE model. To achieve this, mechanical testing of proximally fractured humeri which had been tested with one of the three commercially available plates (S3-, Fx- and the PHILOS plate) was first conducted, as described in chapters 3 and 4. Based on the knowledge of the literature and practical experience, a clinically-relevant experimental protocol was finalised wherein the bone-plate constructs elastic (varus, valgus, extension and flexion) and plastic (varus) bending performances were determined. The mechanical effects of the removal of different screw zones on the bone-plate constructs stability were also investigated. It was found that the treatment of the fractures with the S3 plate yielded the stiffest constructs, possibly due to its thicker cross-section and the 135 o inclination of its screws with respect to the humeral shaft. Removal of the inferomedial support had the most impact on the varus bending stiffness, as supported by the literature. However, varus stability was also found to depend on the type of medial support. For example, the medial support provided by inferomedial screws in the Fx plate achieved significantly superior extension, flexion, valgus and varus bending than that by the insertion of a blade. Removal of the large 6.5 mm screw in the Fx plate had a relatively low effect on the mean stiffness in all four directions than that of the blade or the inferomedial screws. With a few exceptions, varus and valgus stability of the bone-plate construct depended more on the zones near the fracture site. The screws placed farther away from the fracture site were found to be important in extension and flexion bending stiffness, possibly due to their nonparallel orientation which increased the zone's second moment of area. Considering the complexity of their design, plates' performance could be affected by these factors: (A) plate position and orientation, (B) plate dimensions, (C) plate surface profile, (D) number and position of screw holes, (E) screw orientation, (F) surface profile, (G) dimensions and (H) use of a blade in place of the screw. The effect of these factors on mechanical performance was discussed individually in chapter 4. In chapter 5, the computational framework was described in detail, particularly the three stages: reverse engineering of bone and plate geometry, creation of initial FE model and the FE model automation and optimisation. As an implementation of the framework, a subject-specific FE 250

251 model of the humerus-plate construct was successfully developed that simulated the in vitro varus bending tests. The maximum bending force (F 5) recorded in the FE model ( N) validated its use for further in silico testing. The model was then used to perform an optimisation study in chapter 6 where a combination of height and divergence angle of S3 plate s inferomedial screws was determined that yielded optimum bone-plate construct stability (minimum fracture gap change). The optimum design (16 o divergence angle and 33 o height angle) yielded the lowest fracture gap change (0.156 mm) which was 4.686% lower than the standard FE model while achieving 5.707% higher varus bending load ( N). By successfully implementing all the stages of the described computational framework, from medical scan data acquisition to the final optimised design, it was hoped that this framework could be used in future to perform multi-objective optimisation studies of multiple design parameters and to design implants for other parts of the human body. In chapter 7, the use of a validated FE model was extended as it was applied to perform an investigation into the controversial issue of using smooth pegs over threaded screws in the S3 plate. A pseudo-threading technique, previously reported in the literature, was first adapted into our FE model, to more accurately model the bone-screw interface. The model was then re-validated and used to test the effects of different percentages of screw threading on the bone-plate construct stiffness. Overall, it was found that threading did lead to higher varus bending stiffness, as an increase of up to 4.55% was achieved by threading a smooth peg. However, it was also found that threading did not affect all screws equally. For example, the three inferiorly position screws exhibited a larger increase in stiffness than superiorly placed screws, despite increasing the threading by the same percentage. In conclusion, the works conducted in chapter 5 demonstrated that the FE model can be used not only for optimisation studies but also to investigate clinically relevant questions. 251

252 8.2 Research Output Arising from this Work 1. Jabran A, Peach C, Zou Z, Ren L. Biomechanical Assessment of Proximal Humerus Plates using an Integrated Experimental and Computational Framework. Postgraduate Summer Research Showcase (PSRS) [Poster Presentation] 2. Jabran A, Peach C, Zou Z, Ren L. Biomechanical Assessment of Proximal Humerus Plates using an Integrated Experimental and Computational Framework. MACE PGR Conference DOI: [Conference Presentation and Article]- Best oral presentation award 3. Jabran A, Ren L, Peach C, Zou Z. A Methodology for Biomechanical Assessment of Proximal Humerus Fractures Using an Integrated Experimental and Computational Framework. Procedia CIRP. 2016;49: [Conference Presentation and Article] 4. Jabran A, Peach C, Zou Z, Ren L. Hybrid Blade and Locking Plate Fixation for Proximal Humerus Fractures: A Comparative Biomechanical Analysis. British Elbow and Shoulder Society (BESS) Annual meeting [Conference Presentation] 5. Jabran A, Peach C, Ren L. Biomechanical analysis of plate systems for proximal humerus fractures: a systematic literature review. Peer J [Submitted, under review] 6. Jabran A, Peach C, Zou Z, Ren L. Hybrid Blade and Locking Plate Fixation for Proximal Humerus Fractures: A Comparative Biomechanical Analysis. Bone and Joint Research [Submitted, under review] 7. Jabran A, Peach C, Zou Z, Ren L. Biomechanical Comparison of Screw-based Zoning of PHILOS and Fx Proximal Humerus Plates. Clinical Biomechanics [To be submitted] 8. Jabran A, Peach C, Zou Z, Ren L. Biomechanical Comparison of Screw-based Zones of a Spatial Subchondral Support Plate for Proximal Humerus Fractures. Annals of Biomedical Engineering [To be submitted] 9. Jabran A, Peach C, Zou Z, Ren L. Finite Element Based Optimisation of Proximal Humeral Plates. Annals of Biomedical Engineering [To be submitted] 252

253 8.3 Future Works Successful implementation of the computational framework for an optimisation study (chapter 6) and a clinical investigation (chapter 7) opens the door for a variety of future works. For example, the optimisation study was aimed to yield minimum fracture gap change. Indirectly, this meant increasing construct stiffness. A study in the literature shows that increasing stiffness is not always beneficial. A biomechanical study by Lill et al. [104] revealed that when tested under cyclic conditions, rigid implants with high initial stiffness exhibit early loosening and failure of the interface between the bone and the implant. Contrarily, the more elastic implants that possessed low stiffness showed a lower decrease in the load and low load level which is promising for long-term stability and makes them more suitable for osteoporotic bones. There exists a window of stiffness values in which ideal mechanical stability and bone healing are obtained. While determining this window can be challenging, using the stress and strain values in conjunction with stiffness can allow us to better define the aim of the optimisation study. Values and distribution of stress were compared in our optimisation study but their use was limited because the model had been only qualitatively validated for stress values. As reported by Sun et al., the elastic moduli of cellular foams are strain-dependent [207]. Further in vitro biomechanical tests should be performed, based on their methodology, to more accurately determine elastic moduli of the polyurethane foams used in the synthetic humeri. In vitro bending tests involving stress or strain measurements are also required for quantitative verification and validation of the stress and strains in our FE model. Doing so would lead to a more robust FE model especially if the bone's viscoelasticity and the local variations in the density and material properties are also modelled. Once re-validated, the model could be used to perform multi-objective studies, where, for example, the aim would be to not only maximise the construct stiffness but also to minimise the stress distribution in the bone fragments. One of the limitations of the FE model was that it only involved varus bending tests. This was primarily due to the clinical importance of this loading condition as well as the time restrictions. Thus, the optimum design can, in fact, be sub-optimal in real-life loading conditions. Further optimisation studies are required that also test other bending directions (extension, flexion and valgus). Since the in vitro test data for these have already been obtained, the FE model would need to be validated for these directions first. Further in vitro tests will need to be performed, involving other loadings such as axial loading and rotation test to be used for FE model validation. For the more complex 253

254 loadings, such as those discussed in the literature review, the surrounding muscles and tendons may also need to be modelled. In our studies, we only tested simple two-part surgical neck fractures while modelling the bone as a homogenous material. The optimum treatment of more complex fracture cases, especially in osteoporotic bone, remains a controversial topic in the literature. It should be noted that the optimum plate design determined from our current FE model is likely to be sub-optimal for other fracture cases and bone types and thus it would have to be updated before performing further in silico testing. For example, bone-mineral density data can be used to model the bone with a more accurate distribution of bone quality. As a result, certain screws, such as those passing through the inferomedial region of the humerus, may be found to be even more important for certain types of loadings. Also, in fractures of the greater tuberosity fractures, some of the screws can end up passing through multiple fracture fragments and thus their mechanical role will be different to that of the screws in surgical neck fractures simulated in our studies. This will also impose further restrictions on their permissible range of height and divergence angles as they would be expected to purchase with multiple fragments simultaneously. Testing for complex fractures will also necessitate the devising of an alternative method of zoning the screws on the plates. The proximity-based zoning adopted in our optimisation study was easy to implement since there was a single fracture plane distance to be used as a reference. However, this would be particularly difficult to achieve if the fractures have more than two fragments and those too along multiple planes. For FE models with multiple fractures, fracture gap change parameter, which was being minimised in our optimisation study, may also be insufficient to describe the construct stability. Instead, we will need to optimise the inter-fragmentary motion of all different fragments, akin to the measurements made by the 3D motion capture systems used in several in vitro biomechanical studies. Similar to other joints of the body, advancements in the development of plate fixation of the proximal humerus owes much to the study of the implant design whether it is in regard to the design of the plates or the screws. After all, it was the breakthrough in research of plate-screw interface design that led to the development of the locking plate and brought plate fixation back into the limelight. It is, therefore, crucial to not only investigate the current design solutions but also propose new ones. In this regard, conducting the in vitro biomechanical tests allowed us to determine a range of design parameters that can affect the mechanical performance of the plate. It is understood that the removal of a screw zone meant changing several design parameters at once. While this made it complex to analyse the obtained results, 254

255 it also represented screw configurations that truly were as used in the clinic, thus ensuring that our tests were clinically relevant. Subsequent optimisation study focused only on one of the many identified design parameters: screw orientation. Further, the screw orientation of only the two screws (screw 4 and 5), those which were found to be more important for varus stability, was changed. A more comprehensive study, on not only the screw orientation of other screws but also on other design parameters, is required to further optimise the plate design. If such a study was to be performed using the in vitro biomechanical testing approach, it would have several drawbacks. For example, the issue of inter-specimen variability would still exist especially in the testing of cadavers. This is in addition to the limits on factors such as cost, time and resource. On the contrary, the in silico testing performed on this framework would allow the performance of destructive testing of the same specimen as many times as required, which could not be achieved in vitro. However, with the success of the computational framework demonstrated in the current study, such a comprehensive optimisation study may be more achievable using the FE analysis. Once an optimum design has been determined and finalised for a given patient case using the computational framework, the next stage involves the manufacturing of the implant. This manufacturing of patient-specific medical implants with complex geometry and features could be achieved with technologies such as additive manufacturing. Next, further in vitro mechanical testing will be used to compare the performance of this manufactured plate with that of the commercially available plates, before its clinical use. 255

256 References 1. Baron JA, Barret JA, Karagas MR. The Epidemiology of Peripheral Fractures. Bone. 1996;18(3 (suppl)):209s 213S. 2. Kettler M, Biberthaler P, Braunstein V, Zeiler C, Kroetz M, Mutschler W. [Treatment of proximal humeral fractures with the PHILOS angular stable plate. Presentation of 225 cases of dislocated fractures]. Unfallchirurg [Internet]. 2006;109(12): Available from: 3. Hagino H, Yamamoto K, Ohshiro H, Nakamura T, Kishimoto H, Nose T. Changing incidence of hip, distal radius, and proximal humerus fractures in Tottori Prefecture, Japan. Bone. 1999;24(3): Kannus P, Palvanen M, Niemi S, Parkkari J, Jarvinen M, Vuori I. Osteoporotic fractures of the proximal humerus in elderly Finnish persons - Sharp increase in and alarming projections for the new millennium. Acta Orthop Scand. 2000;71(5): Kannus P, Palvanen M, Niemi S, Parkkari J, Jarvinen M, Vuori I. Increasing number and incidence of osteoporotic fractures of the proximal humerus in elderly people. BMJ [Internet]. 1996/10/ ;313(7064): Available from: 6. Hutchinson MR. The Burden of Musculoskeletal Diseases in the United States:Prevalance, Societal and Economic Cost. J Am Coll Surg. 2009;208(1):e Court-Brown CM, Garg A, McQueen MM. The translated two-part fracture of the proximal humerus - Epidemiology and outcome in the older patient. J Bone Jt Surgery- British Vol. 2001;83B(6): Codman EA. The Shoulder: Rupture of the Supraspinatus Tendon and Other Lesions in Or about the Subacromial Bursa. Krieger Publications; p. 9. Neer CS. Displaced Proximal Humeral Fractures.1. Classification and Evaluation. J Bone Joint Surg Am. 1970;A 52(6): Sallay PI, Pedowitz RA, Mallon WJ, Vandemark RM, Dalton JD, Speer KP. Reliability and reproducibility of radiographic interpretation of proximal humeral fracture pathoanatomy. J Shoulder Elb Surg. 1997;6(1): Sidor ML, Zuckerman JD, Lyon T, Koval K, Cuomo F, Schoenberg N. The Neer- Classification-System For Proximal Humeral Fractures - An Assessment Of Interobserver Reliability And Intraobserver Reproducibility. J Bone Joint Surg Am. 1993;75A(12): Kristiansen B, Andersen ULS, Olsen CA, Varmarken JE. The Neer Classification Of Fractures Of The Proximal Humerus - An Assessment Of Interobserver Variation. Skeletal Radiol. 1988;17(6):

257 13. Siebenrock KA, Gerber C. The Reproducibility Of Classification Of Fractures Of The Proximal End Of The Humerus. J Bone Joint Surg Am. 1993;75A(12): Neer 2nd CS. Displaced proximal humeral fractures. I. Classification and evaluation. J Bone Joint Surg Am [Internet]. 1970;52(6): Available from: Bernstein J, Adler LM, Blank JE, Dalsey RM, Williams GR, Iannotti JP. Evaluation of the Neer system of classification of proximal humeral fractures with computerized tomographic scans and plain radiographs. J Bone Joint Surg Am. 1996;78A(9): Helmy N, Hintermann B. New trends in the treatment of proximal humerus fractures. Clin Orthop Relat Res [Internet]. 2006;442(442): Available from: Court-Brown CM, McQueen MM. The impacted varus (A2.2) proximal humeral fracture - Prediction of outcome and results of nonoperative treatment in 99 patients. Acta Orthop Scand. 2004;75(6): Lill H, Josten C. Conservative or operative treatment of numeral head fractures in the elderly? Chirurg. 2001;72(11): Koval KJ, Gallagher MA, Marsicano JG, Cuomo F, McShinawy A, Zuckerman JD. Functional outcome after minimally displaced fractures of the proximal part of the humerus. J Bone Joint Surg Am. 1997;79A(2): Mclaughlin HL. Dislocation of the Shoulder with Tuberosity Fracture. Surg Clin North Am. 1963;43(6): Leyshon RL. Closed Treatment of Fractures of the Proximal Humerus. Acta Orthop Scand. 1984;55(1): Handoll HHG, Ollivere BJ, Rollins KE. Interventions for treating proximal humeral fractures in adults. Cochrane Database Syst Rev. 2012;(12). 23. Rangan A, Handoll H, Brealey S, Jefferson L, Keding A, Martin BC, et al. Surgical vs Nonsurgical Treatment of Adults With Displaced Fractures of the Proximal Humerus The PROFHER Randomized Clinical Trial. Jama-Journal Am Med Assoc. 2015;313(10): Bosworth DM. Blade Plate Fixation - Technic Suitable for Fractures of the Surgical Neck of the Humerus and Similar Lesions. Jama-Journal Am Med Assoc. 1949;141(16): Court-Brown CM, Cattermole H, McQueen MM. Impacted valgus fractures (B1.1) of the proximal humerus - The results of non-operative treatment. J Bone Jt Surgery-British Vol. 2002;84B(4): Kristiansen B, Christensen SW. Plate Fixation Of Proximal Humeral Fractures. Acta 257

258 Orthop Scand. 1986;57(4): Cornell CN, Levine D, Pagnani MJ. Internal-Fixation Of Proximal Humerus Fractures Using The Screw-Tension Band Technique. J Orthop Trauma. 1994;8(1): Nho SJ, Brophy RH, Barker JU, Cornell CN, MacGillivray JD. Innovations in the management of displaced proximal humerus fractures. J Am Acad Orthop Surg. 2007;15(1): Nho SJ, Brophy RH, Barker JU, Cornell CN, MacGillivray JD. Management of proximal humeral fractures based on current literature. J Bone Joint Surg Am. 2007;89A: Wagner M. General principles for the clinical use of the LCP. Inj J Care Inj. 2003;34: Wijgman AJ, Roolker W, Patt TW, Raaymakers ELFB, Marti RK. Open reduction and internal fixation of three and four-part fractures of the proximal part of the humerus. J Bone Joint Surg Am. 2002;84A(11): Seide K, Triebe J, Faschingbauer M, Schulz AP, Püschel K, Mehrtens G, et al. Locked vs. unlocked plate osteosynthesis of the proximal humerus - A biomechanical study. Clin Biomech. 2007;22(2): Siffri PC, Peindl RD, Coley ER, Norton J, Connor PM, Kellam JE. Biomechanical analysis of blade plate versus locking plate fixation for a proximal humerus fracture: Comparison using cadaveric and synthetic humeri. J Orthop Trauma. 2006;20(8): Weinstein DM, Bratton DR, Ciccone WJ, Elias JJ. Locking plates improve torsional resistance in the stabilization of three-part proximal humeral fractures. J Shoulder Elb Surg. 2006;15(2): Bjorkenheim JM, Pajarinen J, Savolainen V. Internal fixation of proximal humeral fractures with a locking compression plate - A retrospective evaluation of 72 patients followed for a minimum of 1 year. Acta Orthop Scand. 2004;75(6): Solberg BD, Moon CN, Franco DP, Paiement GD. Surgical Treatment of Three and Four-Part Proximal Humeral Fractures. J Bone Joint Surg Am. 2009;91A(7): Owsley KC, Gorczyca JT. Fracture displacement and screw cutout after open reduction and locked plate fixation of proximal humeral fractures [corrected]. J Bone Joint Surg Am [Internet]. 2008;90(2): Available from: Clavert P, Adam P, Bevort A, Bonnomet F, Kempf JF. Pitfalls and complications with locking plate for proximal humerus fracture. J Shoulder Elb Surg. 2010;19(4): Fankhauser F, Boldin C, Schippinger G, Haunschmid C, Szyszkowitz R. A new locking plate for unstable fractures of the proximal humerus. Clin Orthop Relat Res. 2005;(430):

259 40. Micic ID, Kim KC, Shin DJ, Shin SJ, Kim PT, Park IH, et al. Analysis of early failure of the locking compression plate in osteoporotic proximal humerus fractures. J Orthop Sci. 2009;14(5): Gardner MJ, Weil Y, Barker JU, Kelly BT, Helfet DL, Lorich DG. The importance of medial support in locked plating of proximal humerus fractures. J Orthop Trauma. 2007;21(3): Hepp P, Theopold J, Voigt C, Engel T, Josten C, Lill H. The surgical approach for locking plate osteosynthesis of displaced proximal humeral fractures influences the functional outcome. J Shoulder Elb Surg. 2008;17(1): Jung W-B, Moon E-S, Kim S-K, Kovacevic D, Kim M-S. Does medial support decrease major complications of unstable proximal humerus fractures treated with locking plate? Bmc Musculoskelet Disord. 2013; Sproul RC, Iyengar JJ, Devcic Z, Feeley BT. A systematic review of locking plate fixation of proximal humerus fractures. Inj J Care Inj. 2011;42(4): Robinson CM, Wylie JR, Ray AG, Dempster NJ, Olabi B, Seah KTM, et al. Proximal humeral fractures with a severe varus deformity treated by fixation with a locking plate. J Bone Jt Surgery-British Vol. 2010;92B(5): Gardner MJ, Boraiah S, Helfet DL, Lorich DG. Indirect medial reduction and strut support of proximal humerus fractures using an endosteal implant. J Orthop Trauma. 2008;22(3): Zhang L, Zheng JY, Wang WL, Lin GM, Huang YJ, Zheng J, et al. The clinical benefit of medial support screws in locking plating of proximal humerus fractures: a prospective randomized study. Int Orthop. 2011;35(11): Ponce BA, Thompson KJ, Raghava P. The role of medial comminution and calcar restoration in varus collapse of proximal humeral fractures treated with locking plates. J Bone Joint Surg Am. 2011;95(16):e113(1-7). 49. Erhardt JB, Stoffel K, Kampshoff J, Badur N, Yates P, Kuster MS. The Position and Number of Screws Influence Screw Perforation of the Humeral Head in Modern Locking Plates: A Cadaver Study. J Orthop Trauma. 2012;26(10):E Zettl R, Müller T, Topp T, Lewan U, Krüger A, Kühne C, et al. Monoaxial versus polyaxial locking systems: A biomechanical analysis of different locking systems for the fixation of proximal humeral fractures. In: International Orthopaedics p Hessler C, Schmucker U, Matthes G, Ekkernkamp A, Guetschow R, Eggers C. Results after treatment of instable fractures of the proximal humerus using a fixed-angle plate. Unfallchirurg. 2006;109(10): Gradl G, Stedtfeld HW, Morlock M, Sellenschloh K, Puschel K, Mittlmeier T, et al. Locking plate fixation of humeral head fractures with a telescoping screw. A 259

260 comparative biomechanical study versus a standard plate. Inj J Care Inj. 2012;43(6): Schumer RA, Muckley KL, Markert RJ, Prayson MJ, Heflin J, Konstantakos EK, et al. Biomechanical comparison of a proximal humeral locking plate using two methods of head fixation. J Shoulder Elb Surg. 2010;19(4): Gille J, Schulz a P, Queitsch C, Paech a, Jürgens C. Initial results of the s3-humerus plate. Open Orthop J [Internet]. 2008;2: Available from: rendertype=abstract 55. Stoddard JE, Pryce D, Potter D, Ali AA, Morgan A. Early experience of managing proximal humerus fractures with the S3 plate. Shoulder Elb [Internet] Oct [cited 2017 Mar 25];3(4): Available from: Charalambous CP, Siddique I, Valluripalli K, Kovacevic M, Panose P, Srinivasan M, et al. Proximal humeral internal locking system (PHILOS) for the treatment of proximal humeral fractures. Arch Orthop Trauma Surg. 2007;127(3): Südkamp N, Bayer J, Hepp P, Voigt C, Oestern H, Kääb M, et al. Open reduction and internal fixation of proximal humeral fractures with use of the locking proximal humerus plate. Results of a prospective, multicenter, observational study. J Bone Joint Surg Am. 2009;91(805): Dheerendra SK, Khan WS, Barber J, Goddard NJ, Ahrens PM. Outcomes of locking plates in proximal humeral fractures: a systematic review. Shoulder Elb [Internet] Apr [cited 2017 May 28];3(2): Available from: Kostler W, Strohm PC, Sudkamp NP. [Proximal humerus fracture in advanced age. Treatment with fixed angle plate osteosynthesis]. Chirurg. 2003;74(11): Rehman S, Damron TA, Geel C. Humeral blade plate fixation of intercalary allografts and segmentally comminuted proximal humeral fractures: A preliminary report. Injury. 2000;31(10): Sehr JR, Szabo RM. Semitubular blade plate for fixation in the proximal humerus. J Orthop Trauma. 1988;2: Baron JA, Karagas M, Barrett J, Kniffin W, Malenka D, Mayor M, et al. Basic epidemiology of fractures of the upper and lower limb among Americans over 65 years of age. Epidemiology. 1996;7(6): Harrison JWK, Howcrofr DWJ, Warner JG, Hodgson SP. Internal fixation of proximal humeral fractures. Acta Orthop Belg. 2007;73(1): Plecko M, Kraus A. Internal fixation of proximal humerus fractures using the locking 260

261 proximal humerus plate. Oper Orthop Traumatol [Internet]. 2005;17(1): Available from: Rowles DJ, McGrory JE. Percutaneous pinning of the proximal part of the humerus. An anatomic study. J Bone Joint Surg Am [Internet]. 2001/11/ ;83 A(11): Available from: Fuchtmeier B, May R, Hente R, Maghsudi M, Volk M, Hammer J, et al. Proximal humerus fractures: a comparative biomechanical analysis of intra and extramedullary implants. Arch Orthop Trauma Surg. 2007;127(6): Koval KJ, Blair B, Takei R, Kummer FJ, Zuckerman JD. Surgical neck fractures of the proximal humerus: A laboratory evaluation of ten fixation techniques. J Trauma-Injury Infect Crit Care. 1996;40(5): Kitson J, Booth G, Day R. A biomechanical comparison of locking plate and locking nail implants used for fractures of the proximal humerus. J Shoulder Elb Surg. 2007;16(3): Sanders BS, Bullington AB, McGillivary GR, Hutton WC. Biomechanical evaluation of locked plating in proximal humeral fractures. J Shoulder Elb Surg. 2007;16(2): Brunner F, Sommer C, Bahrs C, Heuwinkel R, Hafner C, Rillmann P, et al. Open reduction and internal fixation of proximal humerus fractures using a proximal humeral locked plate: a prospective multicenter analysis. J Orthop Trauma [Internet]. 2009;23(3): Available from: Egol KA, Ong CC, Walsh M, Jazrawi LM, Tejwani NC, Zuckerman JD. Early complications in proximal humerus fractures (OTA Types 11) treated with locked plates. J Orthop Trauma [Internet]. 2008;22(3): Available from: Koukakis A, Apostolou CD, Taneja T, Korres DS, Amini A. Fixation of proximal humerus fractures using the PHILOS plate: early experience. Clin Orthop Relat Res [Internet]. 2006/01/ ;442: Available from: Esser RD. Open reduction and internal fixation of three- and four-part fractures of the proximal humerus. Clin Orthop Relat Res [Internet]. 1994/02/ ;(299): Available from: Hirschmann MT, Quarz V, Audigé L, Ludin D, Messmer P, Regazzoni P, et al. Internal fixation of unstable proximal humerus fractures with an anatomically preshaped interlocking plate: a clinical and radiologic evaluation. J Trauma [Internet]. 261

262 2007;63(6): Available from: Bae JH, Oh JK, Chon CS, Oh CW, Hwang JH, Yoon YC. The biomechanical performance of locking plate fixation with intramedullary fibular strut graft augmentation in the treatment of unstable fractures of the proximal humerus. J Bone Jt Surgery-British Vol. 2011;93B(7): Instrum K, Fennell C, Shrive N, Damson E, Sonnabend D, Hollinshead R. Semitubular blade plate fixation in proximal humeral fractures: A biomechanical study in a cadaveric model. J Shoulder Elb Surg. 1998;7(5): Wallace MJ, Bledsoe G, Moed BR, Israel HA, Kaar SG. Relationship of Cortical Thickness of the Proximal Humerus and Pullout Strength of a Locked Plate and Screw Construct. J Orthop Trauma. 2012;26(4): Foruria AM, Carrascal MT, Revilla C, Munuera L, Sanchez-Sotelo J. Proximal humerus fracture rotational stability after fixation using a locking plate or a fixed-angle locked nail: The role of implant stiffness. Clin Biomech. 2010;25(4): Chow RM, Begum F, Beaupre LA, Carey JP, Adeeb S, Bouliane MJ. Proximal humeral fracture fixation: Locking plate construct +/- intramedullary fibular allograft. J Shoulder Elb Surg. 2012;21(7): Mathison C, Chaudhary R, Beaupre L, Reynolds M, Adeeb S, Bouliane M. Biomechanical analysis of proximal humeral fixation using locking plate fixation with an intramedullary fibular allograft. Clin Biomech. 2010;25(7): Weeks CA, Begum F, Beaupre LA, Carey JP, Adeeb S, Bouliane MJ. Locking plate fixation of proximal humeral fractures with impaction of the fracture site to restore medial column support: a biomechanical study. J shoulder Elb Surg / Am Shoulder Elb Surg [Internet]. 2013;22(11): Available from: Roderer G, Brianza S, Schiuma D, Schwyn R, Scola A, Gueorguiev B, et al. Mechanical Assessment of Local Bone Quality to Predict Failure of Locked Plating in a Proximal Humerus Fracture Model. Orthopedics. 2013;36(9):E Roderer G, Scola A, Schmolz W, Gebhard F, Windolf M, Hofmann-Fliri L. Biomechanical in vitro assessment of screw augmentation in locked plating of proximal humerus fractures. Inj J Care Inj. 2013;44(10): Schliemann B, Seifert R, Rosslenbroich SB, Theisen C, Wahnert D, Raschke MJ, et al. Screw augmentation reduces motion at the bone-implant interface: a biomechanical study of locking plate fixation of proximal humeral fractures. J Shoulder Elb Surg. 2015;24(12): Yoon RS, Dziadosz D, Porter DA, Frank MA, Smith WR, Liporace FA. A comprehensive update on current fixation options for two-part proximal humerus fractures A 262

263 biomechanical investigation. Inj J Care Inj. 2014;45(3): Gillespie R, Ramachandran V. Biomechanical Evaluation of 3-Part Proximal Humerus Fractures: A Cadaveric Study. Orthopedics. 2009;32(11): Hymes RA, Levine MJ, Schulman JE, Westbrook RD, Li JH. Mechanisms of Failure of Locked-Plate Fixation of the Proximal Humerus: Acoustic Emissions as a Novel Assessment Modality. J Orthop Trauma. 2013;27(7): Gradl G, Knobe M, Stoffel M, Prescher A, Dirrichs T, Pape HC. Biomechanical Evaluation of Locking Plate Fixation of Proximal Humeral Fractures Augmented With Calcium Phosphate Cement. J Orthop Trauma. 2013;27(7): Chudik SC, Weinhold P, Dahners LE. Fixed-angle plate fixation in simulated fractures of the proximal humerus: A biomechanical study of a new device. J Shoulder Elb Surg. 2003;12(6): Burke NG, Kennedy J, Cousins G, Fitzpatrick D, Mullett H. Locking plate fixation with and without inferomedial screws for proximal humeral fractures: a biomechanical study. J Orthop Surg. 2014;22(2): Dietz SO, Hartmann F, Schwarz T, Nowak TE, Enders A, Kuhn S, et al. Retrograde nailing versus locking plate osteosynthesis of proximal humeral fractures: a biomechanical study. J Shoulder Elb Surg. 2012;21(5): Hessmann MH, Hansen WSM, Krummenauer F, Pol TF, Rommens M. Locked plate fixation and intramedullary nailing for proximal humerus fractures: A biomechanical evaluation. J Trauma-Injury Infect Crit Care. 2005;58(6): Maldonado ZM, Seebeck J, Heller MOW, Brandt D, Hepp P, Lill H, et al. Straining of the intact and fractured proximal humerus under physiological-like loading. J Biomech. 2003;36(12): Edwards SL, Wilson NA, Zhang LQ, Flores S, Merk BR. Two-part surgical neck fractures of the proximal part of the humerus - A biomechanical evaluation of two fixation nchniques. J Bone Joint Surg Am. 2006;88A(10): Huff LR, Taylor PA, Jani J, Owen JR, Wayne JS, Boardman ND. Proximal humeral fracture fixation: A biomechanical comparison of two constructs. J Shoulder Elb Surg. 2013;22(1): Roderer G, Gebhard F, Krischak G, Wilke HJ, Claes L. Biomechanical in vitro assessment of fixed angle plating using a new concept of locking for the treatment of osteoporotic proximal humerus fractures. Int Orthop. 2011;35(4): Ruch DS, Glisson RR, Marr AW, Russell GB, Nunley JA. Fixation of three-part proximal humeral fractures: A biomechanical evaluation. J Orthop Trauma. 2000;14(1): Unger S, Erhart S, Kralinger F, Blauth M, Schmoelz W. The effect of in situ augmentation on implant anchorage in proximal humeral head fractures. Inj J Care Inj. 263

264 2012;43(10): Yamamoto N, Hongo M, Berglund LJ, Sperling JW, Cofield RH, An KN, et al. Biomechanical analysis of a novel locking plate with smooth pegs versus a conventional locking plate with threaded screws for proximal humerus fractures. J Shoulder Elb Surg. 2013;22(4): Kralinger F, Unger S, Wambacher M, Smekal V, Schmoelz W. The medial periosteal hinge, a key structure in fractures of the proximal humerus: a biomechanical cadaver study of its mechanical properties. J Bone Jt Surgery-British Vol. 2009;91B(7): Carrera EF, Nicolao FA, Netto NA, Carvalho RL, dos Reis FB, Giordani EJ. A mechanical comparison between conventional and modified angular plates for proximal humeral fractures. J Shoulder Elb Surg. 2008;17(4): Kwon BK, Goertzen DJ, O Brien PJ, Broekhuyse HM, Oxland TR. Biomechanical evaluation of proximal humeral fracture fixation supplemented with calcium phosphate cement. J Bone Joint Surg Am. 2002;84A(6): Lever JP, Aksenov SA, Zdero R, Ahn H, Mckee MD, Schemitsch EH. Biomechanical analysis of plate osteosynthesis systems for proximal humerus fractures. J Orthop Trauma. 2008;22(1): Lill H, Hepp P, Korner J, Kassi JP, Verheyden AP, Josten C, et al. Proximal humeral fractures: how stiff should an implant be? A comparative mechanical study with new implants in human specimens. Arch Orthop Trauma Surg. 2003;123(2 3): Duda GN, Epari DR, Babst R, Lambert SM, Matthys R, Sudkamp NP. Mechanical evaluation of a new minimally invasive device for stabilization of proximal humeral fractures in elderly patients - A cadaver study. Acta Orthop. 2007;78(3): Bai L, Fu ZG, An S, Zhang PX, Zhang DY, Jiang BG. Effect of Calcar Screw Use in Surgical Neck Fractures of the Proximal Humerus With Unstable Medial Support: A Biomechanical Study. J Orthop Trauma. 2014;28(8): Katthagen JC, Schwarze M, Meyer-Kobbe J, Voigt C, Hurschler C, Lill H. Biomechanical effects of calcar screws and bone block augmentation on medial support in locked plating of proximal humeral fractures. Clin Biomech. 2014;29(7): Katthagen JC, Schwarze M, Warnhoff M, Voigt C, Hurschler C, Lill H. Influence of plate material and screw design on stiffness and ultimate load of locked plating in osteoporotic proximal humeral fractures. Inj J Care Inj. 2016;47(3): Lescheid J, Zdero R, Shah S, Kuzyk PRT, Schemitsch EH. The Biomechanics of Locked Plating for Repairing Proximal Humerus Fractures With or Without Medial Cortical Support. J Trauma-Injury Infect Crit Care. 2010;69(5): Zhang W, Zeng LQ, Liu YJ, Pan Y, Zhang W, Zhang CQ, et al. The Mechanical Benefit of Medial Support Screws in Locking Plating of Proximal Humerus Fractures. PLoS One. 264

265 2014;9(8) Kralinger F, Gschwentner M, Wambacher M, Smekal V, Haid C. Proximal humeral fractures: what is semi-rigid? Biomechanical properties of semi-rigid implants, a biomechanical cadaver based evaluation. Arch Orthop Trauma Surg. 2008;28(2): Walsh S, Reindl R, Harvey E, Berry G, Beckman L, Steffen T. Biomechanical comparison of a unique locking plate versus a standard plate for internal fixation of proximal humerus fractures in a cadaveric model. Clin Biomech. 2006;21(10): Voigt C, Kreienborg S, Megatli O, Schulz AP, Lill H, Hurschler C. How Does a Varus Deformity of the Humeral Head Affect Elevation Forces and Shoulder Function? A Biomechanical Study With Human Shoulder Specimens. J Orthop Trauma. 2011;25(7): Voigt C, Hurschler C, Rech L, Vosshenrich R, Lill H. Additive fiber-cerclages in proximal humeral fractures stabilized by locking plates No effect on fracture stabilization and rotator cuff function in human shoulder specimens. Acta Orthop. 2009;80(4): Rose DM, Sutter EG, Mears SC, Gupta RR, Belkoff SM. Proximal Humeral Fractures: A Biomechanical Comparison of Locking Plate Constructs in a Cadaveric 3-Part Fracture Model. Geriatr Orthop Surg Rehabil. 2010;1(2): Sinatra PM, Jernick ML, Bledsoe G, Kaar SG. No Contribution of Tension-Reducing Rotator Cuff Sutures on Locking Plate Fixation in a 2-Part Proximal Humerus Fracture Model. J Orthop Trauma. 2014;28(8): Brunner A, Resch H, Babst R, Kathrein S, Fierlbeck J, Niederberger A, et al. The Humerusblock NG: a new concept for stabilization of proximal humeral fractures and its biomechanical evaluation. Arch Orthop Trauma Surg. 2012;132(7): Kathrein S, Kralinger F, Blauth M, Schmoelz W. Biomechanical comparison of an angular stable plate with augmented and non-augmented screws in a newly developed shoulder test bench. Clin Biomech. 2013;28(3): da Graca E, Okubo R, Shimano AC, Mazzer N, Barbieri CH. Biomechanics of Four Techniques for Fixation of the Four-Part Humeral Head Fracture. Acta Ortop Bras. 2013;21(1): Osterhoff G, Baumgartner D, Favre P, Wanner GA, Gerber H, Simmen HP, et al. Medial support by fibula bone graft in angular stable plate fixation of proximal humeral fractures: an in vitro study with synthetic bone. J Shoulder Elb Surg. 2011;20(5): Clavert P, Hatzidakis A, Boileau P. Anatomical and biomechanical evaluation of an intramedullary nail for fractures of proximal humerus fractures based on tuberosity fixation. Clin Biomech (Bristol, Avon) [Internet]. 2016;32: Available from: 265

266 122. Hughes RE, An KN. Force analysis of rotator cuff muscles. Clin Orthop Relat Res. 1996;(330): Poppen N, Walker P. Forces at the Glenohumeral Joint in Abduction. Clin Orthop Relat Res [Internet]. 1978;(135): Available from: Bergmann G, Graichen F, Bender A, Kaab M, Rohlmann A, Westerhoff P. In vivo glenohumeral contact forces - Measurements in the first patient 7 months postoperatively. J Biomech. 2007;40(10): Bergmann G, Graichen F, Bender A, Rohlmann A, Halder A, Beier A, et al. In vivo gleno-humeral joint loads during forward flexion and abduction. J Biomech. 2011;44(8): Westerhoff P, Graichen F, Bender A, Halder A, Beier A, Rohlmann A, et al. In vivo measurement of shoulder joint loads during activities of daily living. J Biomech. 2009;42(12): Konrad G, Markmiller M, Rüter a, Südkamp N. [Biomechanical evaluation of glenohumeral stability through muscle force vector analysis. Effect of a decreased glenoid inclination in shoulders with global rotator cuff tears]. Unfallchirurg [Internet]. 2007;110(2): Available from: Halder AM, O Driscoll SW, Heers G, Mura N, Zobitz ME, An KN, et al. Biomechanical comparison of effects of supraspinatus tendon detachments, tendon defects, and muscle retractions. J Bone Joint Surg Am [Internet]. 2002;84 A(5): Available from: Veeger HE, Van der Helm FC, Van der Woude LH, Pronk GM, Rozendal RH. Inertia and muscle contraction parameters for musculoskeletal modelling of the shoulder mechanism. J Biomech [Internet]. 1991;24(7): Available from: Rose DM, Gupta RR. Regarding Proximal humeral fracture fixation: a biomechanical comparison of two constructs. J Shoulder Elb Surg. 2013;22(9):E20 E Terrier A, Reist A, Merlini F, Farron A. Simulated joint and muscle forces in reversed and anatomic shoulder prostheses. J Bone Jt Surg - Br Vol [Internet]. 2008;90 B(6): Available from: Hepp P, Lill H, Bail H, Korner J, Niederhagen M, Haas NP, et al. Where should implants be anchored in the humeral head? Clin Orthop Relat Res [Internet]. 2003;(415): Available from: Hessmann MH, Sternstein W, Mehler D, Korner J, Hofmann A, Rommens PM. Are angle-fixed implants with elastic properties advantageous for the internal fixation of 266

267 proximal humerus fractures? Biomed Tech. 2004;49(12): Claes LE, Heigele CA. Magnitudes of local stress and strain along bony surfaces predict the course and type of fracture healing. J Biomech. 1999;32(3): Noordeen MHH, Lavy CBD, Shergill NS, Tuite JD, Jackson AM. Cyclical Micromovement and Fracture-Healing. J Bone Jt Surgery-British Vol. 1995;77B(4): Schindele S, Hackenbruch W, Sutter F, Scharer M, Leutenegger A. [Migration of Kirschner wires after surgical stabilization of injuries in the area of the shoulder--4 case reports]. Swiss Surg [Internet]. 1999;5(6): Available from: Lyons FA, Rockwood Jr. CA. Migration of pins used in operations on the shoulder. J Bone Joint Surg Am [Internet]. 1990;72(8): Available from: Kopylov P, Adalberth K, Jonsson K, Aspenberg P. Norian SRS versus functional treatment in redisplaced distal radial fractures: A randomized study in 20 patients. J Hand Surgery-British Eur Vol. 2002;27B(6): Liu D, Lei W, Wu ZX, Gao MX, Wan SY, Fu SC, et al. Augmentation of Pedicle Screw Stability With Calcium Sulfate Cement in Osteoporotic Sheep Biomechanical and Screw-bone Interfacial Evaluation. J Spinal Disord Tech. 2011;24(4): Lobenhoffer P, Gerich T, Witte F, Tscherne H. Use of an injectable calcium phosphate bone cement in the treatment of tibial plateau fractures: A prospective study of twentysix cases with twenty-month mean follow-up. J Orthop Trauma. 2002;16(3): Mattsson P, Larsson S. Unstable trochanteric fractures augmented with calcium phosphate cement. A prospective randomized study using radiostereometry to measure fracture stability. Scand J Surg [Internet]. 2004/11/ ;93(3): Available from: Faraj D, Kooistra BW, Vd Stappen WAH, Werre AJ. Results of 131 consecutive operated patients with a displaced proximal humerus fracture: An analysis with more than two years follow-up. Eur J Orthop Surg Traumatol. 2011;21(1): Geiger E, Maier M, Kelm A, Wutzler S, Seebach C, Marzi I. Functional outcome and complications following PHILOS plate fixation in proximal humeral fractures. Acta Orthop Traumatol Turc. 2010;44(1): Cartner JL, Hartsell ZM, Ricci WM, Tornetta P. Can We Trust Ex Vivo Mechanical Testing of Fresh-Frozen Cadaveric Specimens? The Effect of Postfreezing Delays. J Orthop Trauma. 2011;25(8): Inman VT, Saunders JBD, Abbott LC. Observations on the function of the shoulder joint. J Bone Jt Surg. 1944;26:

268 146. Klages A, Hurschler C, Wulker N, Windhagen H. [Muscle efficiency in total shoulder prosthesis implantation: dependence on position of the humeral head and rotator cuff function]. Biomed Tech [Internet]. 2001;46(9): Available from: Kedgley AE, Geoffrey AM, Ferreira LM, Darren SD, King GJW, Kenneth JF, et al. The effect of muscle loading on the kinematics of in vitro glenohumeral abduction. J Biomech. 2007;40(13): Sing SL, An J, Yeong WY, Wiria FE. Laser and electron-beam powder-bed additive manufacturing of metallic implants: A review on processes, materials and designs. Vol. 34, Journal of Orthopaedic Research p Amaritsakul Y, Chao C-K, Lin J. Multiobjective Optimization Design of Spinal Pedicle Screws Using Neural Networks and Genetic Algorithm: Mathematical Models and Mechanical Validation. Comput Math Methods Med [Internet]. 2013;2013:1 9. Available from: Exactech. Equinoxe FX Plate - Operative Technique [Internet] Available from: Synthes. PHILOS- Operative Technique [Internet] Available from: Biomet Trauma. S3 Humerus Plating System - Surgical Technique Commission IE. Safety of laser products - Part 1: Equipment classification and requirements : Creaform. ExaScan Scanner-Technical Specifications [Internet] Available from: Gil-Negrete N, Alonso A. Constitutive Models for Rubber VIII. CRC Press; Zhou Y, Yang L, Huang Y. Micro- and Macromechanical Properties of Materials. CRC Press; Fuchtmeier B, Hammer J, Nerlich M. Biomechanical behaviour of implant-reinforced subcapital humeral fractures. Zeitschrift Fur Met. 2002;93(8): Gautier E. Biomechanics of Osteosynthesis by Screwed Plates. In: Poitout DG, editor. Biomechanics and Biomaterials in Orthopedics [Internet]. London: Springer London; p Available from: Smith WR, Ziran BH, Anglen JO, Stahel PF,.,., et al. Locking plates: tips and tricks. J Bone Joint Surg Am [Internet]. 2007;89(10): Available from: Ellis T, Bourgeault C a, Kyle RF. Screw position affects dynamic compression plate strain in an in vitro fracture model. J Orthop Trauma. 2001;15(5):

269 161. Egol KA, Kubiak EN, Fulkerson E, Kummer FJ, Koval KJ. Biomechanics of locked plates and screws. J Orthop Trauma. 2004;18(8): Gautier E, Sommer C. Guidelines for the clinical application of the LCP. Injury [Internet]. 2003;34 Suppl 2:B Available from: Stoffel K, Dieter U, Stachowiak G, Gächter A, Kuster MS. Biomechanical testing of the LCP - How can stability in locked internal fixators be controlled? Injury. 2003;34(SUPPL. 2) Sinha S, Kelly CP. controversial topics in surgery. Ann R Coll Surg Engl. 2010;92(8): Cohen M, Amaral MV, Monteiro M, Brandão BL, Motta Filho GR. Osteosynthesis of proximal humeral end fractures with fixed-angle plate and locking screws: technique and results. Rev Bras Ortop [Internet] Jan 16;44(2): Available from: Strauss EJ, Schwarzkopf R, Kummer F, Egol K a. The current status of locked plating: the good, the bad, and the ugly. J Orthop Trauma [Internet]. 2008;22(7): Available from: Gardner MJ, Helfet DL, Lorich DG. Has locked plating completely replaced conventional plating? Am J Orthop (Belle Mead NJ). 2004;33(9): Gardner MJ, Nork SE, Huber P, Krieg JC. Less rigid stable fracture fixation in osteoporotic bone using locked plates with near cortical slots. Injury. 2010;41(6): Bottlang M, Doornink J, Byrd GD, Fitzpatrick DC, Madey SM. A Nonlocking End Screw Can Decrease Fracture Risk Caused by Locked Plating in the Osteoporotic Diaphysis. J Bone Joint Surg Am [Internet]. 2009;91(3): Available from: Tingart MJ, Lehtinen J, Zurakowski D, Warner JJP, Apreleva M. Proximal humeral fractures: Regional differences in bone mineral density of the humeral head affect the fixation strength of cancellous screws. J Shoulder Elb Surg. 2006;15(5): Boileau P, Walch G. The three-dimensional geometry of the proximal humerus. Implications for surgical technique and prosthetic design. J Bone Joint Surg Br. 1997;79(5): Robertson DD, Yuan J, Bigliani LU, Flatow EL, Yamaguchi K. Three-dimensional analysis of the proximal part of the humerus: relevance to arthroplasty. J Bone Joint Surg Am. 2000;82 A(11): Yewlett A, King A, Brooks F, Evans R, Williams R. What is the Single Most Important 269

270 Technical Aspect When Fixing a Proximal Humeral Fracture with A PHILOS Plate? MOJ Orthop Rheumatol [Internet]. 2016;6(2):2 6. Available from: Fuchs M, Burchhardt H, Losch A, Stürmer K-M. Open Reduction and Internal Fixation of Proximal Humerus Fractures with a Cannulated Blade Plate. Orthop Traumatol [Internet]. 2002;10(4): Available from: Badman BL, Mighell M, Drake GN. Proximal Humeral Nonunions: Surgical Technique with Fibular Strut Allograft and Fixed-Angle Locked Plating. Tech Shoulder Elb Surg [Internet]. 2006;7(2). Available from: _Nonunions Surgical_Technique.5.aspx 176. Morelli R, Travizanuto RED. Proximal Humerus Fractures: Comparative Study Of Two Different Fixation Methods. Acta Ortop Bras. 2010;18(2): Olerud P, Ahrengart L, Ponzer S, Saving J, Tidermark J. Internal fixation versus nonoperative treatment of displaced 3-part proximal humeral fractures in elderly patients: a randomized controlled trial. J Shoulder Elbow Surg [Internet]. 2011;20(5): Available from: Frich LH, Jensen NC. Bone properties of the humeral head and resistance to screw cutout. Int J Shoulder Surg [Internet]. 2014;8(1):21 6. Available from: rendertype=abstract 179. Meier RA, Messmer P, Regazzoni P, Rothfischer W, Gross T. Unexpected high complication rate following internal fixation of unstable proximal humerus fractures with an angled blade plate. J Orthop Trauma [Internet]. 2006;20(4): Available from: Liew ASL, Johnson JA, Patterson SD, King GJW, Chess DG. Effect of screw placement on fixation in the humeral head. J Shoulder Elb Surg. 2000;9(5): Badman BL, Mighell M. Fixed-angle locked plating of two-, three-, and four-part proximal humerus fractures. J Am Acad Orthop Surg [Internet]. 2008;16(5): Available from: Raphael JS, Booker MR, Saing MH, Beaver AB, Marvil SC, Puglisi AS, et al. Locked plate fixation of proximal humerus fractures using cannulated subchondral support pegs. Shoulder Elb [Internet]. 2013;5(3): Available from: Kelly BP, Bennett CR. Design and validation of a novel Cartesian biomechanical testing 270

271 system with coordinated 6DOF real-time load control: application to the lumbar spine (L1-S, L4-L5). J Biomech. 2013;46(11): Voigt C, Hurschler C, Althainz J, Vosshenrich R, Lill H. Effect of additive fiber cerclage of the rotator cuff in angle-stable plate-stabilized proximal humeral fractures. A biomechanical study of human shoulder specimens. Unfallchirurg. 2008;111(7): Agudelo J, Schurmann M, Stahel P, Helwig P, Morgan SJ, Zechel W, et al. Analysis of efficacy and failure in proximal humerus fractures treated with locking plates. J Orthop Trauma. 2007;21(10): Aksu N, Göğüş A, Kara AN, Işiklar ZU. Complications encountered in proximal humerus fractures treated with locking plate fixation. Acta Orthop Traumatol Turc [Internet]. 2010;44(2): Available from: MacLeod AR, Simpson AHRW, Pankaj P. Reasons why dynamic compression plates are inferior to locking plates in osteoporotic bone: a finite element explanation. Comput Methods Biomech Biomed Engin [Internet]. 2015;18(16): Available from: Aalco Metals Ltd. Stainless Steel - Grade 316 Datasheet [Internet] p Available from: Sawbones. Biomechanical Test Materials [Internet] p. Available from: Rho JY, Ashman RB, Turner CH. Young s modulus of trabecular and cortical bone material: Ultrasonic and microtensile measurements. J Biomech. 1993;26(2): Pal S. Mechanical Properties of Biological Materials. In: Design of Artificial Human Joints & Organs [Internet] p Available from: Cegoñino J, García Aznar JM, Doblaré M, Palanca D, Seral B, Seral F. A comparative analysis of different treatments for distal femur fractures using the finite element method. Comput Methods Biomech Biomed Engin [Internet]. 2004;7(5): Available from: Lovald ST, Khraishi T, Wood J, Wagner J, Baack B, Kelly J. Finite Element Analysis of Screw-Plate Systems for Fixation of Parasymphyseal Fractures of the Mandible. J Mech. 2007;23(1): Zhou JJ, Zhao M, Liu D, Liu LH-Y, Du C-F. Biomechanical Property of a Newly Designed Assembly Locking Compression Plate: Three-Dimensional Finite Element Analysis. J Healthc Eng. 2017; Feerick EM, Kennedy J, Mullett H, FitzPatrick D, McGarry P. Investigation of metallic and carbon fibre PEEK fracture fixation devices for three-part proximal humeral 271

272 fractures. Med Eng Phys. 2013;35(6): Yang P, Zhang Y, Liu J, Xiao J, Ma LM, Zhu CR. Biomechanical effect of medial cortical support and medial screw support on locking plate fixation in proximal humeral fractures with a medial gap: A finite element analysis. Acta Orthop Traumatol Turc. 2015;49(2): He Y, He J, Wang F, Zhou D, Wang Y, Wang B, et al. Application of Additional Medial Plate in Treatment of Proximal Humeral Fractures With Unstable Medial Column: A Finite Element Study and Clinical Practice. Medicine (Baltimore) [Internet] Oct;94(41):e1775. Available from: Zhang YK, Wei HW, Lin KP, Chen WC, Tsai CL, Lin KJ. Biomechanical effect of the configuration of screw hole style on locking plate fixation in proximal humerus fracture with a simulated gap: A finite element analysis. Injury. 2016;47(6): Kayabasi O, Ekici B. The effects of static, dynamic and fatigue behavior on threedimensional shape optimization of hip prosthesis by finite element method. Mater Des [Internet]. 2007;28(8): Available from: Er MS, Verim O, Eroglu M, Altinel L, Gokce B, Tasgetiren S. Biomechanical evaluation of syndesmotic screw design via finite element analysis and Taguchi s method. J Am Pod Med Assoc. 2015;105(1): Haglin JM, Eltorai AEM, Gil JA, Marcaccio SE, Botero-Hincapie J, Daniels AH. Patient- Specific Orthopaedic Implants. Vol. 8, Orthopaedic Surgery p Chen X, He K, Chen Z. A Novel Computer-Aided Approach for Parametric Investigation of Custom Design of Fracture Fixation Plates. Comput Math Methods Med. 2017; Chen X, He K, Chen Z, Wang L. A parametric approach to construct femur models and their fixation plates. Biotechnol Biotechnol Equip [Internet] May 3;30(3): Available from: Freude T, Schroeter S, Plecko M, Bahrs C, Martetschlaeger F, Kraus TM, et al. Dynamic-locking-screw (DLS)-leads to less secondary screw perforations in proximal humerus fractures. BMC Musculoskelet Disord [Internet]. 2014;15(1):194. Available from: rendertype=abstract 205. Davenport SR, Lindsey RW, Leggon R, Miclau T, Panjabi M. Dynamic compression plate fixation: A biomechanical comparison of unicortical vs bicortical distal screw fixation. J Orthop Trauma [Internet]. 1988;2(2): Available from:

273 &partnerID=40&md5=457bb950c6d4ce798bf33cb3aa Beaupre GS, Giori NJ, Caler WE, Csongradi J. A comparison of unicortical and bicortical end screw attachment of fracture fixation plates. J Orthop Trauma. 1992;6(3): Sun Y, Amirrasouli B, Razavi SB, Li QM, Lowe T, Withers PJ. The variation in elastic modulus throughout the compression of foam materials. Acta Mater. 2016;110: Inzana JA, Varga P, Windolf M. Implicit modeling of screw threads for efficient finite element analysis of complex bone-implant systems. Journal of Biomechanics. 2016; 209. McDonald E, Kwiat D, Kandemir U. Geometry of Proximal Humerus Locking Plates. J Orthop Trauma [Internet]. 2015;29(11):e Available from: Lumelsky VJ. On fast computation of distance between line segments. Inf Process Lett [Internet] Aug [cited 2017 May 10];21(2): Available from: Chittawadigi RG, Saha SK. An analytical method to detect collision between cylinders using dual number algebra. In: IEEE International Conference on Intelligent Robots and Systems p Ketchel JS, Larochelle PM. Collision Detection of Cylindrical Rigid Bodies Using Line Geometry. 29th Mech Robot Conf [Internet]. 2005; Available from: Ketchel J, Larochelle P. Collision detection of cylindrical rigid bodies for motion planning. Proc - IEEE Int Conf Robot Autom. 2006;2006: Hertel R, Knothe U, Ballmer FT. Geometry of the proximal humerus and implications for prosthetic design. J Shoulder Elb Surg. 2002;11(4): Chu C, BelavýDL, Armbrecht G, Bansmann M, Felsenberg D, Zheng G. Fully automatic localization and segmentation of 3D vertebral bodies from CT/MR images via a learning-based method. PLoS One. 2015;10(11) Xia Y, Fripp J, Chandra SS, Schwarz R, Engstrom C, Crozier S. Automated bone segmentation from large field of view 3D MR images of the hip joint. Phys Med Biol [Internet]. 2013;58(20): Available from: Chao CK, Lin J, Putra ST, Hsu C-C. A neurogenetic approach to a multiobjective design optimization of spinal pedicle screws. J Biomech Eng [Internet]. 2010;132(9): Available from: Lee CH, Hsu C-C, Huy DC. An Optimization Study of the Screw Orientation on the Interfacial Strength of the Anterior Lumbar Plate System Using Neuro-Genetic Algorithms and Experimental Validation. J Biomech Eng [Internet]. 2014;136:

274 Available from: Choi JW, Kim N. Clinical application of three-dimensional printing technology in craniofacial plastic surgery. Vol. 42, Archives of Plastic Surgery p He J, Li D, Lu B, Wang Z, Tao Z. Custom fabrication of composite tibial hemi-knee joint combining CAD/CAE/CAM techniques. Proc Inst Mech Eng H. 2006;220(8): Harrysson OLA, Hosni YA, Nayfeh JF. Custom-designed orthopedic implants evaluated using finite element analysis of patient-specific computed tomography data: femoralcomponent case study. BMC Musculoskelet Disord [Internet]. 2007;8(iv):91. Available from: rendertype=abstract 222. Dai KR, Yan MN, Zhu ZA, Sun YH. Computer-Aided Custom-Made Hemipelvic Prosthesis Used in Extensive Pelvic Lesions. J Arthroplasty. 2007;22(7): Owsley KC, Gorczyca JT. Displacement/Screw Cutout After Open Reduction and Locked Plate Fixation of Humeral Fractures. J Bone Jt Surg. 2008;90: Aaron D, Shatsky J, Paredes JC, Jiang C, Parsons BO, Flatow EL. Proximal humeral fractures: internal fixation. J Bone Jt Surg - Ser A [Internet]. 2013;94(24): Available from: Bengard MJ, Gardner MJ. Screw Depth Sounding in Proximal Humerus Fractures to Avoid Iatrogenic Intra-articular Penetration. J Orthop Trauma. 2011;25(10): Jin L, Guo J, Guo J, Yin Y, Hou Z, Zhang Y. Clinical Effects of the Probing Method with Depth Gauge for Determining the Screw Depth of Locking Proximal Humeral Plate. Biomed Res Int. 2016;2016: van Rooyen RS, Marsh A, Strick NE, Hong T, Meeghan CO. Humeral head collapse causing locked screw penetration of the glenohumeral joint. Inj Extra [Internet]. 2006;37(12): Available from: van Rooyen RS, Marsh A, Strick NE, Hong T, Meeghan CO. Humeral head collapse causing locked screw penetration of the glenohumeral joint. Inj Extra. 2006;37(12): Poitout DG. Biomechanics and biomaterials in orthopedics, second edition. Biomechanics and Biomaterials in Orthopedics, Second Edition p Kowalski RJ, Ferrara LA, Benzel EC. Biomechanics of the spine. Neurosurg Q [Internet]. 2005;15(1): Available from: s &doi= %2f01.wnq e&partnerid=40&md5= 274

275 7ad7d05d5daa32e7c0e9cb89a7f8763b 231. Lin S-J, Tsai Y-H, Yang T-Y, Shen S-H, Huang K-C, Lee MS. Medial Calcar Support and Radiographic Outcomes of Plate Fixation for Proximal Humeral Fractures. Biomed Res Int [Internet]. 2015;2015:1 8. Available from: Kowalski RJ, Ferrara LA, Benzel EC. Biomechanics of the spine. Neurosurg Q. 2005;15(1): Namdari S, Lipman AJ, Ricchetti ET, Tjoumakaris FP, Russell Huffman G, Mehta S. Fixation strategies to prevent screw cut-out and malreduction in proximal humeral fracture fixation. Clin Orthop Surg. 2012;4(4): Badman BL, Mighell M. Fixed-angle locked plating of two-, three-, and four-part proximal humerus fractures. J Am Acad Orthop Surg. 2008;16(5): Mehling I, Klitscher D, Mehling AP, Nowak TE, Sternstein W, Rommens PM, et al. Volar fixed-angle plating of distal radius fractures: screws versus pegs--a biomechanical study in a cadaveric model. J Orthop Trauma [Internet]. 2012;26(7): Available from: Weninger P, DallʼAra E, Leixnering M, Pezzei C, Hertz H, Drobetz H, et al. Volar Fixed- Angle Plating of Extra-Articular Distal Radius Fractures A Biomechanical Analysis Comparing Threaded Screws and Smooth Pegs. J Trauma Inj Infect Crit Care [Internet]. 2010;69(5):E Available from: Martineau PA, Waitayawinyu T, Malone KJ, Hanel DP, Trumble TE. Volar Plating of AO C3 Distal Radius Fractures: Biomechanical Evaluation of Locking Screw and Locking Smooth Peg Configurations. J Hand Surg Am. 2008;33(6): Yao J, Park MJ, Patel CS. Biomechanical comparison of volar locked plate constructs using smooth and threaded locking pegs. Orthopedics [Internet]. 2014;37(2):e Available from: Boretto JG, Pacher N, Giunta D, Gallucci GL, Alfie V, De Carli P. Comparative clinical study of locking screws versus smooth locking pegs in volar plating of distal radius fractures. J Hand Surg Eur Vol [Internet]. 2014;39(7): Available from: Karunratanakul K, Schrooten J, Van Oosterwyck H. Finite element modelling of a unilateral fixator for bone reconstruction: Importance of contact settings. Med Eng Phys. 2010;32(5): Kennedy J, Feerick E, McGarry P, FitzPatrick D, Mullett H. Effect of calcium triphosphate cement on proximal humeral fracture osteosynthesis: a finite element 275

276 analysis. J Orthop Surg (Hong Kong) [Internet]. 2013;21(2): Available from: MacLeod AR, Pankaj P, Simpson AHRW. Does screw-bone interface modelling matter in finite element analyses? J Biomech. 2012;45(9): Mehling I, Klitscher D, Mehling AP, Nowak TE, Sternstein W, Rommens PM, et al. Volar fixed-angle plating of distal radius fractures: screws versus pegs--a biomechanical study in a cadaveric model. J Orthop Trauma. 2012;26(7): Weninger P, DallʼAra E, Leixnering M, Pezzei C, Hertz H, Drobetz H, et al. Volar Fixed- Angle Plating of Extra-Articular Distal Radius Fractures A Biomechanical Analysis Comparing Threaded Screws and Smooth Pegs. J Trauma Inj Infect Crit Care. 2010;69(5):E Yao J, Park MJ, Patel CS. Biomechanical comparison of volar locked plate constructs using smooth and threaded locking pegs. Orthopedics. 2014;37(2):e Boretto JG, Pacher N, Giunta D, Gallucci GL, Alfie V, De Carli P. Comparative clinical study of locking screws versus smooth locking pegs in volar plating of distal radius fractures. J Hand Surg Eur Vol. 2014;39(7): Jost B, Spross C, Grehn H, Gerber C. Locking plate fixation of fractures of the proximal humerus: Analysis of complications, revision strategies and outcome. J Shoulder Elb Surg. 2013;22(4): Shulman BS, Egol KA. Open reduction internal fixation for proximal humerus fractures: Indications, techniques, and pitfalls. Bull NYU Hosp Jt Dis [Internet]. 2013;71(SUPPL. 2):S54 9. Available from: Perren SM. Evolution of the internal fixation of long bone fractures. The scientific basis of biological internal fixation: choosing a new balance between stability and biology. J Bone Joint Surg Br. 2002;84(8): Bottlang M, Feist F. Biomechanics of Far Cortical Locking. J Orthop Trauma [Internet]. 2011;25:S21 8. Available from: Bottlang M, Doornink J, Fitzpatrick DC, Madey SM. Far Cortical Locking Can Reduce Stiffness of Locked Plating Constructs While Retaining Construct Strength. J Bone Joint Surg Am [Internet]. 2009;91(8): Available from: Döbele S, Gardner M, Schröter S, Höntzsch D, Stöckle U, Freude T. DLS 5.0--the biomechanical effects of dynamic locking screws. PLoS One [Internet]. 276

277 2014;9(4):e Available from: rendertype=abstract 253. Pohlemann T, Gueorguiev B, Agarwal Y, Wahl D, Sprecher C, Schwieger K, et al. Dynamic locking screw improves fixation strength in osteoporotic bone: an in vitro study on an artificial bone model. Int Orthop. 2015;39(4): Foundation AO. Müller AO Classification of Fractures Long Bones [Internet]. AO Foundation; p. Available from: Clamping-Solutions-SA. Pipe Hanger [Internet]. Available from: Transhow-Industries-Ltd. Pipe Clamp with Rubber [Internet]. Available from: BJM. Flexseal Flexible Coupling-MDC32 [Internet]. Available from: Badger-Ladder. Steel Base Plate [Internet]. Available from: 08d6e5fb8d27136e95/b/a/base_plate_w.jpg 259. Creaform. Photograph: ExaScan [Internet] Available from: Gooding BWT, Geoghegan JM. (i) Plate fixation of proximal humeral fractures. Orthop Trauma [Internet]. 2013;27(3): Available from: host-live 261. Synthes. DLS Dynamic Locking Screw-Combined with LCP Locking Compression Plate [Internet] Available from: Mobile/Synthes International/Product Support Material/legacy_Synthes_PDF/ pdf 262. RDP-Electronics. Model 41 Tension/Compression Load Cell (Universal Load Cell) [Internet]. Available from: 277

278 Appendices Appendix I: Technical Drawings of Compression and Bending Test Rig Figure. A.1. Technical drawing of front view of the test rig designed for compression and bending 278

279 Figure. A.2. Technical drawing of top and bottom view of the test rig designed for compression and bending 279

280 Appendix II: Technical Drawings of Axial Rotation Test Rig Figure. A.3. Technical drawing of left side view of the test rig designed for axial rotation 280

281 Figure. A.4. Technical drawing of front view of the test rig designed for axial rotation 281

282 Figure. A.5. Technical drawing of top view of the test rig designed for axial rotation 282

283 Appendix III: Experimental Protocol (Proposed: August 2014) Comparison Tests To compare the performance of synthetic humeri that we will use with the performance of cadaveric humeri used in the literature, comparison tests will be conducted. These tests will involve axial compression, rotation and eccentric loading. For the first, the procedure will be adapted from that of Wallace et al. [77] and for the latter two, it will be from Sanders et al. [69]. A total of twenty synthetic humeri (Model 1028; Pacific Research Laboratories, Inc., Vashon Island, USA) will be obtained. Axial Compression Ten synthetic humeri will be obtained and a six-hole proximal humerus locking plate (Synthes Inc., Paoli, PA, USA), will be implanted onto to each one using standard orthopaedic fracture fixation techniques. A 15-mm resection osteotomy will be created using an oscillating saw at the humeral surgical neck starting at the level of the most distal humeral head screw hole to simulate a comminuted two-part surgical neck fracture. Both the humeral head and the distal end will be potted within methyl-methacrylate blocks and glued to ensure that they are securely fixed. With the distal end fixed and the humeral shaft perpendicular to the ground, the axial load will be applied. The material testing machine will be set at displacement control mode and an increasing axial load will be applied to the specimen at a displacement rate of 5 mm/min until the load displacement curve showed a clear deviation from linearity. The range of this linear region would have been determined from the preliminary experiments. Linear elastic stiffness will then be calculated from the load displacement plot and compared with the values from the study by Wallace et al. Eccentric Loading and Axial Rotation Similarly, ten synthetic humeri will be used for rotation and eccentric loading. A three-part fracture will be simulated by making cuts through the intertubercular groove and the surgical neck, 1 cm distal to the articular surface resulting in separated humeral head, greater tuberosity, and humeral shaft. Humeri will also be cut 23 cm distal from the apex of the humeral head and as for the compression test; both the head and distal end will be potted in methylmethacrylate blocks. The specimen would be instrumented with the same locking plate as for compression test. 283

284 For eccentric loading, the load will be applied onto the block, 2.5 cm away from the humeral shaft axis in the corresponding directions, to achieve varus, valgus, flexion and extension. These loading conditions, especially varus bending have direct clinical relevance, as mentioned in the introduction. For all four cases, material testing machine will be set at a loadcontrol mode at load rate of 10 N/s and loaded up to the maximum value of 80 N. By loading up to 80 N, will allow us to test the same specimen under all four load cases as it will still be within the elastic region. As before, the elastic region of the specimen would be determined from the preliminary experiments. Load-displacement curves will allow calculation of the stiffness for varus, valgus, flexion and extension which will be compared with stiffness values from the study of Sanders et al. The same ten specimens from eccentric loading will be tested under axial rotation in a similar fashion. With the distal end fixed, the humeral head block will be rotated both internal and externally by 8 o at a rate of 1 o /s with the machine set at displacement control mode. The gradient of the torque-angular displacement curves, rotational stiffness, will be calculated for both internal and external rotation compared with the values from those that Sanders et al. calculated. Main Experiments Once the comparison tests have been performed and the experimental results have been compared with the results from the literature, the main tests will be performed. A total of forty synthetic humeri will be obtained and for each humerus, a two-part fracture will be simulated by using a surgical saw to create a transverse osteotomy at the level of surgical neck, leaving a 10-mm fracture gap. Specimens will also be cut 17 cm distal to the apex of the humeral head to ensure that they are all of the same length and the humeral head and the distal end will be potted within methyl-methacrylate blocks and glued to ensure that they are securely fixed. These forty specimens will be divided into four groups of tens and each group undergoing osteosynthesis with one of the four plates: six-hole proximal humerus locking plate (Synthes Inc., Paoli, PA, USA), PHILOS plate (Synthes Inc., Paoli, PA, USA), S3 locking plate (DePuy Orthopaedics, Warsaw, IN, USA) and Equinoxe Fx Plate (Exactech Inc. Gainesville, FL, USA). These plates have been selected to represent modern implant designs. With the specimen fixed, each specimen will undergo axial compression, rotation and eccentric loading by the use of the same steps described previously for comparison tests. 284

285 From these, obtained load-deformation curves will be used to determine the corresponding construct stiffness. In addition to the stiffness, two other measurements will be made. A crack opening gauge will be attached to both sides of the fracture gap to monitor the fracture change in displacement of the two fragments over time. Also, three strain gauges will be attached to the humeral shaft, spread evenly across the distance to record the strain values at the desired point. All the stiffness values along with the fracture gap displacement and the data from the three strain gauges will be used for the validation of the finite element model. 285

286 Appendix IV: Experimental Protocol (Proposed: November 2014) First, a set of tests will be performed reproducing biomechanical studies from the literature that used cadaveric humeri. Results from these comparison tests will be compared with results from the literature to determine how similar the mechanical behaviour of synthetic humeri is to that of the cadaveric ones. Comparison Tests Comparison tests will involve axial compression as performed by Wallace et al. A total of ten synthetic humeri (Model 1028; Pacific Research Laboratories, Inc., Vashon Island, USA) will be obtained. Each specimen will be implanted with a PHILOS plate (Synthes Inc., Paoli, PA, USA). A 15-mm resection osteotomy will be created using an oscillating saw at the humeral surgical neck starting at the level of the most distal humeral head screw hole to simulate a comminuted two-part surgical neck fracture. The load will be applied to the humeral head with the distal end glued in a pot to ensure secure fixing. With the distal end fixed and the humeral shaft perpendicular to the base, the axial load will be applied. The material testing machine will be set at displacement control mode and an increasing axial load will be applied to the specimen at a displacement rate of 5 mm/min until the load displacement curve shows a clear deviation from linearity. The range of this linear region would have been determined from the preliminary experiments. Linear elastic stiffness will then be calculated from the load displacement plot and compared with the values from the study by Wallace et al. Figure A.6. Set-up for axial compression, oriented loading and cantilever bending (L to R) 286

287 Main Experiments Once the comparison tests have been performed and the experimental results have been compared with the results from the literature, tests will be performed to investigate in vitro biomechanical performance of different proximal humerus plates. A total of thirty synthetic humeri will be obtained and for each humerus, a two-part fracture will be simulated by using a surgical saw to create a transverse osteotomy at the level of surgical neck, leaving a 10 mm fracture gap. Specimens will also be cut 17 cm distal to the apex of the humeral head to ensure that they are all of the same length. These thirty specimens will be divided into three groups of tens with each group undergoing osteosynthesis with one of the three plates: PHILOS plate (Synthes Inc., Paoli, PA, USA), S3 locking plate (DePuy Orthopaedics, Warsaw, IN, USA) and Equinoxe Fx Plate (Exactech Inc. Gainesville, FL, USA). These plates have been selected to represent modern implant designs. With the specimen fixed, each specimen will undergo axial compression as performed by Wallace et al. as well as oriented and cantilever loading. For oriented loading, the humeral shaft will be oriented at 20 o of abduction and vertical load will be applied to the superior aspect of the humeral head (Fig. A.6.). The specimen will be incrementally loaded to failure at a rate of 10 cm/min with the failure defined as a marked decrease or discontinuity in the load displacement curve. For cantilever testing, humeral head will be fixed (Fig. A.6). Distal end will be cyclically loaded in a cantilever fashion to ±5 mm of submaximal displacement rate of 1 mm/s for 100 cycles in the sagittal plane for flexion/extension bending and in frontal plane for varus/valgus bending. Then, each specimen will be loaded at 1 mm/s in varus until failure. For all three tests, load-deformation curves will be used to determine the corresponding construct stiffness. For the cyclic steps, the load will be plotted against time and displacement after a set number of cycles will be recorded. Also, with the 3D laser scanners available in our research laboratory, loading will be paused regularly during the tests to take 3D scans of the specimens. 3D scans will be compared to determine the 3D deformation of the specimen and the fracture gap closure. 287

288 Appendix V: Technical Drawings of Testing Machine Components Load Cell Drawing Figure A.7. Drawing of the load cell ordered for ESH testing machine (Model 41, RDP Electronics, West Midlands, UK) [262] Load Cell-Testing Machine Adaptor Design Figure A.8. Design of the adaptor to connect the RDP load cell to the ESH testing machine 288