Modeling the length of the care episode after hip fracture: does the type of fracture Matter?

Transcription

1 Scandinavian Journal of Surgery 98: , 009 Modeling the length of the care episode after hip fracture: does the type of fracture Matter? r. sund 1, J. riihimäki, M. Mäkelä 1,3, a. Vehtari, p. lüthje 4, t. huusko 5, u. häkkinen 1 1 National Institute for Health and Welfare, Helsinki, Finland Department of Biomedical Engineering and Computational Science, Helsinki University of Technology, Helsinki, Finland 3 Services for the Aged and Disabled, City of Vantaa, Vantaa, Finland 4 Kuusankoski Regional Hospital, Kuusankoski, Finland 5 The Social Insurance Institution, Health Department/Rehabilitation Section, Helsinki, Finland abstract Background and Aims: hip fractures are common events that require intensive operative hospital care and alengthy rehabilitation. the effect of hip fracture type on successful rehabilitation is not well known. the aim of this study is to model and compare the length of the care episodes between intra- and extracapsular hip fractures in finland. Material and Methods: hip fracture patients living at home in finland were followed using register-based data. patient characteristics, outcomes, and length of stay (los) distributions were analyzed using a bayesian nonparametric multilayer perceptron (Mlp) network model. Results: Mortality was similar in intra- and extracapsular hip fractures. patients were more likely to need long-term care after extracapsular hip fracture. the average los at the surgical ward was similar for intra- and extracapsular fractures (1.7 weeks), but there was aconsiderable difference for the total inpatient los between the groups (5. weeks vs. 6.9 weeks). intracapsular fractures had asimple unimodal los distribution, whereas the los distribution for the extracapsular fractures was multimodal with two clear peaks. patients with more comorbidities required a longer los. Conclusions: the causes for differences in los between fracture types were most likely due to the different surgical methods and rehabilitation practices for the fracture types. as national guidelines suggest similar rehabilitation for all hip fracture patients, there is aneed for early and aggressive rehabilitation of patients with extracapsular fractures, including full-weight bearing for all but selected patients. Key words: Intracapsular/extracapsular hip fractures; care episodes; rehabilitation; length of stay; register-based data; Bayesian modeling Correspondence: Reijo Sund, D.Soc.Sc. National Institute for Health and Welfare (THL) P.O. Box 30 FI-0071 Helsinki Finland reijo.sund@thl.fi

2 170 R. Sund, J. Riihimäki, M. Mäkelä, A. Vehtari, P. Lüthje, T. Huusko, U. Häkkinen INTRODUCTION Among the ageing population, hip fractures are common events that can be devastating and costly (1). Besides having a high mortality rate, they also require intensive operative hospital care and alengthy rehabilitation (). In anumber of cases, rehabilitation is unsuccessful and the patient requires long term residential care, leading to even higher costs of care (3). As the incidence of hip fractures increases sharply with age, and as populations age, the outcome ofhip fracture care is critical for the health system and elderly care (1). Hip fracture events serve as an excellent model for studying the determinants of successful geriatric rehabilitation, as the diagnosis of hip fractures is usually straightforward, best practices are well established, and critical outcomes are readily defined. The success of rehabilitation after ahip fracture has been shown to depend on the comorbidity and cognitive impairment level of the patient (4, 5), but also on the way the rehabilitation care isprovided (6 8). The term hip fracture refers to a fracture of the upper end of the femur. Hip fractures are commonly classified into intra- and extracapsular according to their relationship to the capsular attachment of the hip. Patients with intracapsular fractures tend to be younger, more mobile, and are less likely to use walking aids or live in residential accommodation; they also have aconsiderably shorter length of hospital stay than those with extracapsular fractures (9 1). However, the effect of hip fracture type on successful rehabilitation is not well known. The aim of this study is to model and compare the length of the care episodes between intra- and extracapsular hip fractures in Finland. MATERIAL AND METHODS The total population of hip fracture patients in was identified in the Finnish Health Care Register using the 10 th revision of the International Classification of Diseases (ICD-10) diagnosis codes S7.0 (fracture of the neck of femur), S7.1 (trochantheric fracture) and S7. (subtrochanteric fracture). Data on the use of residential care and deaths for this hip fracture population were extracted from the Finnish Health Care Register and from the national Causes of Death Statistics using the unique personal identification number of each patient. Records in these registers include a lot of data: patient and provider ID-numbers, age, sex, area codes, and diagnosis and operation codes, as well as dates of admission, operation and discharge (or death). The validity of Finnish register-data for hip fracture follow-up studies is very good (13). The actual study population was restricted to patients aged 65 or older having their first hip fracture and living at home at the time of fracture (n=15544). Hip fractures were classified into intra-capsular (S7.0) and extracapsular fractures (S7.1, S7.), using diagnosis codes recorded for the primary operative care period. Other covariates included were age at the time of fracture, sex, operation type, hospital district, days of inpatient care during one year before the fracture, and the place from which the patient was admitted to the surgical ward (home, nursing home, primary care inpatient ward, hospital). Also certain preexisting comorbid conditions were identified from the data in a manner similar to that used in previous register-based hip fracture studies (14). There are three main outcomes for hip fracture treatment: 1) the patient is able to return home, ) the patient becomes along-term patient, or 3) the patient dies. Discharge home may take place directly from the surgical ward orafter some inpatient rehabilitation. The hip fracturecareepisode was defined as acontinuous episode of inpatient treatment, beginning with operative treatment on the surgical ward and ending with discharge to home. If the rehabilitation period exceeded four months, the patient was classified as along-term patient (15). The hip fracture care episode was considered successful if the patient was still staying at home two weeks after the discharge home. STATISTICAL METHODS ABayesian nonparametric multilayer perceptron (MLP) network model was used for the statistical modeling (16, 17). Bayesian formalism gives aprincipled approach to combine the MLP model description,prior information and the data. The MLP model allows flexible ways to present distributions without fixing the functional forms in advance, and makes it possible to deal with nonlinearities and complex interactions between covariates in the data. Modeling was formulated as aclassification task. For the length of stay (LOS) distributions each class corresponded to the length of stay at an accuracy of one week. The probability that a class target y had value j (out of k possible outputs) was computed using softmax likelihood exp( f j ( x, w)) p( y j x, w). exp( f ( x, w)) k k The latent function was modelled with an MLP function, corresponding to the network with one hidden layer of the form k m d k 0 wkj tanh( w j0 w ji xi) j1 i1 f ( x, w) w, where a d-dimensional input vector is denoted by x, and w represents weight parameters in the model. Indices i and j correspond to input and hidden layers, respectively. The prior was set indirectly via the network and weight priors toafunction space. The following hierarchical structure was used w ~ N( 0, ) ~ Inv gamma( i ji ave i ave, v ) ~ Inv gamma(, v 0,ave ), which favours smooth solutions since small weights produce smooth functions (16, 17). The separate prior variance i for each input automatically determines the relevance of various inputs. The variance hyperparameter i is controlled by the next level hyperparameters ave and v, that are determined by the third level hyperparameters 0 and v, ave.the suggested fixed values for the highest level hyperparameters were used (16). An integration of parameters posterior was made using stochastic Markov chain Monte Carlo (MCMC) simulation techniques. The computations were made using the MCMC Methods for MLP and GP and Stuff (for Matlab) -toolbox V.1 ( The relative risks for the background characteristics and process variables between the intracapsular and extraca-

3 Modeling LOS after hip fracture 171 psular groups were estimated using abinomial model for discrete variables, and anormal distribution for continuous variables. In both these cases a noninformative prior was chosen, and credible intervals were obtained using simulation samples from the posterior distributions. To analyze the differences between the binary valued outcome variables of intra- and extracapsular groups, the MLP model with a single output was used. In this case the probability that a binary class target y had value 1was computed using logistic transformation 1 p( y 1 x, w). 1 exp( f ( x, w)) This allowed us to obtain the relative risks, adjusted for background characteristics and process variables. To illustrate the effects of the covariates on the probability of discharge during certain weeks, we computed average predictive comparisons for the probability of discharge during these weeks for the variables of interest. The comparisons wereobtained for alarge simulation test group by observing the average difference in probability corresponding to aunit difference in each of the input covariates (18). For the two covariates (the place from which the patient was admitted to the surgical ward and operation type), the average difference was compared to the average value. RESULTS The basic characteristics of the hip fracture population are reported in Table 1. Intracapsular fractures were more common, accounting for about 63% of all hip fractures. Patients in both fracture groups had similar background characteristics.osteoarthritis was more common among the extracapsular patients. There were no differences in mortality figures between intra- and extracapsular hip fractures (Table 1). Ahigher proportion of patients with intracapsular hip fractures were discharged home directly from the surgical ward orduring the rehabilitation period. Thus, it was more likely that a patient with an extracapsular hip fracture required long-term care. The proportion of patients dying before returning home or becoming a long-term patient was 16.% and 16.6% for intra- and extracapsular patients, respectively. The proportion of patients at home one year after fracture did not differ between fracture types. The average LOS at the surgical ward was similar for both groups (1.7 weeks), but there was a significant difference in the length of total inpatient treatment between intra- and extracapsular fractures (5. weeks vs. 6.9 weeks). The mode class for the length of stay was 3 weeks for intracapsular fractures, while the LOS distribution had a typical right-skewed shape, where the mode is reached quite quickly and the probabilities for longer stays steadily decrease (Fig. 1A). For extracapsular hip fractures, the shape of the LOS distribution was non-standard, having no systematic decrease in probabilities after reaching a peak at weeks until the mode in the distribution seen at 7 8 weeks. The disparity between the shapes of the distributions for intracapsular and extracapsular fractures was clear (Fig. 1A). For both of the fracture types, the MLP model was also used to examine how the differences in the observed LOS and predicted LOS relate to the success of discharge home. The discharge was considered successful if the patient was staying at home two weeks after discharge. In Fig. 1B, the proportions of successful discharges home are shown as afunction of the residual (the difference in the observed LOS and expected LOS) for each fracture type. The result was computed using the MLP classifier to adjust for the effect of covariates and to obtain a smoothed result. For extracapsular fractures, a shorter than expected LOS was associated with aslightly higher proportion of successful discharges, while alonger than expected LOS resulted in alower proportion of successful discharges. In other words, there was no indication that shorter inpatient stay would lead to worse results among the patients with extracapsular fractures. The results for intracapsular fractures demonstrated that somewhat shorter than expected LOS was associated with ahigher proportion of successful discharges, indicating that the available registerbased data had not enabled the model to identify some early-discharge patients with aclearly better prognosis. The simulated effects of the covariates on the probability of an early discharge are presented in Fig. with corresponding 95% credible intervals. It seemed that patients with more frailties required longer LOS. DISCUSSION In this study, the length of stay until discharge home was found to be significantly longer for the extracapsular fractures. Asimilar pattern has been detected in previous studies (9, 11, 19). The novel finding in this study was the markedly different shape of LOS distributions for intra- and extracapsular fractures. The LOS distribution for the intracapsular fractures was shown to have a simple unimodal shape, whereas the shape of the LOS distribution for extracapsular fractures was found to be multimodal with two clear peaks, suggesting (at least) two separate subpopulations. Determining the total length of stay in inpatient institutions in a reliable way for all hip fracture patients required a reconstruction of the treatment chains, made possible by the comprehensive Finnish register-based data. Although the registers do not contain all clinically relevant variables, and certain comorbidity prevalences are probably underestimated because of missed diagnoses or underreporting (6, 0), the validity of the register data for follow-up purposes has been demonstrated to be even better than with separate audit data collection (13). The anomaly in the shapes of LOS distribution would probably have been missed without the use of ahierarchical nonparametric Bayesian model. The actual reasons for the detected differences in LOS are largely unknown. They were not due to differences in background characteristics, as the groups were very similar in this respect and there were no differences in mortality figures. Because operation techniques vary between the types of hip fractures,

4 17 R. Sund, J. Riihimäki, M. Mäkelä, A. Vehtari, P. Lüthje, T. Huusko, U. Häkkinen TABLE 1 Characteristics of and outcomes for hip fracture population by hip fracture type Intracapsular hip fracture Extracapsular hip fracture Relative risk between groups (95% CI) Number of hip fracture patients aged 65 or more living at home at the time of fracture Background characteristics: Age (SD) 80.5 (7.1) 81.0 (7.3) 1.01 ( ) Female sex, % ( ) Admitted from home, % ( ) Weeks of inpatient care during year preceding fracture (SD) 0.5 (4.5) 0.6 (4.6) 1.01 ( ) Malignancy, % ( ) Chronic obstructive pulmonary disease, % ( ) Ischemic heart disease, % ( ) Previous myocardial infarction, % ( ) Congestive heart failure, % ( ) Cerebrovascular disease, % ( ) Peripheral vascular disease, % ( ) Diabetes, % ( ) Osteoarthritis, % ( ) Parkinson s disease, % ( ) Dementia, % ( ) Alcoholism, % ( ) Process variables: Operation type, % Internal fixation ( ) Hemiarthoplasty ( ) Total hip arthoplasty ( ) Other ( ) No operation ( ) Outcomes*: 30 day mortality, % ( ) 10 day mortality, % ( ) 365 day mortality, % ( ) Proportion of patients at home one month after fracture ( ) Proportion of patients at home 4 months after fracture ( ) Proportion of patients at home one year after fracture ( ) Discharged home from surgical ward, % ( ) Discharged home during rehabilitation period, % ( ) Becomes a long term care patient, % ( ) Died during treatment, % ( ) Mean week of discharge from surgical ward (SD) 01.7 (1.6) 01.7 (1.6) 1.01 ( ) Mean week of discharge from total inpatient stay (SD) 05. (3.4) 06.9 (4.) 1.3 ( ) Mean week of readmission to hospital after discharge home (SD) 14.3 (13.4) 13.8 (1.7) 1.00 ( ) Mean weeks alive and out of inpatient care during 60 days following discharge home (SD) Proportion of patients staying at home two weeks after discharge home, % 08.1 (1.9) 08.1 (.0) 1.00 ( ) ( ) *adjusted for covariates listed in the table these may have an influence on the LOS. In Finland, hemiarthoplasty has been the primary mode of treatment for intracapsular fractures, and osteosynthesis has been used mainly for younger patients (1). Extracapsular fractures have been typically treated with a sliding hip screw or gamma nail (). Total hip arthoplasty is used only for patients with osteoarthritis of the hip or rheumatoid arthritis. As the femoral head is removed and the femoral stem is typically cemented in hemiarthoplasty, there may be less pain in ambulation and full-weight bearing may be easier than for patients who underwent internal fixation. This interpretation is supported by the fact that patients with trochanteric fractures are known to expe-

5 Modeling LOS after hip fracture 173 A) B) week of discharge Fig. 1 A) Distributions of lengths of stay in inpatient care by fracture type B) Proportion of successful discharges to home presented as a function of residual with corresponding 95% credible intervals by fracture type residual: observed minus expected LOS (weeks) Fig.. Simulated average predictive comparisons (95% credible intervals) for the probability of discharge during the weeks 1 5, for background characteristics and process variables average change in early discharge probability Background characteristics Intracapsular fracture Five years increase in mean age Female sex Admitted from home One week increase in mean weeks of preceding care Malignancy Chronic obstructive pulmonary disease Ischemic heart disease Previous myocardial infarction Congestive heart failure Cerebrovascular disease Peripheral vascular disease Diabetes Osteoarthritis Parkinson s disease Dementia Alcoholism Process variables Operation type Internal fixation Hemiarthoplasty Total hip arthoplasty Other No operation

6 174 R. Sund, J. Riihimäki, M. Mäkelä, A. Vehtari, P. Lüthje, T. Huusko, U. Häkkinen rience more pain on walking during the first couple of postoperative weeks (3). Moreover, osteosynthesis was found to be a significant predictor for a longer LOS in this study. However, the multimodal shape of the LOS distribution for extracapsular fractures was unexpected, because similar postoperative rehabilitation practices, including immediate weight-bearing (7, 8), has been promoted in Finland for all patients regardless of the fracture type or operation performed (4). Certain subtrochanteric fractures are an exception, while multifragmented trochanteric fractures may sometimes result in an unstable osteosynthesis that makes immediate weight-bearing more difficult (19); it is unlikely however that these special cases would explain the multimodality. On the other hand, most hip fracture patients are referred to rehabilitation units or to a local health centre hospital for rehabilitation immediately after operative treatment in the surgical ward (5), and different local service structures and resources lead to variations in rehabilitation practices (). Amore plausible explanation for the multimodality is therefore that hospitals may be using outdated practices, such as instructing most of the patients with extracapsular fractures to start rehabilitation with partial weight-bearing. In fact, the examination of residuals in this study suggested that no harm would arise from shortening the LOS of patients with extracapsular fractures, and actually a somewhat shorter than expected LOS might result in better success of discharge to home, as was the case with the intracapsular fractures. In conclusion, the results confirmed that patients with extracapsular fractures require more health system resources during the hip fracture treatment process, since the rehabilitation is significantly slower than in the intracapsular group. As theredonot seem to be any obvious adverse effects in shortening the LOS for patients with extracapsular fractures, and as the recent Finnish clinical guideline suggests similar rehabilitation for all hip fracture patients (4), early and aggressive rehabilitation of patients with extracapsular fractures is needed, including full-weight bearing for all but selected patients. ACKNOWLEDGEMENTS The work of Jaakko Riihimäki was supported by Finnish Funding Agency for Technology and Innovation (project TERANA). The work of Aki Vehtari was supported by Academy of Finland (Academy Research Fellowship). The funding body played no role in the formulation of the design, methods, data analysis, or the preparation of this paper. REFERENCES 01. Marks R, Allegrante JP, Ronald MacKenzie C et al: Hip fractures among the elderly: causes, consequences and control. Ageing Res Rev 003;: Holmberg S,Thorngren KG: Rehabilitation after femoral neck fracture patients followed for 6 years. Acta Orthop Scand 1985;56: Nurmi I, Narinen A, Lüthje Pet al: Cost analysis of hip fracture treatment among the elderly for the public health services: a 1-year prospective study in 106 consecutive patients. Arch Orthop Trauma Surg 003;13: Huusko TM, Karppi P, Avikainen V et al: Randomised, clinically controlled trial of intensive geriatric rehabilitation in patients with hip fracture: subgroup analysis of patients with dementia. Bmj 000;31: Rolland Y, Pillard F, Lauwers-Cances V et al: Rehabilitation outcome of elderly patients with hip fracture and cognitive impairment. Disabil Rehabil 004;6: Huusko TM, Karppi P, Avikainen V et al: Intensive geriatric rehabilitation of hip fracture patients: a randomized, controlled trial. Acta Orthop Scand 00;73: Raivio M, Korkala O, Pitkälä K et al: Rehabilitation outcome in hip-fracture: Impact of weight-bearing restriction Apreliminary investigation. Phys Occup Ther Geriatr 005;: Heinonen M, Karppi P, Huusko Tetal: Post-operative degree of mobilization at two weeks predicts one-year mortality after hip fracture. Aging Clin Exp Res 004;16: Parker MJ, Pryor GA, Anand JK et al: Acomparison of presenting characteristics of patients with intracapsular and extracapsular proximal femoral fractures. J R Soc Med 199;85: Cornwall R, Gilbert MS, Koval KJ et al: Functional outcomes and mortality vary among different types of hip fractures: a function of patient characteristics. Clin Orthop Relat Res 004: Thorngren KG, Norrman PO, Hommel Aetal: Influence of age, sex, fracture type and pre-fracture living on rehabilitation pattern after hip fracture in the elderly. Disabil Rehabil 005;7: Haentjens P, Autier P, Barette Metal: Survival and functional outcome according to hip fracture type: Aone-year prospective cohort study in elderly women with an intertrochanteric or femoral neck fracture. Bone 007;41: Sund R, Nurmi-Lüthje I, Lüthje Petal: Comparing properties of audit data and routinely collected register data in case of performance assessment of hip fracture treatment in Finland. Methods Inf Med 007;46: Sund R, Liski A: Quality effects of operative delay on mortality in hip fracture treatment. Qual Saf Health Care 005;14: Heikkinen T, Jalovaara P: Four or twelve months follow-up in the evaluation of functional outcome after hip fracture surgery? Scand J Surg 005;94: Lampinen J, Vehtari A: Bayesian approach for neural networks review and case studies. Neural Networks 001;14: Neal RM: Bayesian Learning for Neural Networks. Volume 118 of Lecture Notes in Statistics. Springer, New York, Gelman A, Hill J: Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge University Press, Cambridge, Jarnlo GB, Ceder L, Thorngren KG: Early rehabilitation at home of elderly patients with hip fractures and consumption of resources in primary care. Scand JPrim Health Care 1984; : Laurila JV, Pitkälä KH, Strandberg TEetal: Detection and documentation of dementia and delirium in acute geriatric wards. Gen Hosp Psychiatry 004;6: Heikkinen T, Parker M, Jalovaara P: Hip fractures in Finland and Great Britain--a comparison of patient characteristics and outcomes. Int Orthop 001;5: Heikkinen T, Willig R, Hänninen Aet al: Hip fractures in Finland--a comparison of patient characteristics and outcomes in six hospitals. Scand J Surg 004;93: Thorngren KG: Optimal treatment of hip fractures. Acta Orthop Scand Suppl 1991;41: Duodecim Study Group, Finnish Orthopedic Society: [Care of patients with hip fractures]. Duodecim 006;1: Huusko T, Karppi P, Avikainen Vetal: Significant changes in the surgical methods and length of hospital stay of hip fracture patients occurring over 10 years in Central Finland. Ann Chir Gynaecol 1999;88:55 60 Received: May 30, 008 Accepted: March 5, 009