Magnetic Resonance Imaging in Acute Hamstring Injury: Can We Provide a Return to Play Prognosis?

Transcription

1 Sports Med (2015) 45: DOI /s SYSTEMATIC REVIEW Magnetic Resonance Imaging in Acute Hamstring Injury: Can We Provide a Return to Play Prognosis? Gustaaf Reurink Elisabet G. Brilman Robert-Jan de Vos Mario Maas Maarten H. Moen Adam Weir Gert Jan Goudswaard Joannes L. Tol Publised online: 15 August 2014 Ó Springer International Publising Switzerland 2014 Abstract Background Sports pysicians are increasingly requested to perform magnetic resonance imaging (MRI) of acute amstring muscle injuries and to provide a prognosis of te time to return to play (RTP) on te basis of teir findings. Objectives To systematically review te literature on te prognostic value of MRI findings for time to RTP in acute amstring muscle injuries. Data Sources Te databases of PubMed, EMBASE, CI- NAHL, Web of Science and Cocrane Library were searced in June Study Eligibility Criteria Studies evaluating MRI as a prognostic tool for determining time to RTP in atletes wit acute amstring injuries were eligible for inclusion. Data Analysis Two autors independently screened te searc results and assessed risk of bias using criteria for G. Reurink (&) E. G. Brilman Department of Ortopedics, Erasmus Medical Centre, Room Hs-104, PO Box 2040, 3000 CA Rotterdam, Te Neterlands g.reurink@erasmusmc.nl R.-J. de Vos Department of Sports Medicine, Medical Centre Te Hague, Antoniusove, Leidscendam, Te Neterlands M. Maas Department of Radiology, Academic Medical Centre University of Amsterdam, Amsterdam, Te Neterlands M. H. Moen Department of Sports Medicine, Bergman Clinics, Naarden, Te Neterlands M. H. Moen Te Sports Pysician Group, Amsterdam, Te Neterlands A. Weir G. J. Goudswaard J. L. Tol Department of Sports Medicine, Aspetar Ortopedics and Sports Medicine Hospital, Doa, Qatar quality appraisal of prognosis studies. A best-evidence syntesis was used to identify te level of evidence. Results Of te 12 studies included, one ad a low risk of bias and 11 a ig risk of bias. Tere is moderate evidence tat injuries witout yperintensity on fluid-sensitive sequences are associated wit a sorter time to RTP and tat injuries involving te proximal free tendon are associated wit a longer time to RTP. Limited evidence was found for an association of central tendon disruption, injury not affecting te musculotendinous junction and a total rupture wit a longer time to RTP. Te oter MRI findings studied sowed eiter no association or tere was conflicting evidence. Conclusion Tere is currently no strong evidence for any MRI finding tat gives a prognosis on te time to RTP after an acute amstring injury, owing to considerable risks of bias in te studies on tis topic. Key Points Based on te current evidence, tere is no strong evidence for any magnetic resonance imaging (MRI) finding tat can guide sports pysicians and radiologists in predicting prognosis for te time to return to play (RTP) after an acute amstring injury. Tere is considerable risk of bias in te majority of current studies on te prognostic value of MRI for te time to RTP in acute amstring injuries. Tere is moderate evidence tat injuries witout yperintensity on fluid-sensitive sequences are associated wit a sorter time to RTP and tat injuries involving te proximal free tendon are associated wit a longer time to RTP.

2 134 G. Reurink et al. 1 Background Hamstring injuries are te most prevalent time-loss injuries in major sports suc as American and Australian football, soccer and track and field atletics [1 5]. After injury, te main question of te atlete, coacing staff and press is: wen can e or se return to play? Magnetic resonance imaging (MRI) is more readily available tan ever before and plays an increasing role in diagnosing and predicting prognosis in amstring muscle injuries, especially in te elite atlete [6, 7]. Sports pysicians and radiologists are increasingly asked to assess MRIs of tese injuries and to elp provide a prognosis in te time to return to play (RTP) on te basis of teir findings. In te last two decades, a number of studies ave been publised on te prognostic value of MRI in acute amstring injuries tat reported multiple findings as indicators for te time to RTP, but te large variation of te time to RTP from 1 day [3] all te way up to 104 weeks [8] makes estimating te prognosis a callenge. Te purpose of tis paper was to systematically review te literature on te prognostic value of MRI findings for time to RTP in acute amstring injuries. 2 Metods All reviewers involved in te literature searc, study selection, data extraction and risk of bias assessment were medical doctors wit at least 2 years of experience as a clinical researcer in sports medicine. 2.1 Literature Searc Te databases of PubMed, EMBASE, CINAHL, Web of Science and Cocrane Library were searced witout any time limits in June An overview of te complete electronic searc is sown in Table 1. Additional citation tracking was performed by manual screening of te reference lists of te eligible studies. 2.2 Study Selection Two reviewers independently assessed all records identified by te searc strategy. Studies were eligible if tey met te following criteria: subjects ad a clinical diagnosis of an acute amstring injury; MRI examination of te acute injury was performed; MRI findings as a prognostic tool for time to RTP were studied, injury time or time to return to pre-injury level were studied; te study ad to be an original report; full text of te article ad to be available; te article was written in Englis, Dutc or German. Te two reviewers read all relevant full text articles to assess weter tey met te eligibility criteria. If tere was a difference in opinion on eligibility, a consensus was reaced by te two reviewers. If no consensus was reaced, te independent opinion of a tird reviewer was decisive. 2.3 Data Extraction One reviewer recorded te population, details of te MRI protocol, MRI findings, time to RTP and te outcome of te analysis of association between MRI findings and time to RTP using standardised data extraction forms. Autors of te eligible studies were contacted if additional information was required. 2.4 Risk of Bias Assessment Two reviewers independently assessed te potential risk of bias of te studies included, using te criteria of te consensus statement of Hayden et al. [9]. Tis risk of bias assessment tool assesses six potential bias domains, eac consisting of specific items for opportunity of bias (Table 2). If tere was a difference in opinion on an item, a consensus was reaced by te two reviewers. If no consensus was reaced, te independent opinion of a tird reviewer was decisive. As sown in Table 2, eac of te six potential bias domains consists of tree to five specific items. Wen C75 % of tese items witin a domain was fulfilled, we considered te bias low in tat domain. To ave overall low risk of bias, a study sould ave low bias on: 1. At least five out of te six domains. and 2. Te outcome measurement time to RTP (domain 4). 2.5 Best-Evidence Syntesis Because of te eterogeneity of te MRI findings, outcome measures and metodological quality, we refrained from statistical pooling of te data. We used a best-evidence syntesis, consisting of a five levels of evidence based qualitative analysis [10]: 1. Strong evidence: provided by two or more studies wit low risk of bias and by generally consistent findings in all studies (C75 % of te studies reported consistent findings). 2. Moderate evidence: provided by one study wit low risk of bias and/or two or more studies wit ig risk of bias and by generally consistent findings in all

3 MRI for Return to Play Prognosis in Acute Hamstring Injury 135 Table 1 Searc strategy a Searc strategy Records PubMed 246 ((amstring*[tiab])) AND ( magnetic resonance imaging [m] OR diagnostic imaging[m:noexp] OR (magnetic resonance imaging[tiab] OR mri[tiab] OR radiodiagnos*[tiab] OR imaging[tiab])) AND ( Wounds and Injuries [m] OR (injur*[tiab] OR tear*[tiab] OR strain*[tiab] OR rupture*[tiab] OR trauma*[tiab])) NOT (animals[m] NOT umans[m]) Embase 415 (amstring/exp OR (amstring*):ab,ti) AND ( nuclear magnetic resonance imaging /exp OR radiodiagnosis/de OR diagnostic imaging /de OR ( magnetic resonance imaging OR mri OR radiodiagnos* OR imaging):ab,ti) AND (injury/exp OR (injur* OR tear* OR strain* OR rupture* OR trauma*):ab,ti) NOT ([animals]/lim NOT [umans]/lim) Cocrane central 7 ((amstring*):ab,ti) AND (( magnetic resonance imaging OR mri OR radiodiagnos* OR imaging):ab,ti) AND ((injur* OR tear* OR strain* OR rupture* OR trauma*):ab,ti) Web of science 224 TS = (((amstring*)) AND ((magnetic resonance imaging OR mri OR radiodiagnos* OR imaging)) AND ((injur* OR tear* OR strain* OR rupture* OR trauma*)) NOT ((animal* OR mouse OR mice OR rat? OR nonuman OR dog? OR rabbit? OR cicken? OR swine? OR cat? OR rodent?) NOT (uman* OR patient*))) CINAHL 177 TX ((amstring*)) AND (MH magnetic resonance imaging? OR MH diagnostic imaging? OR TX (magnetic resonance imaging OR mri OR radiodiagnos* OR imaging)) AND (Wounds and Injuries? OR TX (injur* OR tear* OR strain* OR rupture* OR trauma*)) NOT (MH animals? NOT MH Humans) a Searc performed June 2013 studies (C75 % of te studies reported consistent findings). 3. Limited evidence: provided by only one study wit ig risk of bias. 4. Conflicting evidence: inconsistent findings in multiple studies (\75 % of te studies reported consistent findings). 5. No evidence: wen no studies could be found. 3 Results 3.1 Literature Searc Figure 1 sows te study selection flow diagram; 12 studies met te inclusion criteria [11 22]. 3.2 Description of Included Studies Table 3 presents te caracteristics of te studies included [11 22]. Two reports [15, 16] used te same data set and are terefore considered as one study (confirmed by te corresponding autor). Table 4 presents an overview of te MRI protocols used in te studies included. 3.3 Risk of Bias Assessment Te scores on te potential risk of bias domains of te studies included are sown in Table 5. One study ad a low risk of bias [20] and 11 studies ad a ig risk of bias [11 19, 21, 22]. Te detailed score seets for eac individual study are presented in Electronic Supplementary Material Appendix S1. Tere was 100 % agreement between te two reviewers on te classification of te studies into ig or low risk of bias. For te specific items for opportunity of bias tere was disagreement on 18 out of te 264 assessed items (6.8 %), for wic consensus was reaced by te two reviewers. 3.4 MRI Finding and Association wit Time to Return to Play Table 6 presents an overview of all te reported MRI findings, teir association wit RTP and te corresponding level of evidence according to te best-evidence syntesis MRI Negative Injury Moderate evidence was found tat te absence of any focal yperintensity on fluid-sensitive sequences (MRI negative injury) is associated wit a reduced time to RTP. Six studies sowed tat MRI-negative injuries ad a significantly sorter time to RTP tan MRI-positive injuries [13, 15 18, 21, 22] Number of Muscles Involved Tere is conflicting evidence for te association of te number of muscles injured and time to RTP, as tere were

4 136 G. Reurink et al. Table 2 Risk of bias assessment tool Potential bias domain Items for assessment of potential opportunity for bias Yes No/not reported 1. Study participation Te study sample represents te population of interest on key caracteristics, sufficient to limit potential bias of te results Yes No 2. Study attrition Loss to follow-up is not associated wit key caracteristics (i.e. te study data adequately represent te sample), sufficient to limit potential bias Yes No 3. Prognostic factor measurement Te prognostic factor of interest is adequately measured in study participants to sufficiently limit potential bias Yes No 4. Outcome measurement Te outcome of interest is adequately measured in study participants to sufficiently limit potential bias Yes No 5. Confounding measurement and account Important potential confounders are appropriately accounted for, limiting potential bias wit respect to te prognostic factor of interest Yes No Te source population is adequately described for te key caracteristics of type and level of sport Inclusion and exclusion criteria are adequately described Tere is adequate participation in te study by eligible individuals Te baseline study sample is adequately described for te key caracteristics: sex, age, type and level of sport Losses to follow-up are reported Reasons for loss to follow-up are described Loss to follow-up is less tan 20 % To assess wen lost to follow-up is more tan 20 %: tere are no important differences between key caracteristics and te prognostic MRI measures in participants wo completed te study and tose wo did not Te prognostic MRI measures are adequately defined or described Te prognostic MRI measures and metods are adequately valid and reliable to limit misclassification bias (may refer to relevant outside sources of information on measurement properties) Te prognostic MRI measures are blinded for te outcome measure time to RTP or injury time More tan 80 % of te study sample as complete data for te prognostic MRI measures Te outcome measure time to RTP or injury time is adequately defined or described Criteria for RTP clearance or recovery of te injury are clearly described and are te same for all study participants Te clinicians/terapists involved in te reabilitation and/or RTP decision and te subjects are blinded to te prognostic MRI measure Important potential confounders: Type of sport or injury mecanism Index injury being a re-injury (assessed at least for te 2 previous monts) Reabilitation protocol

5 MRI for Return to Play Prognosis in Acute Hamstring Injury 137 Table 2 continued Potential bias domain Items for assessment of potential opportunity for bias Yes No/not reported 6. Analysis Te statistical analysis is appropriate for te design of te study, limiting potential for presentation of invalid results Yes No Te important potential confounders measured are adequately defined or described All important potential confounders are measured and accounted for in te study design or analysis Tere is a description of te association of te prognostic MRI measure and te outcome measure time to RTP or injury time, including information about te statistical significance Te selected model is adequate for te design of te study To assess wen multivariate models are used: te strategy of inclusion of variables is appropriate and is based on a conceptual framework or model Tere is no selective reporting of results Continuous variables are reported or appropriate (i.e. not data dependent) cut-off points are used To score yes on a potential bias item at least 75 % of te assessment criteria sould be scored yes RTP return to play, MRI magnetic resonance imaging no consistent findings in te two studies reporting tis finding. Silder et al. [20] reported tat a iger number of muscles injured was significantly correlated wit a longer time to RTP (r = 0.50, p = 0.010). Gibbs et al. [18] reported no difference in te time to RTP between te atletes wit a single muscle and tose wit two muscles injured (p = 0.73) Muscle Involved Moderate evidence was found tat tere is no association between involvement of te different amstring muscles and time to RTP. Connell et al. and Scneider-Kolsky et al. [15, 16] reported tat an injury of te biceps femoris was associated wit longer time to RTP (p = 0.049). Tree studies reported no difference between time to RTP and involvement of te different amstring muscles (p = ) [14, 17, 21] Distance of Injury to te Muscle Origin Conflicting evidence was found tat te distance of te injury to te muscle origin is associated wit time to RTP. Different metods were used to measure te distance of te injury to te muscle origin. Four studies measured te distance between te iscial tuberosity and te most cranial point of te yperintensity [11 13] or te maximum yperintensity [20]. Tree of tese studies reported a significant association of te distance to te iscial tuberosity Fig. 1 Study selection flow diagram. MRI magnetic resonance imaging (r = , p = ) wit a longer time to RTP [11, 13, 20] and one found no association (p [ 0.05) [12]. Slavotinek et al. assessed weter te injury was observed proximal or distal in te amstring, wit using te

6 138 G. Reurink et al. Table 3 Caracteristics of included studies References Population, timing of MRI examination and RTP caracteristics Askling et al. [11] n = 17 Sprinters, national or international; Timing MRI: 4 d after injury; RTP: train or compete at pre-injury level; Time to RTP: median 16 w (range 6 50) Askling et al. [12] b n = 12 Dancers, professional Timing MRI: 4 d after injury; RTP: train or compete at pre-injury level Time to RTP: median 50 w (range 30 76) Askling et al. [13] n = 75 Football, professional; Timing MRI: B5 d after injury; RTP: full participation in team training and availability for matc selection; Time to RTP: L-protocol mean 28 d (±15) C-protocol mean 51 d (±21) Comin et al. [14] n = 62 Australian Football and Rugby, national; Timing MRI: NA; RTP: return to competition; Time to RTP: median 21 d (IQR 14 42) MRI finding Association wit time to RTP Significant association a Non-significant association Proximal free tendon involvement Involved 34.8 w vs. not involved 13 w (p = 0.009) r = (p = 0.044) Distance to origin Proximal 25.6 w vs. distal 9.5 w (p = 0.028) Longitudinal lengt r = (p = 0.055) Cross-sectional area r = (p = 0.004) Volume r = (p = 0.016) Antero-posterior extent r = (p = 0.022) Medio-lateral extent r = (p = 0.146) Involved muscle site r = (p = ) for Distance to origin all MRI findings Longitudinal lengt Cross-sectional area Volume Antero-posterior extent Medio-lateral extent Hyperintensity Absence 6 d (±3) vs. presence 23 d (±11) (p \ 0.001) c Proximal free tendon L-protocol: involved [ not involved (p \ 0.01) involvement C-protocol: involved [ not involved (p \ 0.001) Distance to origin L-protocol: r = (p \ 0.001) C-protocol: r = (p \ 0.001) Longitudinal lengt L-protocol: r = (p \ 0.001) C-protocol: r = (p \ 0.001) Muscle injured BF 21 d (IQR 12 56) vs. SM 32 d (IQR 21 35) vs. ST 14 d (IQR 12 22) (p = 0.33) Central tendon disruption Disruption 72 d (IQR ) vs. no disruption 21 d (IQR 9 28) (p \ 0.01)

7 MRI for Return to Play Prognosis in Acute Hamstring Injury 139 Table 3 continued References Population, timing of MRI examination and RTP caracteristics Connell et al. [15] and Scneider-Kolsky et al. [16] d n = 58 Australian Football, professional; Timing MRI: B3 d after injury; RTP: return to competition; Time to RTP: median 21 d (IQR 4 56) Ekstrand et al. [17] n = 207 Football, professional; Timing MRI: 1 2 d after injury; RTP: clearance medical team for full training participation and matc selection; Time to RTP: mean 19 d (±17) Gibbs et al. [18] n = 31 Australian Football, professional; Timing MRI: 1 3 d after injury; RTP: full participation in team training; Time to RTP: median 18 d (IQR 14 27) Rettig et al. [19] e n = 21 American football, professional; Timing MRI: NA; RTP: NA; Time to RTP: NA Silder et al. [20] n = 25 Sports requiring ig-speed running; Timing MRI: B10 d after injury; RTP: Completion of reabilitation; Time to RTP: median 23 d (IQR 20 23) MRI finding Association wit time to RTP Significant association a Non-significant association Hyperintensity Absence 7 d (IQR 7 14) vs. presence 21 d (IQR 4 56) (p \ 0.001) Location witin muscle Musculotendinous junction involved \ not involved r = NA (p \ 0.05) Cross-sectional area r = NA (p \ 0.05) Intramuscular fluid collection Extramuscular fluid collection Multivariate: r 2 = 37.9 % Longitudinal lengt (p = 0.001) Muscle injured BF involved (p = 0.049) Hyperintensity Absence (=grade 0) vs. presence (=grade 1 3) (p \ 0.05) Grading Grade 0 8 d (±3), grade 1 17 d (±10), grade 2 22 d (±11), grade 3 73 d (±60) (p \ 0.001) Pairwise comparison (p \ 0.05), except for grade 1 vs. grade 2 r=na (p [ 0.05) r = NA (p [ 0.05) Grade 1 vs. grade 2 (p = 0.053) Muscle injured BF 21 d (±19) vs. SM 19 d (±11) vs. ST 17 d (±11) (p = 0.79) Hyperintensity Absence 6.6 d (±8.2) vs. presence 20.2 d (±52.3) (p \ 0.001) Longitudinal lengt r = 0.84 (p \ 0.001) Cross-sectional area r = 0.78 (p \ 0.001) Number of muscles Single vs. double (p = 0.73) Longitudinal lengt Tendon separation at musculotendinous junction Distance to origin r =-0.44 (p = 0.043) f Longitudinal lengt r = 0.41 (p = 0.040) Cross-sectional area r = 0.30 (p = 0.182) f Number of muscles r = 0.50 (p = 0.010) f

8 140 G. Reurink et al. Table 3 continued References Population, timing of MRI examination and RTP caracteristics MRI finding Association wit time to RTP Significant association a Non-significant association Slavotinek et al. [21] n = 30 Australian Football, national or state; Timing MRI: 2 6 d after injury; RTP: return to competition; Time to RTP: median 27 d (range 13 48) Verrall et al. [22] n = 83 Australian Football, National or state; Timing MRI: 2 6 d after injury; RTP: return in competition; Time to RTP: NA Hyperintensity Absence vs. presence: c =-0.69 (p = 0.04) Muscles injured BF vs. ST (p = 0.86) Distance to origin Proximal vs. distal (p = 0.17) Cross-sectional area r = 0.63 (p \ 0.001) Volume r = 0.46 (p = 0.01) Extramuscular fluid r = 0.33 (p = 0.12) collection Hyperintensity Absence 16 d vs. presence 27 d (p \ 0.01) RTP return to play, MRI magnetic resonance imaging, w week, d days, IQR interquartile range, BF biceps femoris, SM semimembranosus, ST semitendinosus, NA not available Presented association measures are tested univariate, unless oterwise specified, p \ 0.05 is considered statistical significant, r correlation coefficient, c gamma statistics, r 2 variance in time to RTP explained by te multivariate model Contact wit corresponding autor: no data available for correlation of eac MRI finding and RTP separately As part of a randomised controlled trial, prognostic MRI variables assessed separately for treatment groups. L-protocol lengtening exercises, C-protocol conventional exercises a b c d Studies of Connell et al. [15] and Scneider-Kolsky et al. [16] used te same dataset and are terefore considered one study (confirmed by te corresponding autor) No statistical testing reported e Association not determined in te original reports. A reviewer (GR) analysed te association using te data presented in te report f

9 MRI for Return to Play Prognosis in Acute Hamstring Injury 141 Table 4 Magnetic resonance imaging protocols used References Sides scanned Magnetic field strengt Coil Sequences TR/TE TI, ETL, flip angle Tickness sections/gap (mm) Field of view (cm) Matrix (pixels) Askling et al. [11] Bilateral 1.0 T [11, 12] Pasedarray Coronal STIR 4,000/30 TI 150 ms 5/ [13] and 1.5 T [13] spine Sagittal STIR 4,000/30 TI 150 ms 5/ Axial STIR 5,035/30 TI 150 ms 5/ Axial T1 722/20 5/ Axial T2 5,500/110 5/ Comin et al. [14] NR NR NR Proton density fat saturation NR NR NR NR NR Connell et al. [15] and Scneider- Kolsky et al. [16] NR 1.5 T Pasedarray soulder Ekstrand et al. [17] NR Minimum required: 1.5 T Axial and coronal oblique fast spin-eco Axial and coronal oblique fast spin-eco IR NR Minimum required: Axial and coronal T1 and T2 fat saturation or STIR Gibbs et al. [18] Bilateral 1.5 T NR Coronal T1 Coronal STIR Axial T2 fat suppression 4,000/45 5,000 6,500 /35 55 ETL /0 5/0 NR NR NR NR NR NR NR 4/1.5 10/0 7/ NR NR Rettig et al. [19] Bilateral NR NR NR NR NR NR NR Silder et al. [20] Bilateral 1.5 T Pasedarray torso Slavotinek et al. [21] NR 1.5 T Polarized body array Axial T2 Coronal T2 Axial T1 Axial IR T2 Sagittal T1 Sagittal IR T2 Axial gradient-eco Verrall et al. [22] NR 1.5 T NR Axial and sagittal T1, T2 and gradient eco 2,200 3,200 / /12 5,032/30 676/12 5,000/30 610/18 NR 5/0 4/0.4 ETL 3 ETL 7; TI 150 ms ETL 3 ETL 7; TI 150 ms Flip angle 20 10/2 10/2 7/1.4 7/1.4 10/ NR NR NR NR NR NR TR time to repetition, TE time to eco, TI time to inversion, ETL eco train lengt, T tesla, STIR sort tau inversion recovery, NR not reported, IR inversion recovery

10 142 G. Reurink et al. Table 5 Risk of bias assessment References femoral origin of te sort ead of te biceps femoris as a reference point. Tey reported no difference in time to RTP between proximally and distally located injuries (p = 0.17) [21] Proximal Free Tendon Involvement Moderate evidence was found for an association between involvement of te proximal tendon and time to RTP. Two studies reported tat time to RTP was significantly longer in injuries wit proximal tendon involvement tan witout (p \ 0.01) [11, 13]. Te proximal free tendon was considered injured if it was tickened, ad an intratendinous ig signal or a collar of ig signal around it on a fluidsensitive sequence Central Tendon Disruption Potential risk of bias item Askling et al. [11]??? -?? Hig Askling et al. [12]? - - -?? Hig Askling et al. [13]?? - -?? Hig Comin et al. [14] - -? - -? Hig Connell et al. [15]??? - -? Hig Scneider-Kolsky et al.? ? Hig [16] Ekstrand et al. [17] ? Hig Gibbs et al. [18]? ? Hig Rettig et al. [19] Hig Silder et al. [20]???? -? Low Slavotinek et al. [21] - -? ±?? Hig Verrall et al. [22]? ? Hig Risk of bias a?, potential risk of bias limited sufficiently; -, potential risk of bias; ±, potential risk of bias limited sufficiently, except for te finding yperintensity absence or presence a Low risk of bias requires positive scores for a minimum five out of six items and for item 4 Limited evidence was found tat involvement of te central tendon is associated wit a longer time to RTP. Comin et al. [14] reported tat injuries wit MRI findings of central tendon disruption, determined by te presence of a focal defect separating proximal and distal parts of te tendon or waviness of te tendon, ad significantly longer time to RTP tan tose injuries witout tese findings (p \ 0.01) Musculotendinous Junction Involvement Limited evidence was found tat injuries not affecting te musculotendinous junction are associated wit a longer time to RTP. Connell et al. [15] reported tat injuries at te musculotendinous junction ad a significant longer recovery time tan tose tat did not affect te musculotendinous junction (p \ 0.05) Longitudinal Lengt Conflicting evidence was found for an association between te longitudinal lengt of yperintensity on fluid-sensitive sequences and te time to RTP. In an univariate analysis, a larger longitudinal lengt was sown to be associated wit a longer time to RTP in tree studies (r = , p = ) [13, 18, 20]. No association was found in two studies (r = 0.51, p [ 0.05) [11, 12]. In a multivariate analysis, te longitudinal lengt was found to be independently associated wit time to RTP (p = 0.001) [15, 16] Cross-Sectional Area Conflicting evidence was found for an association of te cross-sectional area of yperintensity on fluid-sensitive sequences wit time to RTP. All studies used a similar definition of te cross-sectional area: te maximal muscle cross-sectional area of yperintensity expressed as a percentage of te total cross-sectional muscle area on te same axial image, measured on a fluid-sensitive sequence. Four studies [11, 15, 16, 18, 21] reported a significant association wit a longer time to RTP (r = , p = ) and two studies [12, 20] found no association wit te time to RTP (r = 0.30, p = and p [ 0.05) Volume Conflicting evidence was found for an association between te volume of te yperintensity on fluid-sensitive sequences and time to RTP. Te volume in all tree studies was calculated using te formula of an ellipsoid (volume & cranio-caudal 9 antero-posterior 9 mediolateral 9 0.5) [11, 12, 21]. Two studies reported an association of a larger volume wit a longer time to RTP (r = , p = ) [11, 21]. In a coort of dancers Askling et al. [12] found no significant correlation between te volume of te yperintensity on fluid-sensitive sequences and time to return to pre-injury level (p [ 0.05).

11 MRI for Return to Play Prognosis in Acute Hamstring Injury 143 Table 6 Overview of te studied MRI findings and teir association wit te time to return to play, and te corresponding level of evidence according to te best-evidence syntesis MRI finding Univariate Multivariate Best-evidence syntesis a Low risk of bias Hig risk of bias Hig risk of bias Association Level of evidence Hyperintensity absence -[13, 15 18, 21, 22] Yes Moderate Number of injured muscles?[20] =[18] Unknown Conflicting Location Muscle injured =[14, 17, 21]?[15, 16] No Moderate Distance to origin?[20]?[11, 13], =[12, 21] Unknown Conflicting Proximal free tendon involvement?[11, 13] Yes Moderate Central tendon disruption?[14] Yes Limited Musculotendinous junction involvement -[15, 16] Yes Limited Hyperintensity extent Longitudinal lengt?[20]?[13, 18], = [11, 12]?[15, 16] Unknown Conflicting Cross-sectional area =[20]?[11, 15, 16, 18, 21], =[12] Unknown Conflicting Volume?[11, 21], =[12] Unknown Conflicting Antero-posterior (dept)?[11], =[12] Unknown Conflicting Medio-lateral (widt) = [11], =[12] No Moderate Fluid collection Intramuscular =[15, 16] No Limited Extramuscular =[15, 16, 21] No Moderate Grading Grade 0 3?[17] Yes Limited Grade 1 vs. grade 2 =[17] No Limited MRI magnetic resonance imaging -, association wit sorter time to return to play (negative association);?, association wit longer time to return to play (positive association); =, no association wit time to return to play a Te studies of Connell et al. [15] and Scneider-Kolsky et al. [16] used te same dataset and are terefore considered as one study in te bestevidence syntesis Medio-Lateral Extent Moderate evidence was found tat tere is no association between te maximal medio-lateral extent of yperintensity measured on te axial images of fluid sensitive sequences and time to RTP, as Askling et al. [11, 12] found no significant correlations in bot coorts of sprinters and dancers (r = 0.40, p = and p [ 0.05) Antero-Posterior Extent Conflicting evidence was found for an association between te antero-posterior extent of yperintensity measured on te axial images of fluid-sensitive sequences and time back to pre-injury level. A study by Askling et al. [11] in sprinters sowed an association between a larger antero-posterior extent and a longer time to RTP (r = 0.58, p = 0.022). Conversely, a study of Askling et al. [12] in dancers sowed no significant association (p [ 0.05) Fluid Collection Tere is moderate and limited evidence tat extramuscular and intramuscular fluid collections respectively seen on MRI, suggestive for ematoma, are not associated wit te time to RTP (r = 0.33, p = 0.12 and p [ 0.05) [15, 16, 21]. Connell et al. and Scneider-Kolsky et al. [15, 16] defined ematoma as a collection of fluid wit abnormal signal intensity. Slavotinek et al. [21] considered extramuscular T2 yperintensity to be extramuscular fluid Grading Grading was studied in one report [17], using te following classification: (grade 0) negative MRI witout any visible patology, (grade 1) yperintensity on fluid-sensitive sequences witout evidence of a macroscopic tear, (grade 2) yperintensity on fluid-sensitive sequences wit a partial tear, (grade 3) total muscle or tendon rupture. Pairwise comparison sowed tat tere was a significant difference

12 144 G. Reurink et al. in time to RTP between te grades of injury (p \ 0.001), except between grade 1 and 2 injuries (p = 0.053). Tis implies tat tere is limited evidence for an association wit te time to RTP and a grading tat differentiates between: (i) MRI-negative injuries, (ii) MRI-positive injuries witout a total muscle or tendon rupture, and (iii) injuries wit a total muscle or tendon rupture. 4 Discussion Te major findings of our systematic review are tat tere is moderate evidence tat te absence of any yperintensity on fluid-sensitive sequences is associated wit a sorter time to RTP and tat proximal free tendon involvement is associated wit a longer time to RTP. Tere is currently no strong evidence for any MRI finding tat can guide sports pysicians and radiologists in predicting te prognosis for te time to RTP after an acute amstring injury, as only one of te 12 studies included ad a low risk of bias. In te current clinical setting, MRI is considered as a valuable tool in atletes wit amstring injuries and tere are ig demands from te atletes and teir medical staff to provide a prognosis on recovery time based on te MRI findings. However, our review sows tat te sports pysicians and radiologists cannot satisfy tese ig expectations from an evidence-based point of view. 4.1 Return to Play Te definition for RTP differed in te included studies: return to competition [14 16, 21, 22], return to full team training [18], full training participation and availability for matc selection [13, 17], performing at a pre-injury level [11, 12] and completion of reabilitation [20] Tese different definitions for time to RTP complicate comparison of te studies. RTP is generally accepted as te primary outcome measure for acute muscle injuries, as it is te most clinically relevant outcome in atletes wit tese injuries [23, 24]. However, tere are still no validated objective criteria to guide progression troug reabilitation protocols and assess readiness for RTP. Decision making for progression troug reabilitation protocols and clearance for RTP are terefore substantially affected by subjective judgements of atletes and medical staff involved. In te absence of well-defined RTP criteria, knowledge about te results of te MRI findings introduces a major potential source of bias. We terefore considered adequately measured time to RTP, by clearly defined RTP criteria and blinding of subjects and clinicians involved in te reabilitation or RTP decisions, compulsory for a low risk of biased results (domain 4 of te risk of bias assessment tool). Only two of te studies included reported blinding of te subjects and managing clinicians for te prognostic MRI findings studied [20, 21]. 4.2 Confounding Factors Te type of sport or injury mecanism, weter it was a new or recurrent injury and te treatment/reabilitation protocol were considered to be important potential confounders in te prognostic value of MRI findings for time to RTP tat sould be appropriately accounted for in te analysis or study design to sufficiently limit potentially biased results (domain 5 of te risk of bias assessment tool). Askling et al. [11, 12] reported more extensive MRI abnormalities and sorter time to RTP in sprinting atletes compared wit stretc-type injuries in dancers. As tis difference between sprinters and dancers may be caused by te type of sport or te injury mecanism, we considered te potential bias of tis confounder sufficiently limited if eiter te type of sports or te injury mecanism was accounted for. Re-injuries are a potential confounder, because tey are associated wit bot more extensive MRI abnormalities and a longer time to RTP [25]. To prevent confounding, te treatment/reabilitation protocol sould be te same in all studied subjects or appropriately accounted for in te analysis. In four of te studies, tese important potential confounders were appropriately accounted for [11 13, 21]. 4.3 Reliability of MRI Measures Only one study presented any information on te reliability of te performed MRI measures: Comin et al. [14] reported 100 % agreement between te two radiologists on te presence or absence of central tendon disruption. None of te oter studies included provided or referred to any information on reliability of te MRI measures and metods, introducing a risk of misclassification bias. 4.4 Limitations We performed a qualitative analysis (best-evidence syntesis) instead of a quantitative analysis (meta-analysis of te data), because of te eterogeneity of te studies wit regard to te MRI findings, reported outcome measures and metodological quality. Tis systematic review does not provide a quantitative syntesis on te strengt or magnitude of te associations of te MRI findings wit RTP, but is limited to weter tere is evidence for a statistical significant association or not. Tis limits te interpretation of te magnitude and clinical relevance of te reported

13 MRI for Return to Play Prognosis in Acute Hamstring Injury 145 associations. If we for example consider te difference between MRI-negative and MRI-positive injuries, studies reported a wide range of days to RTP: 6 (±3) vs. 23 (±11) [13], 7 (interquartile range 7 14) vs. 21 (interquartile range 4 56) [15, 16], 8 (±3) vs. 20 (±14) [17], 7 (±8) vs. 20 (±52) [18] and 16 vs. 27 [22] for MRI-negative versus MRI-positive injuries, respectively. One study reported gamma statistics, a measure of rank correlation, wit a correlation coefficient of 0.69 between MRI-positive injuries and time to RTP. Altoug tese different outcome measures cannot be appropriately pooled, a general overview of tese numbers and teir variability measures indicate tat a MRI-negative injury may take several days up to weeks to RTP and MRI-positive injuries may take several days up to monts to RTP. For systematic reviews on prognostic findings, tere is currently no standardized risk of bias assessment metod and tere are no generally accepted limits to determine weter a study as a ig or low risk of bias. Instead, it is recommended tat risk of bias criteria for prognostic studies sould be applied on te basis of te relevance to te researc question [9]. We used risk of bias criteria, wic we tougt te most appropriate for studies on te prognostic value of MRI findings in acute amstring injuries. Wit tis approac we aimed to perform a best available systematic risk of bias analysis. Te sample size of some of te studies included migt ave been insufficient to sow statistical significance of clinically important associations, potentially introducing a type II error. Te sample sizes of studies reporting no significant association on te number of muscles involved, distance to te muscle origin and te yperintensity extent measures varied between 12 and 31 [11, 12, 18, 20, 21], and are terefore unable to detect weak to moderate associations [26]. Wen te sample size is large enoug to sow statistical significance, te outcome of te best-evidence syntesis could cange from conflicting evidence to moderate evidence. However, large sample sizes can lead to statistical significant, but clinically irrelevant associations. Te majority of prognostic MRI findings are analysed wit simple univariate statistical approaces, wit only one of te studies using multivariate statistical analysis [16, 22]. In te absence of multivariate analysis, it remains unknown to wat extent te MRI findings are independently associated wit te time to RTP. Tis is because te majority can be expected to be related to eac oter, for example, larger longitudinal lengt is likely to ave a larger volume. 4.5 Future Directions Tis systematic review sowed a lack of ig-quality studies on te prognostic value of MRI in acute amstring injuries. Common metodological limitations are te low number of participants, insufficient information about losses to follow-up, lack of blinding of subjects and clinicians to te MRI results, insufficient accounting for potential confounders, lack of information on te reliability of MRI measures used, and te use of simple univariate statistical analysis. Future studies sould account for tese metodological flaws. Te use of different definitions for te time to RTP, as an outcome measure, limits te comparability of te studies. Consensus on te definition of RTP is required to improve te comparability of studies using RTP as an outcome measure. 5 Conclusion Tere is currently no strong evidence for any MRI finding tat can guide sports pysicians and radiologists in predicting prognosis for te time to RTP after an acute amstring injury, as only one of te 12 studies included ad a low risk of bias. Tere is only moderate evidence tat injuries witout yperintensity on fluid-sensitive sequences are associated wit a sorter time to RTP and tat injuries involving te proximal free tendon are associated wit a longer time to RTP. Limited evidence was found for an association between central tendon disruption, injury not affecting te musculotendinous junction and total amstring ruptures wit a longer time to RTP. Te oter MRI findings studied sowed eiter no association wit time to RTP or tere was conflicting evidence. Acknowledgments No sources of funding were used to assist in te preparation of tis review. Te autors ave no potential conflicts of interest tat are directly relevant to te content of tis review. References 1. Feeley BT, Kennelly S, Barnes RP, et al. Epidemiology of National Football League training camp injuries from 1998 to Am J Sports Med. 2008;36: Sankar PR, Fields SK, Collins CL, et al. Epidemiology of ig scool and collegiate football injuries in te United States, Am J Sports Med. 2007;35: Ekstrand J, Hägglund M, Waldén M. Epidemiology of muscle injuries in professional football (soccer). Am J Sports Med. 2011;39: Alonso J-M, Edouard P, Fiscetto G, et al. Determination of future prevention strategies in elite track and field: analysis of Daegu 2011 IAAF campionsips injuries and illnesses surveillance. Br J Sports Med. 2012;46: Orcard J, Seward H. Epidemiology of injuries in te Australian Football League, seasons Br J Sports Med. 2002;36: Kerkoffs GMMJ, van Es N, Wieldraaijer T, et al. Diagnosis and prognosis of acute amstring injuries in atletes. Knee Surg Sports Traumatol Artrosc. 2013;21:500 9.

14 146 G. Reurink et al. 7. Koulouris G, Connell D. Hamstring muscle complex: an imaging review. Radiograpics. 2005;25: Askling CM, Tengvar M, Saartok T, et al. Proximal amstring strains of stretcing type in different sports: injury situations, clinical and magnetic resonance imaging caracteristics, and return to sport. Am J Sports Med. 2008;36: Hayden JA, Côté P, Bombardier C. Evaluation of te quality of prognosis studies in systematic reviews. Ann Intern Med. 2006;144: Van Tulder M, Furlan A, Bombardier C, Editorial Board of te Cocrane Collaboration Back Review Group, et al. Updated metod guidelines for systematic reviews in te Cocrane collaboration back review group. Spine. 2003;28: Askling CM, Tengvar M, Saartok T, et al. Acute first-time amstring strains during ig-speed running: a longitudinal study including clinical and magnetic resonance imaging findings. Am J Sports Med. 2007;35: Askling CM, Tengvar M, Saartok T, et al. Acute first-time amstring strains during slow-speed stretcing: clinical, magnetic resonance imaging, and recovery caracteristics. Am J Sports Med. 2007;35: Askling CM, Tengvar M, Torstensson A. Acute amstring injuries in Swedis elite football: a prospective randomised controlled clinical trial comparing two reabilitation protocols. Br J Sports Med. 2013;47: Comin J, Malliaras P, Baquie P, et al. Return to competitive play after amstring injuries involving disruption of te central tendon. Am J Sports Med. 2013;41: Connell DA, Scneider-Kolsky ME, Hoving JL, et al. Longitudinal study comparing sonograpic and MRI assessments of acute and ealing amstring injuries. AJR Am J Roentgenol. 2004;183: Scneider-Kolsky ME, Hoving JL, Warren P, et al. A comparison between clinical assessment and magnetic resonance imaging of acute amstring injuries. Am J Sports Med. 2006;34: Ekstrand J, Healy JC, Waldén M, et al. Hamstring muscle injuries in professional football: te correlation of MRI findings wit return to play. Br J Sports Med. 2012;46: Gibbs NJ, Cross TM, Cameron M, et al. Te accuracy of MRI in predicting recovery and recurrence of acute grade one amstring muscle strains witin te same season in Australian Rules football players. J Sci Med Sport. 2004;7: Rettig AC, Meyers S, Kersey PA, et al. Categorization of amstring strain injuries by MRI and playing time lost in professional football players. NATA News 2008; Silder A, Serry MA, Sanfilippo J, et al. Clinical and morpological canges following 2 reabilitation programs for acute amstring strain injuries: a randomized clinical trial. J Ortop Sports Pys Ter. 2013;43: Slavotinek JP, Verrall GM, Fon GT. Hamstring injury in atletes: using MR imaging measurements to compare extent of muscle injury wit amount of time lost from competition. AJR Am J Roentgenol. 2002;179: Verrall GM, Slavotinek JP, Barnes PG, et al. Diagnostic and prognostic value of clinical findings in 83 atletes wit posterior tig injury: comparison of clinical findings wit magnetic resonance imaging documentation of amstring muscle strain. Am J Sports Med. 2003;31: Reurink G, Goudswaard GJ, Tol JL, et al. Terapeutic interventions for acute amstring injuries: a systematic review. Br J Sports Med. 2012;46: Heidersceit BC, Serry MA, Silder A, et al. Hamstring strain injuries: recommendations for diagnosis, reabilitation, and injury prevention. J Ortop Sports Pys Ter. 2010;40: Koulouris G, Connell DA, Brukner P, et al. Magnetic resonance imaging parameters for assessing risk of recurrent amstring injuries in elite atletes. Am J Sports Med. 2007;35: Bar R, Holme I. Risk factors for sports injuries: a metodological approac. Br J Sports Med. 2003;37: