Background Use of survival models for the analysis of sports injury data has grown rapidly potentially enabling better understanding of time to sports injury events. Choosing an appropriate model is a key consideration for survival analysis of sports injuries.
Objective To identify the application of appropriate survival modelling strategies in sports injury prevention research.
Design Systematic review.
Setting Studies were identified by searching six databases and the table of contents of three journals, and evaluated using the PRISMA statement. Reference lists of selected papers were examined to identify additional relevant articles.The search was limited to: original sports injury studies in humans; English-language; published between Jan 1993 and July 2013, inclusive.
Main outcome measurements Features of survival models, estimators and graphical approaches.
Results A total of 2147 articles were identified, of which 103 met the inclusion criteria. There has been an increasing trend in use of survival analysis of sports injury data, with 88.0% of the reviewed articles published since 2005. Models identified were: Univariate Cox regression in 62 studies (60.2%); Multivariate Cox regression in 12 studies (11.7%); both univariate and multivariate cox regression in 12 studies (11.7%), frailty model with Univariate Cox regression, Andersen Gill model, WLW-TT model and PWP-GT model in two studies (1.9%); frailty model only in three studies (2.9 %); and Hurdle regression was used in one study (1.0%). Kaplan Meier only was applied in 11 studies (10.7%). Time to first event was considered in 48 studies (46.6%), and recurrent events in 14 studies (13.6%). Five studies used survival models specific for recurrent events.
Conclusions Use of survival analysis is currently largely limited to modelling of the time to first injury. Even though sports injuries are often recurrent, only in a small number of studies was the correlation structure of recurrent events considered. Future studies should consider recurrent injuries and apply the appropriate survival model.