Validating models of injury risk prediction in football players

Abstract

Association football (soccer) is a popular sport and there is a high risk of injury for participants. Within the context of professional clubs, the risk of injury is also associated with the risk of financial costs. Therefore, injury reduction processes are considered important, and previous studies have sought to identify and model injury risk factors. Although formal screening tests e.g. The Functional Movement Screen (FMS) and monitoring procedures e.g. Union of European Football Associations (UEFA) have been developed for modelling and predicting injuries, the processes in current use, lack precision or clinical usefulness. The aims of this thesis were therefore to explore why existing methods of screening, measuring and modelling are not effective in predicting injuries. In order achieve this the following things were done;

Literature review to evaluate the UEFA screening process and advocated variables,

Validation of the FMS, the most commonly used exercise screening test, against a 3D photogrammetric system (Vicon (©Vicon Motion Systems Ltd))

Injury modelling on a pre-established database designed in accordance with the UEFA guidelines

The literature review confirmed that the established database was compliant with the UEFA screening guidelines. The most commonly used screening measure (FMS) for injury risk was found to be an invalid measure and therefore removed from the modelling process. The models developed were unable to prospectively model injuries accurately (R = 0.23), and the primary problem was a large number of false positives i.e. those predicted as having risk of injury not sustaining injury. Reasons for poor model performance could be attributed to inappropriate screening methods, inadequate datasets or inadequate modelling methods for rare events.
Future work should focus on addressing the limitations in the existing UEFA screening framework and simultaneously develop better methods of rare event modelling from small datasets.