Résumé
BACKGROUND: This study investigates the key variables influencing injury occurrence in elite-level female volleyball players. It aims to evaluate three hypotheses: (1) the quantification of workload using the "System Training Response" score provides a superior explanation and prediction of injury occurrence compared to traditional methods such as sum or mean; (2) both menses and external workload serve as primary variables that explain injury occurrence; and (3) non-linear models yield better explanatory and predictive capabilities for injury occurrence than linear models. METHODS: Nineteen elite female volleyball players were monitored throughout a 190-day competitive season, during which various training-related parameters were collected. These parameters included internal and external workload quantification and menses data. To analyze and predict injury occurrence, machine learning techniques were employed, with a particular emphasis on Random Forest models. RESULTS: The Random Forest model demonstrated superior performance in describing injury occurrence, achieving an area under the receiver operating characteristic curve of 0.87. Key variables identified as significant contributors to injury occurrence included the players’ age, menses status, and the percentage of intense jumps executed. Furthermore, the cross-validation procedure conducted on a reserved portion of the dataset yielded positive results, with an area under the receiver operating characteristic curve of 0.74, indicating a good generalization performance of the model. CONCLUSIONS: The findings of this study suggest that intense training prior to performance may increase the risk of injury, while older players appear to exhibit a lower risk of injury. These insights highlight the importance of tailored training strategies that consider both physiological factors and individual player profiles to mitigate injury risks in elite female volleyball athletes, including the presence or absence of menstruation with associated discomfort, which appears to be a relevant factor.