30 Predicting running-related injuries based on runner characteristics

Introduction Running-related injuries affect up to 40% of runners, underscoring the need for accurate prediction to inform prevention strategies. To our knowledge, this is the second study to investigate multivariable prediction models for individual injury risk, and the first to apply a survival framework, which properly handles censored time-to-event data and thus improves methodological rigor.Materials and Methods Data were derived from the Garmin-RUNSAFE Running Health Study, a cohort including 6.000 adult runners recruited via online platforms. Prediction models for 12- and 26-week injury risk were developed using Cox proportional hazards regression and random survival forests. Models used only baseline self-reported variables, reflecting information typically available to sports professionals in a single consultation.Results Discrimination was modest for both Cox and Random Survival Forest models (AUC ≈ 0.64 at 12 weeks; 0.65 at 26 weeks). The narrow spread of predicted risks (0.12–0.58) suggests minimal differentiation between runners, limiting the practical utility of the models.Conclusion Clinicians should be aware of the limited discriminative ability observed in this study. It indicates that questionnaire data provide only modest guidance for individual risk assessment. This limited discriminative ability likely reflects the multifactorial nature of running injuries, which depend on training habits, recovery, and previous injury. These factors are difficult to capture even with advanced statistical models. Although prediction models may eventually help guide prevention strategies, their current performance does not justify clinical use. Consequently, the present and similar models should be interpreted with caution.