A0962
Title: Comparison of prediction models for survival analysis of running-related injuries
Authors: Katarzyna Szczerba - University of Luxembourg (Luxembourg) [presenting]
Laurent Malisoux - Luxembourg Institute of Health (Luxembourg)
Christophe Ley - University of Luxembourg (Luxembourg)
Abstract: As awareness grows about the benefits of physical activities, there is an increasing need for advanced tools in sports medicine. Running, while popular and beneficial, has a high injury rate, leading to severe long-term impacts. This raises a crucial question: what are the primary risk factors for running injuries, and what preventive measures should be considered? To address this issue, the Luxembourg Institute of Health conducted a randomized controlled trial in 2017 involving 848 runners, with a 6-month follow-up. While several insightful papers resulted, none used machine learning due to its 'black box' nature and initial underperformance compared to traditional Cox models. The aim is to develop a superior machine learning model to identify risk factors for running-related injuries using innovative interpretable machine learning (IML) methods. Using 10-fold cross-validation with the concordance index as the evaluation metric, we found that a gradient-boosted Cox proportional hazards model with regression trees as base learners outperformed all other models. To ensure explainability, SHapley Additive exPlanations (SHAP) is also employed, and additional statistical analyses are conducted. This case study illustrates that sports scientists can achieve deeper insights into their data by employing advanced machine learning models with interpretable machine learning (IML) methods.
