Multidisciplinary prediction of running-related injuries using machine learning

Area of Science:

• Sports Medicine
• Biomechanical Engineering
• Data Science

Background:

• Endurance running-related injuries (RRIs) have complex, multifactorial causes.
• Existing research often overlooks multidisciplinary risk factors for individualized RRI prediction.

Purpose of the Study:

• To create a machine learning-ready dataset for weekly RRI prediction.
• To evaluate the efficacy of machine learning models using multidisciplinary risk factors.

Main Methods:

• Collected data on genetic, historical, biomechanical, physiological, and training factors from 142 competitive runners over 12 months.
• Developed and tested machine learning models using both high-evidence and broader sets of risk factors.
• Prospectively monitored runners for RRIs, accumulating 6181 weekly samples.

Main Results:

• Machine learning models achieved an AUC of 0.784 ± 0.014, showing moderate improvement over previous RRI prediction.
• Random Forest models demonstrated the highest performance (AUC = 0.781 ± 0.016).
• Logistic regression performance significantly improved with a broader set of risk factors.

Conclusions:

• Introduced a reproducible framework for machine learning-based sports injury prediction.
• Provided a valuable dataset for future large-scale sports injury analytics.
• Highlighted the importance of multidisciplinary data and model selection for accurate RRI prediction.