Predicting ACL Injury Using Machine Learning on Data From an Extensive Screening Test Battery of 880 Female Elite...

Area of Science:

• Sports Medicine
• Biomechanical Engineering
• Data Science in Sports

Background:

• Injury risk prediction is an evolving field requiring established best practices for accurate assessment.
• Machine learning (ML) in injury prediction necessitates careful consideration to avoid overinterpreting performance.
• Understanding factors contributing to injuries like anterior cruciate ligament (ACL) tears is crucial.

Purpose of the Study:

• To evaluate the predictive performance of various ML methods for ACL injury using extensive risk factor data.
• To account for chance and random variations in prediction performance within the analysis.
• To investigate the potential of ML in identifying ACL injury risk factors in elite athletes.

Main Methods:

• A case-control study involving 791 elite female handball and soccer players.
• Utilized 3D motion analysis and physical data to predict ACL injuries (n=60).
• Employed four common ML classifiers and assessed performance using average Area Under the Receiver Operating Characteristic Curve (AUC-ROC) across 100 cross-validations, confirmed with permutation tests. Evaluated class imbalance techniques.

Main Results:

• The best performing classifier (linear support vector machine) achieved a mean AUC-ROC of 0.63.
• All classifiers demonstrated predictive ability significantly better than chance.
• AUC-ROC values ranged from 0.51 to 0.69, with substantial variation. Class imbalance handling did not enhance results.

Conclusions:

• The study demonstrated statistically significant predictive ability, suggesting valuable information exists for understanding injury causation.
• Despite statistical significance, the predictive performance was low for clinical applications.
• Current variables and methods are insufficient for practical ACL injury prediction in athletes.