A Machine Learning Model for Post-Concussion Musculoskeletal Injury Risk in Collegiate Athletes.

Area of Science:

• Sports Medicine
• Biostatistics
• Machine Learning in Healthcare

Background:

• Emerging evidence suggests collegiate athletes have an increased risk of post-concussion musculoskeletal injuries.
• Identifying athletes most susceptible to these injuries requires further investigation.

Purpose of the Study:

• To develop a machine learning model for predicting post-concussion musculoskeletal injury risk in collegiate athletes.
• To integrate a comprehensive set of variables into the predictive model.

Main Methods:

• A risk model was developed using a dataset of 194 collegiate athletes.
• 135 variables including health, athletic history, concussion criteria, and assessment outcomes were analyzed.
• Machine learning techniques included weight of evidence transformation, L1/L2-regularized logistic regression, and Akaike Information Criterion for model selection.

Main Results:

• The final model, with 48 predictive variables, demonstrated significant predictive performance (Area Under the Curve = 0.82).
• Key predictors included baseline and acute cognitive, balance, and reaction assessments.
• The model achieved 79% sensitivity and 95% precision at a 6.67% false-positive rate.

Conclusions:

• The study supports the development of a sensitive and specific injury risk model for post-concussion musculoskeletal injuries.
• This machine learning approach, using comprehensive data, can help clinicians identify and target high-risk student athletes for injury prevention.