Predicting Ankle Injury Severity In High School Basketball Players Through Machine Learning Techniques

Severe injury in high school sports can be detrimental to developing athletes. Basketball is one of the most popular sports to play in high school and presents a significant injury exposure risk. In high school basketball, the ankle is one of the most injured locations. To inform preventative care efforts, we analyzed factors surrounding injury to determine if certain variables predispose athletes to severe injury. PURPOSE: The purpose of this study is to apply machine learning models onto a national high school basketball injury database to predict ankle injury severity. METHODS: All primary basketball ankle injuries were acquired from the High School Reporting Information Online (RIO) Study, a database documenting high school athletic exposures and injuries from 2005-2020. A total 160 predictor variables were identified from this database. The primary outcome was ankle injury resulting in prolonged return to sport (RTS), defined as needing ≥22 days, including medical disqualification for season. A total 4 machine learning models—Logistic Regression (LR), Random Forest (RF), Support Vector Classifier (SVC), and eXtreme Gradient Boosting (XGBoost)—were selected and optimized to maximize the area-under-the-receiver-operating-curve (AUC). Models were trained using an 80/20 train/test split. Feature importance analysis was performed on the highest performing model to identify most contributing predictor variables via SHAP scores. RESULTS: 4300 basketball-related ankle injuries were identified, with an average age of 15.93 years (SD = 1.29). 52.77% of all basketball athletes were male. XGBoost had the best performance (AUC: 0.87 ± 0.02), followed by LR (AUC: 0.86 ± 0.02), RF (AUC: 0.86 ± 0.02), and SVC (AUC: 0.84 ± 0.02). For basketball-specific variables, court surface outside of wood or synthetic (coefficient = 0.9713) and dribbling during fast break with ankle inversion (coefficient = 0.8893) were the highest contributing predictors. CONCLUSION: Machine learning was successfully used to predict injury severity in high school basketball ankle injuries.