Can Machine Learning Predict Knee Injury Severity?: A Review Of High School Basketball Athletes From 2005-2020

Abstract

Severe high school (HS) sports injuries can be devastating for developing athletes and young adults. Within HS athletics, basketball is one of the most popular sports to play and thus, presents a significant injury exposure risk. Across all basketball levels, the knee is one of the most injured locations. To inform preventative care efforts and reduce the incidence of knee injuries, 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 knee injury severity. METHODS: The High School Reporting Information Online (RIO) Study is a database documenting high school athletic exposures and injuries from 2005-2020. All primary basketball knee injuries were procured from this database and a total 160 predictor variables were identified from this database. The primary outcome was knee 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: 1924 basketball-related knee injuries were identified, with an average age of 15.94 years (SD = 1.76). Across all basketball athletes, 52.77% were male. RF had the best performance (AUC: 0.8436 ± 0.02), followed by XGBoost (AUC: 0.8402 ± 0.02), LR (AUC: 0.8275 ± 0.02), and SVC (AUC: 0.8179 ± 0.02). For basketball-specific variables, playing a position on the baseline (coefficient = 0.8953), shooting a layup (coefficient = 0.8907), and contact with the floor (coefficient = 0.7196) were the highest contributing predictors. CONCLUSION: Machine learning was successfully used to predict injury severity in high school basketball knee injuries. Specifically, playing a position on the baseline and shooting a layup were the top predictors of severe knee injury.

Ähnliche Arbeiten