Related Experiment Video
Updated: May 26, 2026
11:32
A Neuroscientific Approach to the Examination of Concussions in Student-Athletes
Published on: December 8, 2014
Predicting time to clearance of sport-related concussions using machine learning
Megan Tran1, Jessica Holler2, Byron Moran2
1Bellini College of Artificial Intelligence, Cybersecurity, and Computing, University of South Florida, Tampa, FL, USA.
Digital Health
|May 25, 2026
Summary
Integrating longitudinal data modestly improves machine learning predictions for sport-related concussion (SRC) recovery. Headache during VOR testing is a key predictor of prolonged recovery, aiding early risk stratification.
Area of Science:
- Sports Medicine
- Neurology
- Machine Learning in Healthcare
Background:
- Sport-related concussion (SRC) recovery prediction is challenging.
- Accurate prediction aids in timely medical clearance and rehabilitation planning.
Purpose of the Study:
- To assess if longitudinal clinical data enhances machine learning (ML) predictions of SRC recovery time.
- To identify clinical features predicting prolonged (≥30 days) or normal (<30 days) recovery.
Main Methods:
- Retrospective analysis of 217 athletes' data from the USF Concussion Center (2021-2025).
- Trained six ML classifiers using Visit 1 (n=48) and combined Visit 1+2 (n=95) features.
- Internal validation via Leave-One-Out Cross-Validation (LOOCV).
Main Results:
- Adding Visit 2 data improved ML model accuracy by up to 5% (XGBoost: 0.84).
- Prolonged recovery (81.1%) was predicted by VOR Vertical Headache and its change score.
- Treatment presence between visits predicted normal recovery.
Conclusions:
- Longitudinal data offers modest improvements in ML-based SRC recovery prediction.
- Vestibulo-oculomotor symptoms, especially headache during VOR testing, are strong prognostic indicators.
- Findings support using VOMS subscores for early risk stratification and targeted rehabilitation.
