Time-Domain HRV Metrics as Predictors of Concussion Recovery in Adolescents: A Boosted Tree Approach.

Abstract

Background Concussions in adolescents, especially in sports, represent a major public health issue due to prolonged recovery and diagnostic challenges compared to adults. Current clinical assessments often underestimate the true burden and recovery course, highlighting the need for objective physiological biomarkers. Objective To investigate heart rate variability (HRV) as a potential biomarker for monitoring concussion recovery in adolescents through both traditional statistics and machine learning approaches. Methods Thirty-seven concussed adolescents (23 females, 14 males; mean age 15 ± 2 years) and 37 age-matched healthy controls (14 females, 23 males; mean age 16 ± 2 years) were enrolled. Concussed participants were evaluated within 1 month post-injury and at clinical discharge (mean interval = 9 ± 4 days). Each session included a 5-minute electrocardiogram and Post-Concussion Symptom Scale (PCSS). Time-domain HRV metrics-standard deviation of normal-to-normal intervals (SDNN) and root mean square of successive differences (RMSSD), were severity-adjusted to PCSS severity (SDNNidx, RMSSDidx). A Boosted Tree algorithm predicted clinical outcomes (Good/Bad) based on raw and severity-adjusted HRV features. Results Conventional statistics revealed no significant group differences in raw HRV metrics. However, severity-adjusted indices increased with lower symptom severity. The optimized Boosted Tree model achieved promising discriminative performance (AUC = .88), with 82% to 83% sensitivity, 75% to 79% specificity, and 82% to 84% F1 scores, accurately classifying all controls. Conclusions Machine learning uncovered nonlinear HRV patterns predictive of clinical recovery where standard analyses failed. HRV-based predictive modeling may provide a noninvasive approach for individualized autonomic monitoring and evidence-based concussion management in adolescents.