Multivariate Machine Learning Analysis of M-ECG-derived Heart Rate Variability in TBI Veterans, With and Without Comorbid PTSD

Traumatic brain injury (TBI) and posttraumatic stress disorder (PTSD) frequently co-occur in Veterans, producing overlapping symptoms and shared autonomic dysregulation. Heart rate variability (HRV) offers a noninvasive measure of autonomic function. Univariate HRV analyses often fail to capture complex, multivariate patterns associated with comorbidity. This study applied machine learning to HRV features extracted from MEG-derived electrocardiogram (M-ECG) signals to differentiate Veterans with TBI alone (TBI-alone; n = 42) from those with comorbid PTSD (TBI+PTSD; n = 40). Time-domain, frequency-domain, geometric, and nonlinear HRV metrics were analyzed using nested cross-validated Random Forest and XGBoost classifiers, with Boruta-based feature selection and SHapley Additive exPlanations for model interpretability. Both classifiers achieved above-chance discrimination (Random Forest AUC = 0.663; XGBoost AUC = 0.635). Multivariate models identified distributed autonomic signatures in TBI+PTSD, including altered sympathovagal balance, increased low-frequency proportion, and greater heart rate complexity. In contrast, univariate HRV differences were subtle and did not survive correction for multiple comparisons. These findings demonstrate how using multivariate machine learning HRV analysis could help with detecting comorbidity-specific autonomic patterns, suggesting that HRV-derived signatures may serve as exploratory biomarkers for risk assessment and targeted interventions in Veterans with TBI and PTSD.