Heart rate variability stimulation-to-baseline ratios combined with a support vector machine model predicted 3-month outcomes in patients with disorders of consciousness with 97% accuracy in the training set and 80% accuracy in an independent test set.
Cohort (n=50)
No
Does heart rate variability (HRV) ratio analysis combined with support vector machine (SVM) classification improve prediction of 3-month outcomes in patients with disorders of consciousness?
Heart rate variability ratios between stimulation and baseline combined with machine learning can accurately predict 3-month outcomes in patients with disorders of consciousness.
Disorders of consciousness (DoC), including unresponsive wakefulness syndrome (UWS/VS) and minimally conscious state (MCS), pose significant diagnostic challenges due to their complexity and high misdiagnosis rates. This study investigates the prognostic potential of heart rate variability (HRV) ratios between stimulation and baseline, combined with support vector machine (SVM) classification, to predict outcomes in DoC patients. Fifty patients were enrolled within the first 10 days of hospitalization; 40 were used to train and optimize the SVM model, while 10 served as an independent test group. HRV analysis employed ratios of high-frequency and low-frequency components along with Sample Entropy to capture dynamic autonomic changes. Assessments were conducted weekly over three weeks. The SVM achieved 97% overall accuracy (misclassification 3%), 96% sensitivity, 100% specificity, and 97% balanced accuracy during training (10-fold cross-validation: 0% misclassification). In the independent test set (N = 10), performance was 80% overall accuracy (misclassification 20%), 80% sensitivity, 80% specificity, and 80% balanced accuracy. These results highlight the value of the HRV ratio approach, particularly the early recovery of vagal response followed by sympathetic activation, in predicting patient trajectories. Although the sample size is small, our findings support the integration of HRV analysis with machine learning as a promising tool for enhancing prognostic assessments in DoC. Future research should replicate these findings in larger cohorts and incorporate longitudinal data.
Riganello et al. (Fri,) conducted a cohort in Disorders of consciousness (n=50). Heart rate variability (HRV) stimulation-to-baseline ratios and support vector machine (SVM) classification was evaluated on 3-month outcome prediction accuracy (good vs. bad outcome). Heart rate variability stimulation-to-baseline ratios combined with a support vector machine model predicted 3-month outcomes in patients with disorders of consciousness with 97% accuracy in the training set and 80% accuracy in an independent test set.