Abstract
Background
Seizures are common after out‐of‐hospital cardiac arrest and may exacerbate secondary brain injury, yet early prediction remains challenging. Although heart rate variability has been used to predict seizures in epilepsy, its role and optimal prediction window after cardiac arrest remain unclear. We aimed to develop a heart rate variability–based model for predicting post–cardiac arrest seizures and identify the optimal prediction window.
Methods
This study prospectively collected and retrospectively analyzed data from adults resuscitated from nontraumatic out‐of‐hospital cardiac arrest and admitted to an academic post–cardiac arrest care center. Electrocardiography and electroencephalography were continuously recorded, and seizures were independently confirmed by a neurologist. Five‐minute time‐, frequency‐, and nonlinear‐domain heart rate variability features from 0 to 30 minutes before seizure onset were normalized to baseline and entered into a support vector machine classifier using training, validation, and testing sets.
Results
Among 36 patients, 20 (56%) experienced 93 seizures. Model performance was highest 15 to 20 minutes before seizure onset. At 18 minutes, the model achieved an area under the receiver operating characteristic curve of 0.812 (95% CI, 0.792–0.832), 76% accuracy, 77% sensitivity, and 68% specificity. Compared with baseline, low frequency increased 3.28‐fold, sample entropy 2.58‐fold, high frequency 2.43‐fold, and the low/high frequency ratio 1.47‐fold in seizure segments; corresponding ratios in nonseizure segments remained near baseline. These findings suggest that autonomic function fluctuates substantially before post–cardiac arrest seizures and may provide an early warning window for clinically important neurologic events during intensive care.
Conclusions
Heart rate variability changed before post–cardiac arrest seizures, with 18 minutes before onset appearing optimal for prediction.