Heart Rate Variability Integrated into a Longitudinal Multimodal Model to Predict Delayed Cerebral Ischemia After Aneurysmal Subarachnoid Hemorrhage.

backgroundDelayed cerebral ischemia (DCI) is a major cause of morbidity after aneurysmal subarachnoid hemorrhage (aSAH), yet early prediction remains difficult, especially in comatose patients. Existing risk scores are largely static and do not capture dynamic pathophysiological changes. Heart rate variability (HRV), reflecting autonomic nervous system activity, may provide additional predictive information. We investigated whether integrating HRV into a longitudinal multimodal model improves DCI prediction.

methodsIn this prospective cohort study, continuous ECG data were collected from patients with aSAH admitted to a tertiary neurocritical care unit between 2016 and 2024. HRV features were computed in real time from high-resolution ECG recordings and combined with demographic, clinical, laboratory, and blood gas variables. Four prediction models were evaluated: static clinical variables, laboratory/blood gas data, HRV features, and a multimodal combined model. Performance was assessed using cross-validation, anchored ROC analysis, and leave-one-out simulations.

resultsAmong 101 eligible patients, 35 (34.7%) developed DCI. LFNorm showed the best univariate HRV performance (ROC AUC 0.64). HRV features and laboratory/blood gas models achieved ROC AUCs of 0.60 and 0.63, respectively. The multimodal model achieved the highest overall performance (ROC AUC 0.68), with HRV features accounting for more than half of the model's feature importance. Time-resolved analyses further demonstrated that the contribution of HRV varied across the clinical course, providing the greatest incremental value during the later phase preceding DCI.

conclusionsIntegrating HRV into a multimodal prediction framework provides additional time-dependent information for DCI risk estimation in patients with aSAH.