Multimodal Wrist-Based Prediction of Neurological and Metabolic Events: A Proof-of-Concept Algorithm Study Using Open Clinical Datasets

This preprint presents three proof-of-concept algorithms for multimodal wrist-based prediction and detection of acute physiological events: epileptic seizures, severe hypoglycemia, and atrial fibrillation. Using open-access clinical datasets from PhysioNet, the study evaluates autonomic and cardiovascular signal patterns relevant to pre-event warning. The seizure prediction algorithm generated an alert 7 minutes before documented seizure onset; the hypoglycemia algorithm generated a pre-threshold alert 11 minutes before modeled glucose crossed 3.9 mmol/L; and the atrial fibrillation algorithm detected RR-interval irregularity in real time. These findings provide proof-of-concept support for a multimodal wearable architecture integrating physiological sensing and event-specific prediction algorithms. Prospective clinical validation with dedicated wearable sensors is required.