AI and the World's Hard Problems · Medicine & drug discovery · Nature (computational biology) · 2026-10-22
Predicting acute intracranial vascular events in stroke alerts using large-language models and structured data
Acute stroke alerts are frequently activated for non-cerebrovascular conditions, creating false positives that strain resources. Researchers developed machine learning models combining large language models, structured electronic health record data, and clinical time-series features to predict acute intracranial vascular events at stroke alert activation, using retrospective alerts from the Mount Sinai Health System. Among 16,512 alerts, the multimodal model achieved an AUROC of 0.73 with sensitivity of 0.76 and specificity of 0.56, suggesting it could support post-activation risk stratification and reduce false-positive evaluations.