AI home monitoring for behavioral markers of cerebrovascular disease.

Cerebrovascular disease (CeVD) is a major health concern in aging populations, and early identification is crucial for improving outcomes. Conventional diagnostic approaches are hospital-centered and limited in capturing disease risks from behavioral changes at home. We propose a framework to identify potential CeVD prodromal individuals and estimate diagnostic risk using behavioral and environmental data collected from contactless sensors in real-world homes. We used 13,362 samples (14-day windows) from 1224 older adults (598 healthy, 28 prodromal, 598 diagnosed) in South Korea. In-home behavioral and environmental factors, along with demographics and comorbidities, were used to develop models for three tasks. The framework achieved an area under the precision-recall curve of 0.85 for prodromal identification (Task 1), an area under the receiver operating characteristic curve of 0.91 for classifying diagnosed patients (Task 2), and a sensitivity of 95.12%, specificity of 96.97%, and accuracy of 96.53% for predicting imminent diagnostic risk within the prodromal group (Task 3). Model interpretation identified key digital behavioral markers, including frequent continuous activity and shorter inactive time during bedtime preparation hours (Task 1) and evening hours (Task 3). Our approach offers the potential to facilitate early detection of CeVD at home and requires further validation before clinical application.