AI Identifies Hidden Stroke Risk Through At-Home Behavior Patterns

Subtle, pre-symptomatic changes in daily living often go unnoticed in clinic-based evaluations. Earlier identification could prompt timely medical assessment and prevention. To help address this challenge, researchers have developed an artificial intelligence (AI) approach that analyzes real-life daily activity and environmental data from older adults to flag digital behavioral markers of cerebrovascular disease risk.

KAIST (Daejeon, South Korea) led the development of an AI framework designed to identify the prodromal phase of cerebrovascular disease and to assess imminent diagnostic risk. The work involved collaboration with Sungkyunkwan University’s School of Electronic and Electrical Engineering and the Department of Neurology at Korea University Anam Hospital. The system focuses on detecting risk trajectories at home, where changes emerge gradually and may precede symptom-driven care.

The study used lifelog data gathered in real residential environments from 1,224 older adults collected by LivOn Care Co., Ltd. Investigators analyzed 13,362 two-week lifelog samples to examine whether small day-to-day variations could reveal early warning signals. The analysis demonstrated the feasibility of pre-symptomatic detection based on routine living patterns rather than event-driven care.

The AI evaluates daily activity, sleep, circadian rhythm, and indoor environmental information, together with age and chronic disease data. It applies explainable AI methods to pinpoint which behavior and environment features underlie its judgments. By tracking longitudinal lifestyle patterns, the framework determines risk stages and highlights when a diagnostic encounter may be approaching.

Distinct behavioral signatures emerged. Older adults in the prodromal phase tended to show frequent continuous activity between 10 p.m. and 2 a.m., indicating irregular rhythms with delayed sleep onset and a blunted distinction between day and night activity. As diagnosis approached, evening (6 p.m. to 10 p.m.) continuous activity decreased while inactive time increased, and low indoor humidity also signaled an imminent diagnostic risk.

When lifelog data within four weeks before diagnosis were labeled as an “imminent diagnostic risk period,” and data from 12 weeks prior as a “non-imminent period,” the AI distinguished between these windows with 96.53% accuracy. The researchers emphasized that the technology does not predict the exact onset of disease or replace clinical diagnosis. They noted it is intended to support prevention and early consultation, and that prospective validation in larger patient groups is needed before clinical use.

The findings were published on June 2 in npj Digital Medicine.

“The key point of this study is not that AI should replace a hospital diagnosis, but that it can first detect risk signals in small lifestyle changes at home and help connect patients to medical care at the right time. We expect this technology to contribute to a shift from a healthcare system that treats disease after it occurs to one that supports prevention and early intervention,” said Lisa Lim, Professor, Department of Civil and Environmental Engineering, KAIST.

Related Links: 
KAIST