Key Takeaways
- KAIST researchers created an AI framework that analyzes long‑term lifelog data (activity, sleep, circadian rhythm, indoor environment) from older adults to detect early cerebrovascular disease risk.
- Using data from 1,224 participants (13,362 two‑week samples), the AI distinguished a “prodromal risk group” between healthy and diagnosed individuals.
- The model identified an “imminent diagnostic risk period” (≤ 4 weeks before diagnosis) with 96.53 % accuracy by tracking lifestyle changes over time.
- Explainable AI revealed specific behavioral markers: increased nighttime activity (10 p.m.–2 a.m.), blurred day‑night distinction, reduced evening activity (6 p.m.–10 p.m.), increased inactive time, and low indoor humidity as risk indicators.
- The technology is intended as a supportive, preventive tool—not a replacement for clinical diagnosis—and requires further validation before clinical deployment.
Study Overview and Motivation
Cerebrovascular disease often progresses silently, making early detection challenging. To address this gap, a multidisciplinary team from KAIST, Sungkyunkwan University, and Korea University Anam Hospital leveraged real‑world lifelog data collected in older adults’ homes. Their goal was to uncover subtle, pre‑symptomatic behavioral changes that could signal heightened risk, thereby shifting care from reactive treatment toward proactive prevention.
Data Collection and Sample Size
The researchers partnered with LivOn Care Co., Ltd. to obtain continuous sensor‑based lifelog recordings from 1,224 older adults residing in typical home environments. Over the study period, this yielded 13,362 two‑week segments, encompassing metrics such as physical activity, sleep patterns, circadian rhythms, and indoor conditions (temperature, humidity). The large, longitudinal dataset enabled robust training and testing of machine‑learning models while reflecting real‑life variability.
AI Framework Design
The core of the study is an AI model that fuses multiple data streams: daily activity levels, sleep timing and quality, circadian rhythm strength, indoor environmental factors, plus baseline covariates like age and chronic disease history. By processing these heterogeneous inputs, the algorithm learns patterns associated with the prodromal phase of cerebrovascular disease—periods when pathological changes exist but clinical symptoms are still absent.
Identifying the Prodromal Risk Group
When the AI was applied to the lifelog corpus, it successfully separated participants into three categories: healthy, prodromal (at‑risk), and clinically diagnosed. The prodromal group exhibited distinct digital signatures that were not apparent through conventional clinical screening, demonstrating that home‑based monitoring can capture early disease trajectories missed by intermittent hospital visits.
Defining Imminent Diagnostic Risk
To evaluate whether the AI could forecast an approaching diagnosis, the researchers labeled lifelog windows: data collected within four weeks before a clinical diagnosis marked the “imminent diagnostic risk period,” while data from twelve weeks prior served as a “non‑imminent” baseline. The model discriminated between these periods with an impressive accuracy of 96.53 %, indicating that measurable lifestyle shifts occur shortly before disease manifestation.
Explainable AI Insights – Nighttime Activity
Interpretability analysis revealed that individuals in the prodromal phase often displayed frequent continuous movement between 10 p.m. and 2 a.m., a window when physiological processes normally favor sleep onset and maintenance. This nocturnal activity suggests delayed sleep timing and a weakened demarcation between day and night, both of which emerged as strong early warning signals.
Explainable AI Insights – Evening Inactivity and Environment
As the diagnosis neared, the AI detected a notable decline in continuous activity during the early evening (6 p.m.–10 p.m.), accompanied by increased periods of inactivity. Concurrently, lower indoor humidity—reflecting drier home environments—was consistently associated with imminent risk. These findings highlight how both behavioral retreat and environmental dryness may contribute to or reflect pathophysiological changes preceding cerebrovascular events.
Potential Impact on Digital Healthcare
The researchers envision deploying this AI‑driven monitoring system as a passive, objective health‑surveillance tool for older adults who may struggle to articulate subtle health changes. By delivering timely risk alerts to clinicians and caregivers, the technology could facilitate earlier medical consultation, lifestyle interventions, and preventive care, ultimately reducing the burden of advanced cerebrovascular disease.
Limitations and Future Steps
While promising, the approach does not claim to predict the exact onset of stroke or other cerebrovascular events, nor does it substitute for professional diagnostic evaluation. The authors stress that further prospective validation in larger, more diverse cohorts is essential before the system can be adopted in routine clinical practice. Ongoing work will focus on refining model generalizability, integrating additional biomarkers, and ensuring user privacy and data security.
Researcher Perspective and Conclusion
Professor Lisa Lim emphasized that the study’s primary value lies in using AI to notice modest lifestyle deviations at home, thereby bridging the gap between everyday living and timely medical attention. She anticipates that such tools will help transition healthcare from a reactive model—treating disease after it appears—to a preventive paradigm that supports early intervention and sustained well‑being for aging populations.
Publication and Support
The findings were published on June 2, 2026, in npj Digital Medicine (DOI: 10.1038/s41746-026-02836-7), a Nature‑Portfolio journal with an impact factor of 15.1, placing it among the top 0.3 % of JCR-ranked journals. The work received funding from the National Research Foundation of Korea (grant RS‑2025‑16068234), underscoring national commitment to advancing digital health solutions for cerebrovascular disease prevention.