Analysis of contactless sensor data from 1,224 older adults in South Korea found that an AI model could recognize activity and sleep patterns in the month before diagnosis with high accuracy. It is early research, not a clinical tool.
What is new
A research team in South Korea reports that ordinary, contactless sensors in the home can supply enough signal for an AI system to spot patterns linked to impending cerebrovascular disease. The work centers on older adults who live alone, a group that may miss subtle day-to-day changes without a nearby observer.
How the study worked
- Participants: 1,224 adults aged 65 and older, living alone and able to walk independently.
- Sensors: passive motion, door and temperature or humidity sensors placed in homes to capture activity, inactivity, sleep timing and indoor conditions over long-term monitoring.
- Data scope: analyses referenced more than 13,000 data points aggregated across 14 days.
- Publication status: early-access paper in npj Digital Medicine. The approach is not a product and has not been tested prospectively for clinical outcomes.
What the numbers say
The authors trained and compared several models. The best performer, TabNet, achieved an area under the precision-recall curve of 85 percent for identifying a prodromal group, and an area under the ROC curve of 91 percent for separating people with prior cerebrovascular diagnoses from those with no history.
Focusing on the month leading up to a new diagnosis, the model reached 95.12 percent sensitivity, 96.97 percent specificity and 96.53 percent accuracy, according to the paper. These results come from retrospective testing rather than live deployment.
Signals that stood out
The patterns the model used were grounded in everyday behavior and the indoor environment:
- Sleep schedule and continuity: bedtimes shifted later in the pre-diagnostic group, while people with a previous cerebrovascular diagnosis showed more motion overnight along with more frequent sleep interruptions.
- Activity consolidation: before going to sleep, some individuals showed longer uninterrupted activity runs and spent less time inactive during this phase.
- Environment-behavior linkage: evenings marked by more inactivity and fewer extended activity bouts, together with very low or very high indoor humidity, were associated with a higher likelihood of being within four weeks of diagnosis in this dataset.
Why it could matter for independent living
- Non-wearable appeal: passive, contactless sensors avoid common adherence issues with wearables like charging or remembering to put a device on.
- Actionability if validated: earlier detection could prompt medication reviews, blood pressure checks, atrial fibrillation screening or faster clinical assessment before a disabling event.
- Interpretability: features such as bedtime shifts, nocturnal motion and the length of activity bouts are intuitive to clinicians and caregivers, which may help real-world adoption if the approach proves effective.
Across the Asia-Pacific region, many hospitals are adopting AI to speed in-hospital stroke care. This home-based strategy targets an earlier point on the timeline.
Limits and open questions
- Generalization: housing layouts, climate and daily routines in South Korea may differ from other regions, which could affect performance.
- Study design: results are retrospective and do not show that real-time alerts change outcomes.
- Clinical pathway: who receives alerts, how often and with what follow-up need careful design to avoid false alarms and alarm fatigue.
- Privacy and consent: any future system would require clear consent, data minimization and strong governance.
What you can do now
- Keep proven prevention front and center: know your blood pressure, take prescribed medications, and seek urgent care for new neurological symptoms such as sudden weakness, speech difficulty, facial droop or vision loss.
- For those living alone, simple non-camera sensors can support safety check-ins. Treat any alerts about unusual inactivity as a prompt to reach out, not a diagnosis.
- If you already use smart-home sensors, choose practical notifications for patterns you care about, and discuss any data sharing with your clinician first.
- Maintain comfortable indoor conditions for overall wellbeing. The study linked humidity extremes with risk in its dataset, but it did not test whether changing humidity alters outcomes.
Bottom line
This study suggests that ordinary home sensors, paired with AI, might one day provide early warnings for stroke risk in older adults who live alone. The idea is promising, but it needs prospective trials and proof of clinical benefit before it can inform care.
Watch
Find out more
- AI home monitoring for behavioral markers of ... — Publisher page for the npj Digital Medicine paper reporting this study — useful for the study’s methods, sample size (1,224 people) and the behavioural and environmental features the authors analysed.
- Stroke Symptoms and Warning Signs | American Stroke Association — American Stroke Association guidance on stroke symptoms and warning signs (B.E. F.A.S.T.), and clear advice on when to seek emergency care.
- FTC Report on Internet of Things Urges Companies to Adopt Best Practices to Address Consumer Privacy and Security Risks | Federal Trade Commission — Federal Trade Commission report with practical privacy and security recommendations for Internet‑of‑Things devices in the home — relevant if you’re considering passive sensors.
- FDA Releases Artificial Intelligence/Machine Learning Action Plan | FDA — U.S. Food and Drug Administration news release outlining the AI/ML‑Based Software as a Medical Device Action Plan — explains regulatory approaches that apply to AI health tools.
Image: Photo: Centre for Ageing Better / Pexels
