Researchers from Korea University Anam Hospital, the Korea Advanced Institute of Science and Technology (KAIST) and Sungkyunkwan University have developed and evaluated an artificial intelligence system that analyses smart home sensor data to identify behavioural changes associated with an elevated risk of stroke among older adults living alone.
The study included 1,224 participants aged 65 years and older who lived independently and were able to walk without assistance. Participants were categorised into three groups: individuals with no history of cerebrovascular disease, those with a previous diagnosis, and a prodromal group comprising individuals who were subsequently identified as being in the early stage of the disease.
The research team analysed more than 13,000 data points collected over a 14-day period using contactless Internet of Things sensors, including motion detectors, door sensors and temperature and humidity monitors. The dataset captured information on physical activity, inactivity, sleep behaviour and indoor environmental conditions. Six artificial intelligence models were trained to detect patterns associated with stroke risk.
According to findings published in an early-access paper in npj Digital Medicine, the best-performing model, TabNet, achieved an area under the precision-recall curve (AUPRC) of 85% in identifying the prodromal group and an area under the receiver operating characteristic curve (AUROC) of 91% in distinguishing participants with a previous diagnosis from those without cerebrovascular disease.
Retrospective testing showed that the model identified behavioural patterns recorded during the four weeks preceding a stroke diagnosis with 95.12% sensitivity, 96.97% specificity and 96.53% accuracy.
Researchers found that the prodromal group was characterised by more frequent sustained movement before sleep, reduced periods of inactivity and later sleep onset. Participants with a previous diagnosis demonstrated greater overnight activity and more frequent sleep interruptions. The analysis also associated increased evening inactivity, fewer sustained activity periods and unusually low or high indoor humidity with a greater likelihood of stroke diagnosis within four weeks.
Commenting on the findings, Dr Cho Kyung-hee, Professor of Neurology at Korea University Anam Hospital and the study's co-lead investigator, said, "Early intervention is crucial for the prognosis of cerebrovascular disease, but it is easy to miss subtle changes in elderly patients."
The researchers concluded that AI-enabled contactless home monitoring could complement clinical assessment by helping identify early warning signs, particularly among older adults living alone who may not immediately recognise changes in their health.
Click here for the original news story.