Passive Smart Home Monitoring for Delirium-Relevant Anomaly Detection in People Living With Dementia:...

Area of Science:

• Gerontology
• Artificial Intelligence in Healthcare
• Digital Health

Background:

• Delirium superimposed on dementia leads to poor outcomes and is often undetected in home settings.
• Current detection methods, like the Confusion Assessment Method (CAM), are rarely used outside hospitals.
• Passive monitoring offers a potential solution for detecting delirium in community-dwelling individuals.

Purpose of the Study:

• To develop and test a framework for detecting delirium-consistent anomalies using passive smart home sensor data in people with dementia.
• To explore the feasibility of using AI-driven anomaly detection for early delirium identification in a home environment.
• To approximate key CAM criteria using digital markers from sensor data.

Main Methods:

• Utilized the Technology Integrated Health Management dataset of older adults with dementia or mild cognitive impairment.
• Applied Isolation Forest and Long Short-Term Memory (LSTM) models for individualized anomaly detection.
• Incorporated theory-driven digital markers (activity entropy, sleep quality) and clinical indicators (early warning scores, UTIs).

Main Results:

• Both Isolation Forest and LSTM models identified a similar number of anomalies (approx. 15.6-15.8%) in short temporal clusters.
• Feature importance analysis highlighted activity entropy, sleep quality, and early warning scores as key indicators.
• Interfeature correlations were stronger during anomaly periods, suggesting distinct behavioral patterns.

Conclusions:

• Demonstrated the technical feasibility of detecting delirium-related anomalies via passive smart home monitoring.
• The approach shows promise for enabling early intervention in community settings for dementia patients.
• Further validation studies with clinical delirium labels are crucial for confirming accuracy and utility.