Technology and Dementia Preconference | JoVE Visualize

Area of Science:

Gerontology and Artificial Intelligence
Digital Health and Sensor Technology
Machine Learning in Healthcare

Background:

Caregiver burden assessment traditionally relies on subjective self-reports, lacking objective home activity data.
Previous research confirms the feasibility and acceptability of collecting home sensor data over extended periods.
Identifying specific home activities linked to caregiver burden in individuals with dementia remains an open question.

Purpose of the Study:

To develop a "digital signature" using home sensor data to quantify caregiver burden.
To identify objective home activity patterns associated with varying levels of care partner burden in dyads.
To explore the utility of machine learning for analyzing sensor data to predict caregiver burden.

Main Methods:

Analysis of clinical and home motion sensor data from longitudinal studies of dyads (care partner and individual with mild cognitive impairment or dementia).
Machine learning models, including Independent Component Analysis (ICA) and decision trees, were employed to analyze sensor features (e.g., room occupancy, movement patterns).
Care partner burden was measured weekly using the Zarit Burden Interview Short Form (ZBI-12); SHAP values identified key predictive features.

Main Results:

The predictive model achieved 69.8% accuracy in distinguishing low versus high caregiver burden using only motion sensor data from 44 dyads.
ICA identified three components from sensor data; one component strongly correlated with ZBI-12 scores.
Higher caregiver burden was significantly associated with increased number of trips to and average time spent in the bathroom.

Conclusions:

Home sensor data offers a promising avenue for continuous, objective assessment of daily activities contributing to caregiver stress.
Novel machine learning techniques can effectively extract meaningful outcome measures from complex sensor data.
Further research is ongoing to optimize sensor combinations for accurate caregiver burden prediction.