An Instrumental Activities of Daily Living Decline Detection Analysis Pipeline Using Temporal Changes in In-Home Sensor Data

This study proposes a practical analysis pipeline for detecting early decline in instrumental activities of daily living (IADL) from in-home sensor data, aiming to capture behavioral signs that are less affected by individual differences in lifestyle and living environments. Early detection of IADL decline is critical for maintaining independence in older adults, yet conventional assessments based on self-reports or home visits provide only intermittent and subjective information. We instead quantify temporal changes in behavior using data from passive infrared motion sensors installed in the home. Using data from 22 single-living older adults, we extracted 34 input features: temporal differences in 33 daily behavioral indicators between two 28-day observation windows, and one static clinical attribute representing walking ability. A logistic regression model was evaluated with subject-wise leave-one-subject-out cross-validation. The proposed method achieved an area under the receiver operating characteristic curve of 0.858 with a 95% confidence interval of 0.714 to 0.974, indicating promising performance for early screening. Exploratory feature reduction suggested that a subset of 14 features related to four behavioral groups (outing, nighttime toilet use, nighttime awakening, and post-waking toilet use) preserved the performance of the full Delta model, implying the possibility of optimizing sensor placement and reducing implementation costs. These findings support the feasibility of an objective, data-driven approach to early IADL decline detection from unobtrusive in-home monitoring and may contribute to preventive interventions that help older adults maintain independent living.