Detection of depression risk among older adults using home-deployed socially assistive robots: a real-world study.

INTRODUCTION

Monitoring depression among older adults using socially assistive robots provides scalable and continuous health surveillance while reducing the clinical burden on therapists and minimizing delays in treatment. This study aimed to predict depression risk and identify individuals in need of specialized depression care at local healthcare centers, using response and usage log data from the socially assistive robot Hyodol.

METHODS

A total of 215 community-dwelling older adults (170 in the 2024 cohort and 45 in the 2025 cohort) who used Hyodol were recruited. User responses to Hyodol's daily health check-in questions and free conversations, as well as physical interaction and content usage logs, were processed as features. Depression status was determined via clinical surveys and expert-led video consultations. A random forest model to predict depression status, defined as (i) symptomatic of depression and (ii) depression requiring referral to local healthcare centers, was trained on the 2024 cohort and tested on the 2025 cohort.

RESULTS

The model predicted symptomatic participants and participants requiring referral with sensitivities of 0.939 and 0.900, respectively. The model also produced considerable false positives. Features most strongly associated with depression status included engagement with quiz content, frequency of free conversations, positive responses to daily check-ins, regular meal intake, and the frequency of physical interactions with the robot.

DISCUSSION: The preliminary findings suggest that Hyodol-based monitoring may serve as a viable screening tool for detecting depression risk in older adults. Future work should focus on refining the model to replicate current results while minimizing false alarms, incorporating more in-depth content analysis, and developing continuous emergency monitoring and alert systems to enhance clinical utility.