Project 06 · 4 publications

Aging, Smart Health & Cyber-Physical Systems

Sensor-based systems for detecting social isolation and loneliness in elderly communities, fall risk evaluation from electronic health records, and multimodal gait and sleep sensing — developed during postdoctoral research at the University of Pennsylvania in collaboration with clinical and engineering teams.

GerontechnologySocial IsolationFall RiskEHR DataAmbient SensingClinician-in-the-LoopACM/IEEE CHASEICCPS
Overview

Technology for Healthy Aging

Social isolation and loneliness among elderly adults carry health risks rivaling those of smoking and obesity, yet they remain difficult to detect and intervene upon at scale. This research thread, conducted at the University of Pennsylvania in collaboration with clinicians, nursing informaticists, and international teams, developed cyber-physical systems for passive, unobtrusive monitoring of older adults — enabling early detection of social isolation, prediction of fall risk from clinical data, and correlation of mobility signals with health outcomes.

A common theme is the clinician-in-the-loop design philosophy: sensor systems and data-driven models are built to support clinical decision-making rather than replace it, enabling scalable, actionable health interventions in community-dwelling and inpatient settings.

iCareLoop system: Proposes a closed-loop architecture where passive sensing of daily activity patterns feeds a machine learning pipeline, and outputs trigger clinician-mediated interventions — forming a complete loop from detection through actuation for gerontological social isolation and loneliness.

22Elderly participants (US & Japan)
MIMIC-IIIEHR database used (FRED)
4Publications
Research Threads

Three Interconnected Aging-Health Problems

Social Isolation & Loneliness Detection

Passive ambient sensing of daily activity patterns (motion, sleep, social interaction proxies) in elderly homes, combined with machine learning, to predict loneliness scores and trigger clinician-mediated interventions.

Fall Risk Evaluation from EHR

Open-source database (FRED) derived from MIMIC-III integrating Morse Fall Scale records, physician notes, and medication data — enabling data-driven fall risk prediction and prevention research.

Gait & Sleep Sensing Correlation

Pilot study exploring correlations between gait parameters measured by wearable sensors and bed movement signals from smart bed systems — toward unobtrusive, continuous mobility monitoring for older adults.

International Collaboration

Cross-cultural deployment of sensing systems in US and Japan communities, studying loneliness indicators across different social contexts through a US–Japan research partnership.

Publications
ACM/IEEE ICCPS · CPS-IoT Week 2023
iCareLoop: Closed-Loop Sensing and Intervention for Gerontological Social Isolation and Loneliness
X. Ji, A. Yuh, H. Choi, A. Watson, C. Kendell, X. Li, J. Weimer, H. Nagahara, T. Higashino, T. Mizumoto, V. Erdélyi, G. Demiris, O. Sokolsky, I. Lee
Proposes iCareLoop, a closed-loop decision support system with clinician-in-the-loop actuation to mitigate gerontological social isolation and loneliness. Passive sensors deployed in elderly residences capture daily activity data; machine learning models predict social isolation and loneliness levels; targeted interventions are triggered through clinical workflows. Addresses mental health disruptions significantly worsened by the COVID-19 pandemic.
IEEE/ACM CHASE · 2023
Short: Integrated Sensing Platform for Detecting Social Isolation and Loneliness in the Elderly Community
X. Ji, X. Li, A. Yuh, C. Kendell, A. Watson, J. Weimer, H. Nagahara, T. Higashino, T. Mizumoto, V. Erdélyi, G. Demiris, O. Sokolsky, I. Lee
Presents an integrated sensing platform deployed in the residences of elderly community-dwellers to passively detect social isolation and loneliness. The platform captures daily living activity patterns through a suite of ambient sensors and applies machine learning to correlate sensor features with validated loneliness scales — demonstrating the feasibility of technology-based detection as a scalable alternative to clinic-based assessment.
IEEE/ACM CHASE · 2021
FRED: Fall Risk Evaluation Database Based on Electronic Health Record Data
P. Lu, X. Li, S. Jang, A. Lee, S. Pugh, A. Watson, R. I. Bjarnadóttir, R. Lucero, G. Demiris, A. Nenkova, J. Weimer, I. Lee
Introduces FRED, an open-source four-part database for fall risk evaluation derived from MIMIC-III EHR data. Part 1: demographic data and admission IDs. Part 2: Morse Fall Scale intervals and fall risk labels (high/low/increasing/decreasing). Part 3: corresponding physician notes. Part 4: administered medication records. Available on PhysioNet to enable data-driven fall prediction and prevention research across interdisciplinary aging studies.
MidAtlantic Bioinformatics Conference · 2021
Pilot Study to Explore Correlation Between Gait and Bed Movements
X. Li, J. Weimer, G. Demiris, O. Sokolsky, I. Lee
A pilot study investigating the correlation between gait parameters measured by wearable inertial sensors and bed movement signals captured by a smart bed monitoring system in older adults. Explores whether unobtrusive overnight bed sensor data can serve as a proxy for mobility and gait health — enabling continuous longitudinal tracking without requiring active cooperation from participants or wearable compliance during the day.