Validating the Apple Watch for Passive Monitoring of Agitation in Patients with Dementia

Project Accomplishments: This pilot project focuses on validating a machine learning algorithm designed to detect agitation in individuals with Alzheimer’s disease and related dementias using accelerometric data from an Apple Watch. The team finalized the study protocol, secured IRB approval, and developed a custom Apple Watch app to collect sensor data and caregiver‑reported agitation episodes. Recruitment efforts enrolled 17 patient–caregiver dyads, each trained using standardized UCLA Alzheimer’s and Dementia Care Program modules to ensure accurate identification of agitation events. Although seven participants withdrew for reasons including illness, decreased agitation, or death, the project successfully deployed continuous data collection and established a feasible workflow for real‑world monitoring. 

The project’s objectives—validating the algorithm’s sensitivity and specificity and comparing its performance to standardized caregiver‑reported measures—remain unchanged. While full data analysis is pending, early outcomes demonstrate strong feasibility: caregivers can reliably use the technology, recruitment is achievable, and the Apple Watch platform can support continuous agitation monitoring. Prior work with a similar device achieved high accuracy, and this study aims to replicate those results on a more accessible consumer device. Additionally, the pilot has laid the groundwork for larger validation trials and suggests potential expansion to other commercial wearables, broadening future clinical and research applications. 

Initial Proposal Abstract: This project seeks to improve care for older adults with Alzheimer’s disease and related dementias (ADRD). Specifically, we plan to validate a novel machine learning technique that analyzes motion data from an accelerometer in an Apple Watch to identify the onset of an agitation episode in a person with ADRD. 

Validating this technology offers the potential to provide caregivers with an advance warning system that allows them to better manage agitation at home, provide clinicians a tool to improve clinical decision-making, and provide clinical trials developing pharmacologic or other therapies for dementia a better outcome measure.

This study plans to recruit 25 patient/caregiver dyads for approximately 2 months each. This study has two Specific Aims: (1) Demonstrate the algorithm can achieve a specificity > 0.95 and a sensitivity > 0.80 and (2) Demonstrate that the use of the algorithm reveals a weakness in the use of standardized assessments that are based on caregiver recall of events over the recent past. 

Patients will be equipped with an Apple Watch, and the algorithm will provide predictions to caregivers in real-time via a mobile app. Caregivers’ observations will be matched to the algorithm’s predictions to determine its specificity and sensitivity. Caregivers will be asked to fill out the Cohen-Mansfield Agitation Inventory (CMAI-C) and Neuropsychiatric Inventory Questionnaire (NPI-Q) at the beginning and end of the study period. Discrepancies and the direction of those discrepancies will be used to support our second Specific Aim.