Continuous ADL monitoring using computer vision to maintain independence and improve HRQoL in older adults at risk for AD/ADRD - To help older adults age independently at home, effectively monitoring and detecting changes in ADLs are critical for preventing adverse events and maintaining health-related quality of life (HRQoL). However, ADLs are time consuming to capture, highly subjective, and rarely documented in most clinical encounters. Artificial intelligence (AI) computer vision is capable of automatically capturing a continuous timestream of activities and may address these limitations, yet has been criticized for the “blackbox” nature of algorithms. Our preliminary data identified that a unique AI approach using computer vision can capture ADLs without large tagged datasets to learn a behavior while preserving privacy. Our hypothesis is that ADL-related data captured by an explainable AI monitoring system can be a key contributor to preventing ADL-deficit associated adverse events and maintaining HRQoL among individuals with Alzheimer's Disease and Alzheimer's Disease Related Dementias (AD/ADRD) residing in home settings. In the proposed work, we will develop (R21) and assess (R33) a highly personalized and clinically interpretable AI system, known as Cherry AI, to monitor ADLs, detect changes early, predict relevant adverse events, and support healthcare planning for Program for All- inclusive Care for the Elderly (PACE) providers, with an ultimate goal of maintaining HRQoL among PACE enrollees with or without dementia. In the R21 phase (Stage 0), we will refine Cherry AI algorithms and conduct focus groups of PACE clinicians to identify and summarize factors involved in clinical management plans for ADLs. We will enroll PACE enrollees with a history of ADL deficits and varied cognitive profiles [total n=20, 10 w/ mild cognitive impairment; 10 w/ subjective cognitive decline] and monitor ADLs in homes using Cherry AI. PACE clinicians will evaluate participants’ ADLs using the Modified Barthel Index. Correlations between Cherry AI-measured and clinician-rated ADLs will be evaluated. Qualitative focus groups of 10-15 home care clinicians will be used to improve the Cherry AI interface. Specific aims include (1) refining Cherry AI algorithms and (2) enhancing interpretability of the Cherry AI system to help clinicians make ADL related management plans. In the R33 phase (pilot test, Stage I), we will assess the ability of Cherry AI to help maintain or improve HRQoL in PACE enrollees with AD/ADRD by predicting future changes in ADLs and associated adverse events, and assisting with ADL-related management. PACE enrollees (n=80) with a history of ADL deficits will be stratified on cognitive phenotype and randomly assigned to one of two groups: Cherry AI (intervention) vs. usual care (control) in a pilot single-blind randomized controlled trial. We will use linear mixed- effect models to examine Cherry AI’s effect on maintaining HRQoL compared to PACE’s usual care. Specific aims include comparing changes of HRQoL, incidence of adverse events, and changes in PACE management plans between groups. This study will lead to an efficacy trial of Cherry AI monitoring to improve HRQoL for community-dwelling seniors with AD/ADRD.
