An AI-based empathic virtual companion to track changing health states in persons living with dementia

PROJECT ABSTRACT Mentia, a company focused on improving the lives of people with dementia through meaningful engagement with interactive technology, addresses the health inequity of the nation’s 6.8 million dementia population compared to non-ADRD older adults. This group is twice as likely to have multiple chronic conditions than non-AD/ADRD older adults, and health management is complicated by memory, communication, and sensing deficits, contributing to a 2.5 times greater likelihood of visiting Emergency Departments. Once there, patients commonly experience disorientation and anxiety in a stressful, unfamiliar environment, and may experience polypharmacy, excessive testing, and potential hospital admission, which may hasten their cognitive and functional decline. These cascading effects come at great cost to patients, their families, and the healthcare system, with Medicare alone spending $100 billion annually more on seniors with dementia than seniors without. To stem the humanistic and economic crises, we propose a new system for the early detection of acute health events that enables patients to receive timely clinical care in place. Our solution, known as MentiaCompanion™, leverages AI and Mentia’s deep understanding of the engagement process for people who live with AD/ADRD, to track the person’s health state, enabling connected care and clinical health teams to assess risk. The system, to be developed in two phases, is a technology for data collection (Phase I) and the alignment of data sets with health risk assessment (Phase II). Preparatory work on this solution was undertaken in 2021-23 via an NIH SBIR grant and an NIA subaward via Johns Hopkins University's Artificial Intelligence and Technology Collaboratory for Aging (AITC). The goal of Phase I is to achieve frequency of use. Based on the principle that high engagement equates to frequency, Mentia Companion will run autonomously to relieve caregiver load, while empathic and knowledgeable on-screen avatars will hold informed two-way conversations by processing speech and emotion recognition, life stories, and case notes. Mentia Companion will serve both verbal and non-verbal users, exceeding usual chatbot offerings. Further engagement will be achieved through avatar-led activities within a dementia-friendly virtual world. The avatars will guide users through media-rich environments containing meaningful interactive stimuli based on non-pharmacologic approaches to care. A user experience (UX) study to identify optimal ways to start and complete sessions autonomously, along with AI technology development, will culminate in an AD/ADRD cohort study to establish that we have the frequency of use needed to build the data sets in the Phase 2 work that follows.