Development and Pilot Testing of an Artificial Intelligence and Care Coach Intervention for Cognitively Impaired Older Adults

Background Care transitions from the emergency department (ED) to community settings are particularly challenging for persons living with cognitive impairment (PLWCI) and their caregivers. The chaotic ED environment and limited post‐discharge support contribute to poor outcomes and high caregiver burden. This study aimed to develop and pilot test an intervention integrating artificial intelligence (AI) technology and care coaching to enhance post‐ED support among PLWCI and their caregivers. Methods We conducted a three‐phase study using a user‐centered approach. Phase 1 involved focus groups with key informants to explore care transition experiences. Phase 2 included a design thinking workshop with PLWCI, caregivers, clinicians, and technology experts to co‐create the intervention. The final intervention combined a 5‐part AI technology application (termed NeuViCare)—planner, task support, resource advisor, care advisor, and community hub—with access to a care coach trained in occupational therapy or nursing. In Phase 3, we pilot tested the intervention in four EDs, measuring caregiver self‐efficacy (Fortinsky Caregiver Self‐Efficacy scale) and burden (4‐item Zarit Caregiver Burden scale). Results Focus groups including 15 participants highlighted major ED and transition challenges, emphasizing the need for better communication and support. Workshop participants ( n = 23) identified key intervention components. In the pilot, 40 participants enrolled, with 34 completing the 7‐day follow‐up and 29 completing the 30‐day follow‐up. Engagement with NeuViCare was high, with caregiver self‐efficacy improving from 52.0 at Day 0 to 65.8 at Day 30, and caregiver burden decreasing from 9.2 to 8.2. Conclusion Integrating AI‐driven support with human care coaching shows promise in improving ED care transitions for PLWCI and their caregivers. The intervention enhanced caregiver self‐efficacy and modestly reduced burden, suggesting potential for broader implementation.