A Collaborative Approach to Support Medication Management in Older Adults with Mild Cognitive Impairment Using Conversational Assistants (CAs)
Niharika Mathur, Kunal Dhodapkar, Tamara Zubatiy, Jiachen Li, Brian Jones, Elizabeth Mynatt · 2022 · Proceedings of the 24th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS '22) · doi:10.1145/3517428.3544830
Summary
This paper presents MATCHA (Medication Action To Check-In for Health Application), a conversational medication management system built as a Google Action for the Google Home Hub, designed specifically for older adults with Mild Cognitive Impairment (MCI) and their caregivers (referred to as "carepartners"). MCI represents an intermediate stage of cognitive decline between normal aging and dementia, affecting approximately 16.6% of people over 65, with symptoms including memory loss, language problems, and attention difficulties. Medication management is particularly challenging for this population because traditional alarm-based reminders can actually increase the risk of accidental over-medication — when reminded to take medication, a person with MCI may not remember whether they have already taken it and may take it again. The research followed an extensive participatory design process within a hospital-based comprehensive cognitive program. Two focus groups with 18 member-carepartner dyads and a scenario-based design session with 18 participants revealed that users strongly preferred a system that "checks in" rather than "reminds." This distinction is crucial: rather than telling users to take medication, MATCHA asks whether they have taken it, prompting reflection and self-checking behaviour. The design sessions also uncovered preferences for positive reinforcement after confirming medication, multiple notification channels (phone alerts to carepartners when the member is unavailable), personalized greetings and volume settings, and integration with existing habits like pillbox use rather than replacing them. MATCHA was deployed in two phases over 20 weeks with 7 dyads. The system recognized over 200 possible responses and handled scenarios including already-taken medication (positive feedback), not-yet-taken (offer to reschedule), uncertain recall (prompt to check pillbox), and no response (phone notification to both member and carepartner). Phase 1 lasted 4 weeks, after which design revisions were made based on usage data and interviews before deploying Phase 2 for 16 weeks.
Key findings
The check-in approach generated sustained and increasing engagement over the 20-week deployment. The weekly engagement rate rose from approximately 18% in Phase 1 to 67% in Phase 2, with a continued upward trend reaching around 85-90% by the final weeks. This increase was attributed to design revisions (reducing overly enthusiastic positive feedback, adding touch-screen buttons as an alternative to voice, incorporating a "Taking Now" response scenario) and growing comfort with the system. The system achieved strong acceptance scores — SUS scores of 84.66 from members and 86.16 from carepartners in Phase 1. All five dyads who completed Phase 2 expressed desire to continue using MATCHA after the study, and were still using it approximately 55 weeks later at the time of writing. Members reported that MATCHA prompted them to reflect on whether they had actually taken their medication before responding, rather than reflexively acknowledging an alarm. In some cases, this reflection led them to check their pillbox, catching instances where they had not yet taken medication or had potentially already taken it. Carepartners valued receiving phone notifications about their partner's medication status, particularly when they were away from home. The multimodal interaction — combining voice and touch-screen buttons — was appreciated, with touch-based interactions accounting for 20-35% of total interactions. This provided an important fallback when speech recognition failed or when members had difficulty being understood. Personalization of greeting style, volume, and timing based on individual medication schedules and household layouts contributed to a sense that the system understood their routines.
Relevance
This research offers a compelling model for designing assistive technology that supports gradual cognitive decline rather than just addressing a fixed disability state. The distinction between "checking in" and "reminding" is a nuanced but impactful design insight applicable beyond medication management — any system supporting individuals with memory difficulties should consider how its framing affects user agency, self-efficacy, and the risk of error. The finding that traditional reminders can actually increase over-medication risk challenges common assumptions in assistive technology design. The collaborative dyad model — designing for both the person with MCI and their caregiver as a unit — is particularly relevant for accessibility practitioners working with aging populations. The study demonstrates that conversational assistants can serve as an effective intermediary layer in caregiving relationships, providing the person with MCI a sense of independence while giving caregivers peace of mind through notifications. The emphasis on integrating with existing habits (pillboxes, routines) rather than replacing them reflects an assets-based design approach that respects users' existing competencies and reduces adoption barriers.
Tags: mild cognitive impairment · older adults · medication management · conversational assistants · smart speakers · voice interface · aging · caregiving