JMIR Publications

ABSTRACT

Background:

In response to the growing demands of chronic conditions, artificial intelligence (AI)-enhanced remote consultations indicate a promising avenue. However, reports detailing the co-design of assistive AI tools for remote chronic care remain limited. In addition, digital tools that fail to integrate end-user perspectives have been associated with inefficient adoption. As such, the development of AI-assisted virtual consultation tools would benefit from end-user involvement, such as through co-design.

Objective:

To report on the co-design of CARA (Consultation Analysis & Response Assistant), a novel, evidence-based AI prototype to assist in remote chronic care, and explore end-users’ perceptions of the tool.

Methods:

Four participants, including two patients and two healthcare professionals (HCPs), with remote chronic care experience from the public health sector in rural Northwest Ireland, were involved in three iterative co-design workshops. These included self-paced, hands-on sessions with CARA, and interactive group feedback and brainstorming. Discussions were analysed using reflexive thematic analysis to identify themes around usability and acceptability perspectives. The first author led the development of CARA, which relied on natural language processing techniques and incorporated participant feedback during iterative development cycles. The prototype's backend was developed in Python, interfacing with large language models analysing synthetic but representative virtual consultation transcripts, and this AI-generated output would be parsed onto an HTML frontend.

Results:

Incremental changes were made to both the frontend and backend of CARA, based on participants’ feedback. This study has shared the evidence-based features of this bespoke AI prototype, which include patient- and HCP-facing consultation summaries, patient access to additional resources relevant to their condition, familiarising HCPs with a patient’s medical history with a view to improving rapport building, and analysing patients’ sentiments to provide insights for HCPs to improve the interaction quality. This study also provided practical details around the co-design process structure for future reference. Five themes emerged from the discussions, namely: System-level challenges, Remote assessment and interaction assistance, Trust and human oversight over powerful AI assistance, Holistic patient view with emotional dimension, and Usability and accessibility of outputs.

Conclusions:

There was a general acceptance of CARA, despite concerns over the systemic requirements for its implementation in practice. The agile, user-centred development resulted in software that addresses the pertinent needs and preferences of end-users, ensuring that the AI meaningfully supports - rather than undermines - both patient experience and clinical practice. The findings lay the groundwork for subsequent research and eventual practical uses.