JMIR Publications

ABSTRACT

Background:

Advance care planning (ACP), goals-of-care discussions, and serious illness communication are underutilized in routine clinical practice despite well-established benefits for patient-centered, end-of-life care. Artificial intelligence (AI) technologies are increasingly explored as tools to support these communication processes, yet prior reviews in this space have focused primarily on AI-based prediction and patient identification rather than communication itself.

Objective:

This scoping review aimed to map the emerging literature on AI-supported communication in serious illness and end-of-life care, characterizing the types of AI applications used, how they are integrated into communication processes, and their reported outcomes and implementation challenges.

Methods:

Following PRISMA-ScR guidance, PubMed, CINAHL, PsycINFO, Web of Science, IEEE Xplore, and ACM Digital Library were searched from inception to March 31, 2026. Studies were eligible if they evaluated an AI-based technology directly engaging patients, families, or clinicians in serious illness or end-of-life communication.

Results:

Twelve studies published between 2017 and 2025 met inclusion criteria, predominantly conducted in the United States. Three themes were identified: (1) AI for identifying patients and initiating serious illness conversations, most often through machine learning–based mortality prediction models embedded in electronic health records to trigger clinician prompts or care coach outreach (7 studies); (2) conversational AI and virtual agents supporting patient-facing ACP engagement and values clarification (3 studies); and (3) AI-based platforms for clinician communication-skills training (2 studies). The prediction- and identification-focused literature was comparatively mature, with large-scale randomized and stepped-wedge trial evidence, while the conversational AI and training literature remained smaller and methodologically less developed.

Conclusions:

AI is being applied across at least three distinct points in the serious illness communication process. Substantially more evidence supports its use in identifying and prompting conversations than in directly facilitating or training for them. Future research should prioritize rigorous evaluation of AI systems that directly engage patients and clinicians in communication itself, alongside continued development of prediction-based approaches.