Artificial Intelligence for Affective-Domain Development in Healthcare Professions Education: A Systematic Review

Synthesis: Ismail and Zafira (2026) systematically reviewed 17 empirical studies published between 2010 and May 2026 on AI for affective-domain development, including Empathy, in health professions education, searching eight databases and appraising quality with Joanna Briggs Institute tools. The interventions cluster into communication training, virtual patients, reflective feedback, breaking-bad-news simulation, interview rehearsal and OSCE-style assessment. Most studies report gains in confidence, self-efficacy, reflective engagement, perceived communication skill or simulation performance — and the review's central finding is that evidence for the deeper affective outcomes, empathy, emotional responsiveness, relational authenticity and non-verbal communication, remains limited. The authors conclude that AI should complement rather than replace human-facilitated teaching.

Relevance to AI in Education: A compact map of what AI is currently doing in health professions training for affect, and of the gap between the outcomes that are easy to measure and the ones the field actually cares about. It is the best single starting point for anyone designing an AI-supported empathy intervention in medicine or nursing.

Key Findings

• The literature is small and heterogeneous. Seventeen studies across undergraduate and postgraduate learners, spanning a decade and a half, with varied designs, tools and outcome measures — enough for a narrative synthesis, not a meta-analysis.
• Confidence and perceived skill dominate the outcomes. Most studies measured learner confidence, self-efficacy, reflective engagement, perceived communication skills or simulation performance.
• Empathy itself is rarely demonstrated. The review finds limited evidence for empathy, emotional responsiveness, relational authenticity and non-verbal communication as outcomes.
• AI's advantages are logistical. Repeated low-stakes practice, immediate structured feedback, accessibility and scalability recur across studies.
• The recurring concerns are realism and dependence. Authors report worries about emotional realism, over-reliance on AI, and reduced interpersonal authenticity.

What the review covers

Searches ran in PubMed, Scopus, Web of Science, SpringerLink, ScienceDirect, Google Scholar, Wiley Online Library and IEEE Xplore for English-language empirical studies from January 2010 to May 2026. Eligible studies involved healthcare professional learners, used an AI-based educational tool for teaching, simulation, feedback, communication training, reflection or assessment, and reported at least one affective-domain outcome. The protocol was registered with PROSPERO (CRD420261401425) and reporting followed PRISMA 2020. Quality was appraised with JBI critical appraisal tools, with CASP as a supplementary aid for qualitative and mixed-methods work.

The included studies range from small feasibility evaluations to cohort studies of around a hundred learners. Typical tools are ChatGPT-based simulated patients with automated feedback, LLM virtual patients with voice, and OSCE-style AI assessment of history taking. Reported effects are mostly student-reported: high usability, psychological safety, motivation, and increased comfort with history taking. One study of 24 students generating 93 survey responses reported a 14.6% increase in comfort with history taking; another with 106 students found 69% felt a virtual patient increased their confidence.

Why the outcome gap matters

Affective-domain learning is the part of professional education where the measurement problem is hardest. Confidence is cheap to collect and responsive to almost any intervention; empathy is expensive to observe and contested in definition. A literature that reports the first while intending the second will look more successful than it is.

The review is careful about this, which is its main virtue. It does not claim that AI teaches empathy; it claims that AI supports communication rehearsal, reflective learning and formative feedback, and that educators remain essential for empathy, ethical judgment, professional identity and relational care. That is a narrower and more defensible position than much of the primary literature.

The concern about emotional realism cuts in the same direction. If a simulated patient cannot convey fear, embarrassment or ambiguity, then practice with it trains the student to handle a simplified encounter. The skills may transfer, but the review offers no evidence that they do.

What this means for practice

• Use AI where the evidence is strongest: repeated rehearsal, immediate structured feedback, and low-stakes practice before real or standardized patients.
• Do not report self-efficacy as empathy. If the outcome is affective, measure it with observation, standardized patients or patient ratings, not only a confidence scale.
• Keep a human in the loop for the relational parts. The review's conclusion is complementarity, not substitution, and the studies it cites give no basis for replacing facilitated debriefing.
• Design for transfer. Voice, avatar and scenario realism are the levers the primary studies identify, and none of them has been tested against downstream clinical behavior.

Limitations

• No meta-analysis: designs, interventions, populations and outcomes were too heterogeneous for pooling, so the synthesis is narrative and cannot estimate an effect size.
• Restrictions to English-language, empirical studies may exclude relevant work in other languages and unpublished evaluations.
• Most included studies are small and short-term, and few report delayed outcomes or behavioral measures.
• The review reports limited evidence on empathy but does not quantify how many of the 17 studies attempted to measure it.
• Affective-domain outcomes are defined broadly across the included studies, which makes "improvement" mean different things in different rows of the evidence table.

Citation

Ismail, N. A. S., & Zafira, A. (2026). Artificial intelligence for affective-domain development in healthcare professions education: A systematic review. Frontiers in Medicine, 13, 1894820.