MedEasy: Designing AI Standardized Patients for Clinical Consultation Training

Synthesis: Gao et al. (2026) present MedEasy, a multi-agent system that simulates standardized patients with varying conditions for medical consultation training. By organizing virtual-patient practice through patient dialogue, clinical actions, decision submission, documentation and feedback, it outperforms script-based approaches in realism and adaptability — supporting clinical training with Feedback-rich practice.

MedEasy multi-agent system simulates standardized patients with varying conditions for medical consultation training, outperforming script-based approaches in realism and adaptability. AI standardized patients are becoming a setting for professional training in clinical consultation. This paper presents MedEasy, a multi-agent system that organizes virtual-patient practice through patient dialogue, clinical actions, decision submission, documentation, and feedback. Its realism and adaptability make it a valuable tool for AI-guided clinical learning, connecting to affective and equity considerations in medical education.

What this means for practice

• Software developers. Keep intent recognition, case-grounded response generation, and post-session evaluation as separate components, as MedEasy does, so errors in any one stage stay traceable and reviewable against the case record.
• Software developers. Encode missing case states explicitly — a clinically negative fact, information unknown to the patient, an item not yet assessed, and information absent from the authored case — because treating all missing fields as negative or normal injects unsupported clinical information into the simulated patient.
• Instructors. Use AI standardized patients as repeatable rehearsal alongside educators, human SPs, and clinical encounters rather than as a replacement: the system became less convincing as expectations moved toward procedural skill, emotional response, or final judgment on disputed standards.
• Instructors. Schedule practice so learners can repeat a case and compare a new attempt with earlier feedback — participants described rehearsing the order of a consultation before supervised practice and returning to missed steps after teaching as the system's value.
• Administrators. Keep AI-SP practice formative: the Evaluation Agent follows the supplied expert answer rather than comparing competing clinical guidelines, and agreement with generated comments does not establish validity as a measure of clinical competence.

Limitations

• Qualitative and interpretive: the evaluative study involved 12 clinical-year medical students (P1–P12) in a single individual session of approximately 90–120 minutes, so the study does not estimate learning effects or establish that MedEasy improves clinical competence.
• Only two cases were used; both received clinical review, but their complete evaluation criteria were not independently validated as competence assessments.
• MedEasy's predefined intent framework represents mainly history-taking inquiries, maps each utterance to no more than three categories, and collapses ambiguous or unrecognized inputs into a small-talk category.
• Exposure was individual and time-limited, so repeated course use, educator-led debriefing, and group teaching around system traces were not observed.

Citation

Zhiqi Gao, Huarui Luo, Guo Zhu, Bingquan Zhang, Dongyijie Primo Pan, Yizhan Feng, Jiahuan Pei, Jie Li, Benyou Wang (2026). MedEasy: Designing AI Standardized Patients for Clinical Consultation Training.