This study evaluated whether an emotionally and interactionally adaptive real-time voice virtual patient can improve clinical communication training for resident physicians. Communication with patients in emotionally challenging situations, such as breaking bad news, requires clinicians to deliver information clearly, recognize emotional cues, respond empathically, and maintain a coherent interaction. Standardized patient training can support these skills but is resource intensive and difficult to provide repeatedly. Real-time voice virtual patients may offer scalable practice opportunities, but simple spoken interaction alone may not provide patient responses that change meaningfully according to the learner's communication.

The adaptive virtual patient used in this study was designed to update the patient's emotional tone, openness, defensiveness, and interactional style during the conversation based on the learner's communication behavior. The study compared this adaptive virtual patient training with a non-adaptive real-time voice virtual patient and with standardized patient training.

Resident physicians were randomly assigned in a 1:1:1 ratio to one of three groups: emotionally and interactionally adaptive virtual patient training, non-adaptive virtual patient training, or standardized patient training. Participants completed two communication training sessions focused on breaking bad news. After training, participants completed learner-reported measures of perceived response contingency and patient realism. Approximately four weeks later, participants completed a mini-objective structured clinical examination using a human standardized patient to assess communication performance in a delayed near-transfer setting.

The primary outcome was the total mini-OSCE communication score at four weeks. Secondary outcomes included mini-OSCE domain scores for structured delivery, emotional responsiveness, and interactional process; learner-reported response contingency and patient realism; and expert-rated behavioral consistency and clinical credibility of virtual patient interactions.

Inclusion Criteria:

• Resident physicians in postgraduate years 1-3.
• Participants who had received basic clinical communication training or related instruction.
• Participants who had not received systematic, specialized training in breaking bad news communication.
• Able to complete two training sessions and the four-week follow-up mini-OSCE assessment.
• Willing to participate and able to provide written informed consent.

Exclusion Criteria:

• Prior systematic training in breaking bad news, advanced communication skills, or similar specialized communication training.
• Prior participation in pilot testing or system testing related to this study.
• Unable to complete the required training sessions or follow-up assessment.
• Unwilling to have training or assessment sessions recorded for research scoring and analysis.
• Concurrent participation in another communication-skills intervention study that could affect communication performance.
• Any other condition judged by the investigators to make participation inappropriate.

Individual participant data will not be publicly shared because participant consent and institutional ethics approval do not permit unrestricted public release of human-subject data. Summary data and study materials sufficient to interpret the findings will be provided in the publication and supplementary materials. Additional deidentified data may be made available from the corresponding author upon reasonable request, subject to institutional and ethics approval.

Participants and training facilitators could not be masked because the intervention modalities were visibly different. Outcome assessors who rated the four-week mini-OSCE recordings were masked to group assignment. Expert raters evaluating virtual patient interaction quality were also masked to condition where applicable.