Half an hour with an AI patient improved simulated serious-illness conversations

Half an hour of conversation with a virtual patient improved how clinicians and trainees handled a subsequent simulated discussion about serious illness. In a 51-person randomised trial, those who practised with the SOPHIE system made greater gains than participants who spent the same time reading about the communication skills. The result was assessed through conversations with human actors, not by asking the AI to grade its own performance.[1]

SOPHIE stands for Standardized Online Patient for Healthcare Interaction Education. The University of Rochester-led team designed it to combine dialogue with structured feedback on three skills: empathising, being explicit and empowering patients. Each gives the learner something concrete to practise in a conversation where choosing the right words can be difficult.[1]

Thirty minutes, two ways of learning

Fifty-one people enrolled and completed the randomised study, including medical, nursing and physician-assistant students and practising clinicians. The intervention group spent 30 minutes working through three interactive training modules with SOPHIE. The themes were empathising, being explicit and empowering patients. The control group spent an equivalent period reading material on the same framework. The allocation was balanced by training level, and the trial used videoconferencing rather than an in-person hospital ward.[1]

Each participant held two conversations with trained simulated patients, one before and one after training. One actor directly involved and four additional blinded reviewers assessed the recordings, using a rubric to judge how well the participants approached the three communication tasks. Altogether, the paper reports 102 simulated conversations and 506 of an expected 510 ratings. Agreement among the blinded raters was strong. That helps distinguish a measured learning signal from the research team's enthusiasm for its own system.[1]

What improved, and by how much

Both groups improved after their allocated exercise, which makes sense: even focused reading and a second rehearsal can help. The AI-practice group, however, improved more. Average overall rubric scores rose from 0.52 to 0.67 in that group, compared with 0.59 to 0.65 after reading alone. In a mixed-effects analysis, the extra gain attributed to the SOPHIE arm was 0.09 on the normalized scoring scale, with a 95% confidence interval from 0.04 to 0.14 and a P value of .001.[1]

The individual dimensions moved in the same direction. Extra gains were estimated at 0.11 for empowering the patient, 0.08 for being explicit and 0.07 for empathising, although the uncertainty bands differed. Participants also rated their own improvement in confidence more highly after SOPHIE: 4.31 versus 3.83 on a five-point scale. Confidence, though useful to track, is not the same as demonstrated skill. The standardized-patient assessments form the stronger part of this evidence.[1]

Practice has to cross the gap into care

The trial tested a teaching intervention: rehearsal and automated feedback against reading, followed by an immediate assessment. It did not compare SOPHIE with an equal amount of actor-led practice or measure patient outcomes. The next educational question is whether the gains persist and transfer to real consultations, where an anxious patient’s responses rarely fit a short training scenario.[1]

The practical attraction is repetition. Learners could practise explaining a difficult situation, acknowledge an emotional response and try a clearer approach before another person’s care depends on the conversation. The study gives that idea a measurable starting point: better performance on a blinded assessment after a brief session. Its lasting value will depend on what learners carry from the rehearsal into the room.

Report an error or suggest a correction