Quality Assessment of Large Language Model–Generated Medical Dialogue for Clinical Vignettes: Evaluation Study
Bibliographic record
Abstract
BACKGROUND: Traditional clinical vignettes, though widely used in medical education, often focus on prototypical presentations; require substantial time and effort to develop; and fail to represent patient diversity, the complexity of clinical conditions, patients' perspectives, and the dynamic nature of physician-patient interactions. OBJECTIVE: This study aimed to evaluate the quality of Japanese-language physician-patient dialogues produced by generative artificial intelligence (AI), focusing on their medical accuracy and overall appropriateness as medical interviews. METHODS: We created an AI prompt that included a specific clinical history and instructed the model to simulate a cooperative patient responding to the physician's questions to generate a physician-patient dialogue. The target diseases were those covered by the Japanese National Medical Licensing Examination. Each dialogue consisted of 25 turns by the physician and 25 by the patient, reflecting the typical volume of conversation in Japanese outpatient settings. Three internists independently evaluated each generated dialogue using a 7-point Likert scale across 6 criteria: coherence of the conversation, medical accuracy of the patient's responses, medical accuracy of the physician's responses, content of the medical history, communication skills, and professionalism. In addition, a composite score for each dialogue was calculated as the overall mean of these 6 criteria. Each dialogue was also examined for the presence of 5 essential clinical components commonly included in medical interviews: chief concern and clinical course since onset, physical findings, test results, diagnosis, and treatment course. A dialogue was considered to include a component only if all 3 evaluators independently confirmed its presence. RESULTS: The mean composite score was 5.7 (SD 1.0), indicating high overall quality. Mean scores for each criterion were as follows: coherence of the conversation, 5.9 (SD 0.9); medical accuracy of the patient's responses, 6.0 (SD 0.9); medical accuracy of the physician's responses, 5.6 (SD 1.1); content of medical history taking, 5.9 (SD 0.9); communication skills, 5.6 (SD 0.9); and professionalism, 5.5 (SD 1.1). Among the 5 clinical components assessed in each dialogue across 47 clinical cases, chief concern and clinical course were included in all 47 (100%) cases, physical findings in 15 (32%) cases, test results in 27 (57%) cases, diagnosis in 45 (96%) cases, and treatment course in 0 (0%) cases. CONCLUSIONS: While physician oversight remains essential, it is feasible to efficiently create AI-generated educational materials for medical education that overcome the limitations of traditional clinical vignettes. This approach may reduce time and financial burdens, enhancing opportunities to practice clinical interviewing in settings that closely mirror real-world encounters.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.058 | 0.271 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.004 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.003 | 0.002 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".