Physician vs. AI-generated messages in urology: evaluation of accuracy, completeness, and preference by patients and physicians

To evaluate the accuracy, comprehensiveness, empathetic tone, and patient preference for AI and urologist responses to patient messages concerning common BPH questions across phases of care.

Cross-sectional study evaluating responses to 20 BPH-related questions generated by 2 AI chatbots and 4 urologists in a simulated clinical messaging environment without direct patient interaction. Accuracy, completeness, and empathetic tone of responses assessed by experts using Likert scales, and preferences and perceptions of authorship (chatbot vs. human) rated by non-medical evaluators.

Five non-medical volunteers independently evaluated, ranked, and inferred the source for 120 responses (n = 600 total). For volunteer evaluations, the mean (SD) score of chatbots, 3.0 (1.4) (moderately empathetic) was significantly higher than urologists, 2.1 (1.1) (slightly empathetic) (p < 0.001); mean (SD) and preference ranking for chatbots, 2.6 (1.6), was significantly higher than urologist ranking, 3.9 (1.6) (p < 0.001). Two subject matter experts (SMEs) independently evaluated 120 responses each (answers to 20 questions from 4 urologist and 2 chatbots, n = 240 total). For SME evaluations, mean (SD) accuracy score for chatbots was 4.5 (1.1) (nearly all correct) and not significantly different than urologists, 4.6 (1.2). The mean (SD) completeness score for chatbots was 2.4 (0.8) (comprehensive), significantly higher than urologists, 1.6 (0.6) (adequate) (p < 0.001).

Answers to patient BPH messages generated by chatbots were evaluated by experts as equally accurate and more complete than urologist answers. Non-medical volunteers preferred chatbot-generated messages and considered them more empathetic compared to answers generated by urologists.