Accuracy and Reliability of Chatbot Responses to Physician Questions

Abstract

Importance: Natural language processing tools, such as ChatGPT (generative pretrained transformer, hereafter referred to as chatbot), have the potential to radically enhance the accessibility of medical information for health professionals and patients. Assessing the safety and efficacy of these tools in answering physician-generated questions is critical to determining their suitability in clinical settings, facilitating complex decision-making, and optimizing health care efficiency.

Objective: To assess the accuracy and comprehensiveness of chatbot-generated responses to physician-developed medical queries, highlighting the reliability and limitations of artificial intelligence-generated medical information.

Design, setting, and participants: Thirty-three physicians across 17 specialties generated 284 medical questions that they subjectively classified as easy, medium, or hard with either binary (yes or no) or descriptive answers. The physicians then graded the chatbot-generated answers to these questions for accuracy (6-point Likert scale with 1 being completely incorrect and 6 being completely correct) and completeness (3-point Likert scale, with 1 being incomplete and 3 being complete plus additional context). Scores were summarized with descriptive statistics and compared using the Mann-Whitney U test or the Kruskal-Wallis test. The study (including data analysis) was conducted from January to May 2023.

Main outcomes and measures: Accuracy, completeness, and consistency over time and between 2 different versions (GPT-3.5 and GPT-4) of chatbot-generated medical responses.

Results: Across all questions (n = 284) generated by 33 physicians (31 faculty members and 2 recent graduates from residency or fellowship programs) across 17 specialties, the median accuracy score was 5.5 (IQR, 4.0-6.0) (between almost completely and complete correct) with a mean (SD) score of 4.8 (1.6) (between mostly and almost completely correct). The median completeness score was 3.0 (IQR, 2.0-3.0) (complete and comprehensive) with a mean (SD) score of 2.5 (0.7). For questions rated easy, medium, and hard, the median accuracy scores were 6.0 (IQR, 5.0-6.0), 5.5 (IQR, 5.0-6.0), and 5.0 (IQR, 4.0-6.0), respectively (mean [SD] scores were 5.0 [1.5], 4.7 [1.7], and 4.6 [1.6], respectively; P = .05). Accuracy scores for binary and descriptive questions were similar (median score, 6.0 [IQR, 4.0-6.0] vs 5.0 [IQR, 3.4-6.0]; mean [SD] score, 4.9 [1.6] vs 4.7 [1.6]; P = .07). Of 36 questions with scores of 1.0 to 2.0, 34 were requeried or regraded 8 to 17 days later with substantial improvement (median score 2.0 [IQR, 1.0-3.0] vs 4.0 [IQR, 2.0-5.3]; P < .01). A subset of questions, regardless of initial scores (version 3.5), were regenerated and rescored using version 4 with improvement (mean accuracy [SD] score, 5.2 [1.5] vs 5.7 [0.8]; median score, 6.0 [IQR, 5.0-6.0] for original and 6.0 [IQR, 6.0-6.0] for rescored; P = .002).

Conclusions and relevance: In this cross-sectional study, chatbot generated largely accurate information to diverse medical queries as judged by academic physician specialists with improvement over time, although it had important limitations. Further research and model development are needed to correct inaccuracies and for validation.

Figure 1.. Methods

AI indicates artificial intelligence. ^aD.B.J. and L.E.W. scored 2 separate data sets of melanoma and immunotherapy and common conditions questions. ^bRegenerated answers were created 8 to 17 days after initial answers. ^cRegenerated answers were created 90 days after initial answers.

Figure 2.. Accuracy of Chatbot-Generated Answers

Accuracy of artificial intelligence answers from multispecialty questions (A-C [P < .01 for panel C]) or all questions (multispecialty, melanoma and immunotherapy, and common medical conditions; D-F [P = .03 for panel E]). A, Among all descriptive questions in the multispecialty analysis, median accuracy scores were 5.0 (IQR, 3.0-6.0) (mean [SD] score, 4.9 [1.5]) for easy, 5.0 (IQR, 3.0-6.0) (mean [SD] score, 4.4 [1.9]) for medium, and 5.0 (IQR, 3.0-6.0) (mean [SD] score, 4.1 [1.8]) for hard questions (P = .70 determined by the Kruskal-Wallis test). B, Among all binary questions in the multispecialty analysis, median accuracy scores were 6.0 (IQR, 5.0-6.0) (mean [SD] score, 4.9 [1.8]) for easy, 4.0 (IQR, 3.0-6.0) (mean [SD] score, 4.3 [1.6]) for medium, and 5.0 (IQR, 1.0-6.0) (mean [SD] score, 4.2 [1.8]) for hard answers (P = .10 determined by the Kruskal-Wallis test). C, Of 36 questions with accuracy scores of 2 or lower, 34 were requeried or regraded 8 to 17 days later. The median accuracy score for original questions was 2.0 (IQR, 1.0-2.0) (mean [SD] score, 1.6 [0.5]) compared with 4.0 (IQR, 2.0-5.3) (mean [SD] score, 3.9 [1.8]) for rescored answers (P < .01 determined by Wilcoxon signed rank test). D, Among all descriptive questions, median accuracy scores for easy, medium, and hard questions were 5.3 (IQR, 3.0-6.0) (mean [SD] score, 4.8 [1.5]) for easy, 5.5 (IQR, 3.3-6.0) (mean [SD] score, 4.7 [1.7]) for medium, and 5.0 (IQR, 3.6-6.0) (mean [SD] score, 4.5 [1.6]) for hard questions (P = .40 determined by the Kruskal-Wallis test). E, Among all binary questions, median accuracy scores were 6.0 (IQR, 5.0-6.0) (mean [SD] score, 5.3 [1.5]) for easy, 5.5 (IQR, 3.4-6.0) (mean [SD] score, 4.6 [1.6) for medium, and 5.5 (IQR, 4.0-6.0) (mean [SD] score, 4.8 [1.6]) for hard questions, which resulted in a significant difference among groups (P = .03 determined by the Kruskal-Wallis test). F, Median accuracy scores were 5.0 (IQR, 3.4-6.0) (mean [SD] score, 4.7 [1.6]) for all descriptive questions and 6.0 (IQR, 4.0-6.0) (mean [SD] score, 4.9 [1.6]) for binary questions (P = .07 determined by Mann-Whitney U test).