The future of AI clinicians: assessing the modern standard of chatbots and their approach to diagnostic uncertainty.

Artificial intelligence (AI) chatbots have demonstrated proficiency in structured knowledge assessments; however, there is limited research on their performance in scenarios involving diagnostic uncertainty, which requires careful interpretation and complex decision-making. This study aims to evaluate the efficacy of AI chatbots, GPT-4o and Claude-3, in addressing medical scenarios characterized by diagnostic uncertainty relative to Family Medicine residents. Questions with diagnostic uncertainty were extracted from the Progress Tests administered by the Department of Family and Community Medicine at the University of Toronto between 2022 and 2023. Diagnostic uncertainty questions were defined as those presenting clinical scenarios where symptoms, clinical findings, and patient histories do not converge on a definitive diagnosis, necessitating nuanced diagnostic reasoning and differential diagnosis. These questions were administered to a cohort of 320 Family Medicine residents in their first (PGY-1) and second (PGY-2) postgraduate years and inputted into GPT-4o and Claude-3. Errors were categorized into statistical, information, and logical errors. Statistical analyses were conducted using a binomial generalized estimating equation model, paired t-tests, and chi-squared tests. Compared to the residents, both chatbots scored lower on diagnostic uncertainty questions (p < 0.01). PGY-1 residents achieved a correctness rate of 61.1% (95% CI: 58.4-63.7), and PGY-2 residents achieved 63.3% (95% CI: 60.7-66.1). In contrast, Claude-3 correctly answered 57.7% (n = 52/90) of questions, and GPT-4o correctly answered 53.3% (n = 48/90). Claude-3 had a longer mean response time (24.0 s, 95% CI: 21.0-32.5 vs. 12.4 s, 95% CI: 9.3-15.3; p < 0.01) and produced longer answers (2001 characters, 95% CI: 1845-2212 vs. 1596 characters, 95% CI: 1395-1705; p < 0.01) compared to GPT-4o. Most errors by GPT-4o were logical errors (62.5%). While AI chatbots like GPT-4o and Claude-3 demonstrate potential in handling structured medical knowledge, their performance in scenarios involving diagnostic uncertainty remains suboptimal compared to human residents.

Huang RS ，Benour A ，Kemppainen J ，Leung FH ... -  《BMC Medical Education》

Assessment of Resident and AI Chatbot Performance on the University of Toronto Family Medicine Residency Progress Test: Comparative Study.

Large language model (LLM)-based chatbots are evolving at an unprecedented pace with the release of ChatGPT, specifically GPT-3.5, and its successor, GPT-4. Their capabilities in general-purpose tasks and language generation have advanced to the point of performing excellently on various educational examination benchmarks, including medical knowledge tests. Comparing the performance of these 2 LLM models to that of Family Medicine residents on a multiple-choice medical knowledge test can provide insights into their potential as medical education tools. This study aimed to quantitatively and qualitatively compare the performance of GPT-3.5, GPT-4, and Family Medicine residents in a multiple-choice medical knowledge test appropriate for the level of a Family Medicine resident. An official University of Toronto Department of Family and Community Medicine Progress Test consisting of multiple-choice questions was inputted into GPT-3.5 and GPT-4. The artificial intelligence chatbot's responses were manually reviewed to determine the selected answer, response length, response time, provision of a rationale for the outputted response, and the root cause of all incorrect responses (classified into arithmetic, logical, and information errors). The performance of the artificial intelligence chatbots were compared against a cohort of Family Medicine residents who concurrently attempted the test. GPT-4 performed significantly better compared to GPT-3.5 (difference 25.0%, 95% CI 16.3%-32.8%; McNemar test: P<.001); it correctly answered 89/108 (82.4%) questions, while GPT-3.5 answered 62/108 (57.4%) questions correctly. Further, GPT-4 scored higher across all 11 categories of Family Medicine knowledge. In 86.1% (n=93) of the responses, GPT-4 provided a rationale for why other multiple-choice options were not chosen compared to the 16.7% (n=18) achieved by GPT-3.5. Qualitatively, for both GPT-3.5 and GPT-4 responses, logical errors were the most common, while arithmetic errors were the least common. The average performance of Family Medicine residents was 56.9% (95% CI 56.2%-57.6%). The performance of GPT-3.5 was similar to that of the average Family Medicine resident (P=.16), while the performance of GPT-4 exceeded that of the top-performing Family Medicine resident (P<.001). GPT-4 significantly outperforms both GPT-3.5 and Family Medicine residents on a multiple-choice medical knowledge test designed for Family Medicine residents. GPT-4 provides a logical rationale for its response choice, ruling out other answer choices efficiently and with concise justification. Its high degree of accuracy and advanced reasoning capabilities facilitate its potential applications in medical education, including the creation of exam questions and scenarios as well as serving as a resource for medical knowledge or information on community services.

Huang RS ，Lu KJQ ，Meaney C ，Kemppainen J ，Punnett A ，Leung FH ... -  《-》

Evaluation of AI-generated responses by different artificial intelligence chatbots to the clinical decision-making case-based questions in oral and maxillofacial surgery.

This study aims to evaluate the correctness of the generated answers by Google Bard, GPT-3.5, GPT-4, Claude-Instant, and Bing chatbots to decision-making clinical questions in the oral and maxillofacial surgery (OMFS) area. A group of 3 board-certified oral and maxillofacial surgeons designed a questionnaire with 50 case-based questions in multiple-choice and open-ended formats. Answers of chatbots to multiple-choice questions were examined against the chosen option by 3 referees. The chatbots' answers to the open-ended questions were evaluated based on the modified global quality scale. A P-value under .05 was considered significant. Bard, GPT-3.5, GPT-4, Claude-Instant, and Bing answered 34%, 36%, 38%, 38%, and 26% of the questions correctly, respectively. In open-ended questions, GPT-4 scored the most answers evaluated as grades "4" or "5," and Bing scored the most answers evaluated as grades "1" or "2." There were no statistically significant differences between the 5 chatbots in responding to the open-ended (P = .275) and multiple-choice (P = .699) questions. Considering the major inaccuracies in the responses of chatbots, despite their relatively good performance in answering open-ended questions, this technology yet cannot be trusted as a consultant for clinicians in decision-making situations.

Azadi A ，Gorjinejad F ，Mohammad-Rahimi H ，Tabrizi R ，Alam M ，Golkar M ... -  《-》

GPT-4o’s competency in answering the simulated written European Board of Interventional Radiology exam compared to a medical student and experts in Germany and its ability to generate exam items on interventional radiology: a descriptive study.

Ebel S ，Ehrengut C ，Denecke T ，Gößmann H ，Beeskow AB ... -  《-》

Can Artificial Intelligence Pass the American Board of Orthopaedic Surgery Examination? Orthopaedic Residents Versus ChatGPT.

Advances in neural networks, deep learning, and artificial intelligence (AI) have progressed recently. Previous deep learning AI has been structured around domain-specific areas that are trained on dataset-specific areas of interest that yield high accuracy and precision. A new AI model using large language models (LLM) and nonspecific domain areas, ChatGPT (OpenAI), has gained attention. Although AI has demonstrated proficiency in managing vast amounts of data, implementation of that knowledge remains a challenge. (1) What percentage of Orthopaedic In-Training Examination questions can a generative, pretrained transformer chatbot (ChatGPT) answer correctly? (2) How does that percentage compare with results achieved by orthopaedic residents of different levels, and if scoring lower than the 10th percentile relative to 5th-year residents is likely to correspond to a failing American Board of Orthopaedic Surgery score, is this LLM likely to pass the orthopaedic surgery written boards? (3) Does increasing question taxonomy affect the LLM's ability to select the correct answer choices? This study randomly selected 400 of 3840 publicly available questions based on the Orthopaedic In-Training Examination and compared the mean score with that of residents who took the test over a 5-year period. Questions with figures, diagrams, or charts were excluded, including five questions the LLM could not provide an answer for, resulting in 207 questions administered with raw score recorded. The LLM's answer results were compared with the Orthopaedic In-Training Examination ranking of orthopaedic surgery residents. Based on the findings of an earlier study, a pass-fail cutoff was set at the 10th percentile. Questions answered were then categorized based on the Buckwalter taxonomy of recall, which deals with increasingly complex levels of interpretation and application of knowledge; comparison was made of the LLM's performance across taxonomic levels and was analyzed using a chi-square test. ChatGPT selected the correct answer 47% (97 of 207) of the time, and 53% (110 of 207) of the time it answered incorrectly. Based on prior Orthopaedic In-Training Examination testing, the LLM scored in the 40th percentile for postgraduate year (PGY) 1s, the eighth percentile for PGY2s, and the first percentile for PGY3s, PGY4s, and PGY5s; based on the latter finding (and using a predefined cutoff of the 10th percentile of PGY5s as the threshold for a passing score), it seems unlikely that the LLM would pass the written board examination. The LLM's performance decreased as question taxonomy level increased (it answered 54% [54 of 101] of Tax 1 questions correctly, 51% [18 of 35] of Tax 2 questions correctly, and 34% [24 of 71] of Tax 3 questions correctly; p = 0.034). Although this general-domain LLM has a low likelihood of passing the orthopaedic surgery board examination, testing performance and knowledge are comparable to that of a first-year orthopaedic surgery resident. The LLM's ability to provide accurate answers declines with increasing question taxonomy and complexity, indicating a deficiency in implementing knowledge. Current AI appears to perform better at knowledge and interpretation-based inquires, and based on this study and other areas of opportunity, it may become an additional tool for orthopaedic learning and education.

Lum ZC 《-》