Purpose: To evaluate the accuracy of GPT-3.5, GPT-4, and a fine-tuned GPT-3.5 model in applying Fleischner Society recommendations to lung nodules. Methods: We generated 10 lung nodule descriptions for each of the 12 nodule categories from the Fleischner Society guidelines, incorporating them into a single fictitious report (n = 120). GPT-3.5 and GPT-4 were prompted to make follow-up recommendations based on the reports. We then incorporated the full guidelines into the prompts and re-submitted them. Finally, we re-submitted the prompts to a fine-tuned GPT-3.5 model. Results were analyzed using binary accuracy analysis in R. Results: GPT-3.5 accuracy in applying Fleischner Society guidelines was 0.058 (95% CI: 0.02, 0.12). GPT-4 accuracy was improved at 0.15 (95% CI: 0.09, 0.23; P = .02 for accuracy comparison). In recommending PET-CT and/or biopsy, both GPT-3.5 and GPT-4 had an F-score of 0.00. After explicitly including the Fleischner Society guidelines in the prompt, GPT-3.5 and GPT-4 significantly improved their accuracy to 0.42 (95% CI: 0.33, 0.51; P < .001) and to 0.66 (95% CI: 0.57, 0.74; P < .001), respectively. GPT-4 remained significantly better than GPT-3.5 (P < .001). The fine-tuned GPT-3.5 model accuracy was 0.46 (95% CI: 0.37, 0.55), not different from the GPT-3.5 model with guidelines included (P = .53). Conclusion: GPT-3.5 and GPT-4 performed poorly in applying widely known guidelines and never correctly recommended biopsy. Flawed knowledge and reasoning both contributed to their poor performance. While GPT-4 was more accurate than GPT-3.5, its inaccuracy rate was unacceptable for clinical practice. These results underscore the limitations of large language models for knowledge and reasoning-based tasks.

OBJECTIVE: Colonoscopy is commonly used in screening and surveillance for colorectal cancer. Multiple different guidelines provide recommendations on the interval between colonoscopies. This can be challenging for non-specialist healthcare providers to navigate. Large language models like ChatGPT are a potential tool for parsing patient histories and providing advice. However, the standard GPT model is not designed for medical use and can hallucinate. One way to overcome these challenges is to provide contextual information with medical guidelines to help the model respond accurately to queries. Our study compares the standard GPT4 against a contextualized model provided with relevant screening guidelines. We evaluated whether the models could provide correct advice for screening and surveillance intervals for colonoscopy.
METHODS: Relevant guidelines pertaining to colorectal cancer screening and surveillance were formulated into a knowledge base for GPT. We tested 62 example case scenarios (three times each) on standard GPT4 and on a contextualized model with the knowledge base.
RESULTS: The contextualized GPT4 model outperformed the standard GPT4 in all domains. No high-risk features were missed, and only two cases had hallucination of additional high-risk features. A correct interval to colonoscopy was provided in the majority of cases. Guidelines were appropriately cited in almost all cases.
CONCLUSIONS: A contextualized GPT4 model could identify high-risk features and quote appropriate guidelines without significant hallucination. It gave a correct interval to the next colonoscopy in the majority of cases. This provides proof of concept that ChatGPT with appropriate refinement can serve as an accurate physician assistant.