Large Language Models Perform Diagnostic Reasoning

Cheng-Kuang Wu, Wei-Lin Chen, Hsin-Hsi Chen

Keywords: large language models, automatic diagnosis, diagnostic reasoning, medical reasoning

TL;DR: We propose DR-CoT prompting, which elicits the diagnostic reasoning ability of LLMs and exhibit striking empirical improvements for automatic diagnosis.

Abstract: We explore the extension of chain-of-thought (CoT) prompting to medical reasoning for the task of automatic diagnosis. Motivated by doctors' underlying reasoning process, we present Diagnostic-Reasoning CoT (DR-CoT). Empirical results demonstrate that by simply prompting large language models trained only on general text corpus with two DR-CoT exemplars, the diagnostic accuracy improves by 15% comparing to standard prompting. Moreover, the gap reaches a pronounced 18% in out-domain settings. Our findings suggest expert-knowledge reasoning in large language models can be elicited through proper promptings.