SemioLLM: Evaluating Large Language Models for Diagnostic Reasoning from Unstructured Clinical Narratives in Epilepsy

Large Language Models (LLMs) have been shown to encode clinical knowledge. Many evaluations, however, rely on structured question-answer benchmarks, overlooking critical challenges of interpreting and reasoning about unstructured clinical narratives in real-world settings. In this study we task eight Large Language models including two medical models (GPT-3.5, GPT-4, Mixtral-8x7B, Qwen-72B, LlaMa2, LlaMa3, OpenBioLLM, Med42) with a core diagnostic task in epilepsy: mapping seizure description phrases, after targeted filtering and standardization, to one of seven possible seizure onset zones using likelihood estimates. Most models yield results that often match the ground truth and even approach clinician-level performance after prompt engineering. Specifically, clinician-guided chain-of-thought reasoning leading to the most consistent improvements. Performance was further strongly modulated by clinical in-context impersonation, narrative length and language context (13.7%, 32.7% and 14.2% performance variation, respectively). However, expert analysis of reasoning outputs revealed that correct prediction can be based on hallucinated knowledge and inaccurate source citation, underscoring the need to improve interpretability of LLMs in clinical use. Overall, SemioLLM provides a scalable, domain-adaptable framework for evaluating LLMs in clinical disciplines where unstructured verbal descriptions encode diagnostic information. By identifying both the strengths and limitations of LLMs, our work contributes to testing the applicability of foundational AI systems for healthcare.