Answers are grounded in this paper and cite their sources.
OBJECTIVE: To assess how prompt formulation and clinical complexity influence the performance of large language models (LLMs) in real otolaryngology cases. STUDY DESIGN: Comparative observational study. SETTING: University hospital study. METHODS: Clinical information from 80 consecutive outpatients prospectively recruited at two otolaryngology centers and from 40 published case reports was submitted to ChatGPT-5, Claude-Sonnet-4.5, and Gemini-2.5-Pro using two prompt strategies: a physician-oriented prompt with medical terminology, and a patient-oriented prompt using lay language. Cases were classified by two board-certified otolaryngologists as simple, moderate, or complex (case reports). LLM responses were rated by two otolaryngologists using the Artificial Intelligence Performance Instrument (AIPI), with adequate inter-rater reliability (intraclass correlation coefficient [ICC] = 0.855) on a subset of complex, practitioner-prompt cases, and performance was analyzed according to prompt formulation and case complexity. RESULTS: Claude-Sonnet-4.5 and Gemini-2.5-Pro significantly outperformed ChatGPT-5 across differential diagnosis, additional examinations, treatment sub-scores, and total AIPI score (P < .001). Prompt origin significantly influenced performance: patient prompts yielded lower accuracy than practitioner prompts for medical history, symptoms, and primary diagnosis (P < .001). LLMs performed better on simple cases for history consideration, but on moderate-to-complex cases for physical examination, differential diagnoses, and additional examinations. Primary diagnostic accuracy peaked for moderate cases (54.6%) and dropped for complex ones (28.7%; P < .001). CONCLUSION: LLM performance is significantly influenced by prompt formulation and clinical complexity, highlighting the importance of structured prompts and supervised clinical use. These findings suggest that the accuracy and clinical relevance of LLM outputs must be interpreted considering both prompt formulation and the amount of clinical information available to the model-a critical consideration for future integration of these tools into clinical practice.
No takes yet. Share an insight, caveat, or question.
Sogalow et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: