Assessing Clinical Decision-Making Aided by a RAG-Based Dialog in Pre-Examination on Non-Odontogenic Tooth Pain

Differentiating Non-Odontogenic Tooth Pain, a potential symptom of life-threatening conditions like Ischemic Heart Disease, is a critical challenge for dentists, as existing AI tools fail to support their decision-making. To address this, we developed a chatbot using a Retrieval-Augmented Generation (RAG) enhanced Large Language Model (LLM). This study protocol outlines a trial to evaluate the impact of the AI's outputs on dentists' diagnostic accuracy. The study employs a three-arm, within-subjects crossover trial where 128 dentists evaluate simulated patient scenarios under three conditions: RAG, standalone LLM, and conventional interview via a novel Stepwise Information Disclosure Protocol, with diagnostic accuracy as the primary outcome. The findings will present a new framework for evaluating human-AI clinical interaction, with the potential to enhance patient safety by reducing missed diagnoses and unnecessary dental interventions.