Large Language Models in the Emergency Department: Fairness Implications for Clinicians

Applications of artificial intelligence (AI), particularly large language models (LLMs), are rapidly expanding in emergency medicine, but concerns remain regarding their fairness across diverse patient populations. Fairness extends beyond model accuracy and includes equal performance across patient groups, equitable outcomes and allocation of clinical benefits, and procedural fairness throughout the lifecycle of AI implementation. This narrative review aims to provide a practical framework for understanding and evaluating fairness in emergency medicine applications of AI and LLMs. This narrative review organizes fairness risks into three clinically relevant layers. First, datalevel fairness concerns the data used to train, fine-tune, or prompt AI systems. Biased clinical data, including non-random missingness and stigmatizing documentation, can influence model performance. Second, output-level fairness concerns whether the outputs generated by AI systems are equally accurate, safe, and useful across patient groups. In emergency care, subgroup-specific errors in recommendations, predictions, translations, or instructions may arise from aggregation, learning, or evaluation bias, even when overall model performance appears strong. Third, human-interaction fairness concerns whether the benefits of AI are realized equitably in clinical practice and by patients. Even when data and outputs are fair, inequities may arise if clinicians over-rely on or reject useful AI recommendations, or if patients are unable to access, understand, or trust AI-mediated information. Clinicians should consider potential implications for safety, transparency, and fairness alongside overall model accuracy when evaluating AI tools in emergency medicine.