MedGellan is a lightweight, annotation-free framework for generating clinical guidelines from medical records. It uses large-scale language models (LLMs) to generate clinical guidelines used for diagnosis prediction from medical records, assisting physicians in making diagnosis predictions. It takes into account the temporal order of clinical data using a Bayesian-inspired prompting strategy. Initial experimental results show that LLM-generated guidelines using MedGellan improve diagnostic performance, especially in terms of recall and F1 score. Considering the challenges of fully automated systems, we present a hybrid framework that combines human supervision and machine intelligence.
