Language Models as Interfaces, Not Oracles: A Hybrid LLM-ML System for Pediatric Appendicitis

The article introduces ClaMPAPP (Clinical Language-assisted Machine-learning Pipeline for Appendicitis), a hybrid system that utilizes a large language model (LLM) as an interface to extract clinical features from narrative texts, which are then processed by an XGBoost classifier for risk prediction in pediatric appendicitis cases. Evaluations on two independent cohorts demonstrated that ClaMPAPP outperformed end-to-end LLM baselines in diagnostic accuracy, particularly in maintaining robustness against variations in narrative input, thus offering a more reliable and auditable framework for clinical decision support. This approach emphasizes the effective integration of LLMs with structured machine learning, addressing the limitations of direct LLM usage in clinical diagnostics.