Agentic llm predicts pediatric diabetes outcomes better than classic machine learning
Agentic LLM reasoning in pediatric diabetes outcome prediction from multimodal medical records
Computation and Language
Summary
Predicting health outcomes for children with diabetes is hard because there are fewer patients and uncommon events. The authors used a special computer program called CEDAR that reads different types of medical information and explains its reasoning. This program predicted future health problems more accurately than traditional computer methods in young people with type 1 and type 2 diabetes. Doctors who looked only at the program’s explanations agreed with its predictions in most cases.
What this means in practice
- •For hospital data teams: Use agentic LLMs like CEDAR to improve early risk predictions for pediatric diabetes complications from patient records.
- •For developers of clinical decision support: Build explainable AI tools that assist clinicians by citing evidence and rationale for pediatric diabetes outcome predictions.$Commercial implications: Allows creation of AI-based medical software that enhances doctor decision-making with transparent, reliable clinical predictions.
Authors
Daniel Palacios, Guantong Qi, Anna Kadi, Natali Borrego, Sukru Aras, Michael L. Ferm, Elizabeth Kubota-Mishra, Maria J. Redondo, Zhandong Liu, Mustafa Tosur, Hu Chen
Abstract
Large language model (LLM) agents are proposed for clinical prediction but are untested in pediatrics, where cohorts are small and outcomes are uncommon. Using records from 2,489 youth with type 2 diabetes (T2D) and 4,018 with type 1 diabetes (T1D), spanning structured data, note-derived social determinants and, in T1D, continuous glucose monitoring data, we predicted seven T2D outcomes, four also in T1D, at 24 months from 18 months of follow-up. CEDAR (Clinical Evidence-Driven Agent for Risk-prediction), an evidence-citing, self-auditing LLM pipeline, outperformed trained machine learning in both cohorts (pooled area under the receiver operating characteristic curve 0.815 versus 0.656 to 0.700 in T2D and 0.756 versus 0.603 to 0.631 in T1D), most on the sparsely screened outcomes. Only 4 of 70,246 values it cited did not match the chart. Reading only the evidence and reasoning cited by CEDAR, two pediatric endocrinologists reached the same prediction as CEDAR in about nine of ten cases, agreeing with each other on 90.5%.