Agentic LLM reasoning in pediatric diabetes outcome prediction from multimodal medical records
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%.
Related
- APOD: reasoning-guided agentic population ordinary differential equation discovery for pharmacological digital twins
- TRACEDD: A Tool-grounded Reasoning and Agentic Coordination for Explainable Drug Design
- CircuitATLAS: Agentic reasoning over a systems neuroscience knowledge graph for target discovery in circuitopathies
- OmniVCBench: Benchmarking Evidence-Grounded Multimodal Reasoning Towards AI Virtual Cells
- Multimodal reasoning for broadly neutralizing antibody discovery from label-free human B cell repertoires across virus families
BibTeX
@misc{palacios2026agentic,
title = {{Agentic LLM reasoning in pediatric diabetes outcome prediction from multimodal medical records}},
author = {Daniel Palacios and Guantong Qi and Anna Kadi and Natali Borrego and Sukru Aras and Michael L. Ferm and Elizabeth Kubota-Mishra and Maria J. Redondo and Zhandong Liu and Mustafa Tosur and Hu Chen},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.40379},
eprint = {2610.40379},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.40379},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Agentic LLM reasoning in pediatric diabetes outcome prediction from multimodal medical records
AU - Daniel Palacios
AU - Guantong Qi
AU - Anna Kadi
AU - Natali Borrego
AU - Sukru Aras
AU - Michael L. Ferm
AU - Elizabeth Kubota-Mishra
AU - Maria J. Redondo
AU - Zhandong Liu
AU - Mustafa Tosur
AU - Hu Chen
PY - 2026
DA - 2026/10/29
PB - arXiv
DO - 10.48550/arxiv.2610.40379
UR - https://arxiv.org/abs/2610.40379
AB - 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%.
SN - 2331-8422
ER -