EIN-RECAP: Evidence-Grounded Multi-Agent Summarization and Retrieval of Infectious-Disease Expert Discussions
- 1The University of Iowa
Abstract
Infectious disease experts exchange practical knowledge through clinical discussions, but identifying relevant information across years of conversations is difficult. These discussions often contain differing opinions and emerging observations that are not yet reflected in published evidence. We present EIN-RECAP, an AI-assisted platform developed with Emerging Infections Network (EIN) experts and deployed in beta to summarize and search historical discussions. The system combines retrieval-grounded summarization, citation validation, and hybrid search while distinguishing published evidence from community experience. We evaluated summarization methods across 3,185 threads and conducted blinded expert review on a recent subset. Although a multi-agent workflow achieved stronger automated scores, experts favored a simpler generation pipeline and identified repetition, inaccuracies, and overconfident recommendations in the more complex system. This feedback motivated an ensemble that improved discussion coverage and key clinical points on nine high-disagreement threads, although retrieval relevance remained a limitation. Our findings demonstrate the importance of expert involvement in clinical AI development and show why automated evaluation alone is insufficient to guide system selection.
Related
- Real-world evaluation of a multilingual agentic retrieval-augmented clinical AI framework for evidence-grounded medical decision support
- OncoAgentic-RAG: An evidence-grounded multi-agent hybrid retrieval-augmented system for patient-centric clinical oncology question answering
- LitBench: A Benchmark for Retrieval-Grounded Multi-Paper Evidence Synthesis in Epilepsy
- BanglaDiaHallu: Hallucinations of large language models and the effect of guideline-grounded retrieval-augmented generation in Bangla diabetes question answering
- Integrating knowledge graphs and retrieval-augmented generation to identify clinical pharmacology gaps in infectious disease trial protocols
BibTeX
@misc{choudhuri2026einrecap,
title = {{EIN-RECAP: Evidence-Grounded Multi-Agent Summarization and Retrieval of Infectious-Disease Expert Discussions}},
author = {A. Choudhuri and Y. Zhong and S. Beekmann and J. Hodges and K. Jha and A. M. Segre and B. Adhikari and P. M. Polgreen},
howpublished = {medRxiv},
year = {2026},
month = nov,
doi = {10.64898/2026.11.05.26367381},
url = {https://www.medrxiv.org/content/10.64898/2026.11.05.26367381v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - EIN-RECAP: Evidence-Grounded Multi-Agent Summarization and Retrieval of Infectious-Disease Expert Discussions
AU - A. Choudhuri
AU - Y. Zhong
AU - S. Beekmann
AU - J. Hodges
AU - K. Jha
AU - A. M. Segre
AU - B. Adhikari
AU - P. M. Polgreen
PY - 2026
DA - 2026/11/05
PB - medRxiv
DO - 10.64898/2026.11.05.26367381
UR - https://www.medrxiv.org/content/10.64898/2026.11.05.26367381v1
AB - Infectious disease experts exchange practical knowledge through clinical discussions, but identifying relevant information across years of conversations is difficult. These discussions often contain differing opinions and emerging observations that are not yet reflected in published evidence. We present EIN-RECAP, an AI-assisted platform developed with Emerging Infections Network (EIN) experts and deployed in beta to summarize and search historical discussions. The system combines retrieval-grounded summarization, citation validation, and hybrid search while distinguishing published evidence from community experience. We evaluated summarization methods across 3,185 threads and conducted blinded expert review on a recent subset. Although a multi-agent workflow achieved stronger automated scores, experts favored a simpler generation pipeline and identified repetition, inaccuracies, and overconfident recommendations in the more complex system. This feedback motivated an ensemble that improved discussion coverage and key clinical points on nine high-disagreement threads, although retrieval relevance remained a limitation. Our findings demonstrate the importance of expert involvement in clinical AI development and show why automated evaluation alone is insufficient to guide system selection.
ER -