Problem
Research questions and friction points this paper is trying to address.
CRF filling
clinical NLP
structured information extraction
electronic health records
privacy-preserving LLM
sebis at CRF Filling 2026: A Two-Stage Local LLM Pipeline for Medical CRF Filling
🤖 AI Summary
Extracting structured clinical information from unstructured electronic health records faces significant challenges, including privacy risks, high inference costs, and model hallucinations, which hinder the deployment of large language models in clinical settings. This work proposes a fully local two-stage pipeline that first performs item-level binary existence classification using MedGemma-27B and then leverages few-shot in-context learning for value extraction, producing deterministic outputs strictly grounded in textual evidence—without requiring external APIs or model fine-tuning. Evaluated on the official CL4Health 2026 test set, the method achieves a macro F1 score of 0.55, ranking second among all local open-source submissions, thereby effectively balancing privacy preservation, reliability, and performance.
📝 Abstract
The extraction of structured clinical information from unstructured EHR notes is a persistent bottleneck in healthcare informatics. While large language models (LLMs) offer high performance, their deployment in clinical settings is hindered by privacy risks, inference costs, and the tendency to hallucinate beyond textual evidence. We address these challenges for the CL4Health 2026 Case Report Form (CRF) filling task by proposing a fully local, domain-adapted pipeline using the MedGemma-27B model. Our two-stage architecture, which separates binary presence classification from value extraction, enforces strict adherence to textual evidence and ensures deterministic outputs for negated, uncertain, or unknown states. By leveraging item-specific, few-shot in-context learning without external API calls or fine-tuning, our approach achieves a macro-F1 score of 0.55 on the official English test track. This result secures second place among all locally-hosted, open-source submissions. Our work demonstrates that privacy-preserving, on-premise LLM pipelines can achieve near-competitive performance with proprietary frontier models, providing a practical, data-sovereign framework for clinical NLP.
Innovation
Methods, ideas, or system contributions that make the work stand out.
two-stage LLM pipeline
local deployment
evidence-constrained extraction
few-shot in-context learning
clinical CRF filling
🔎 Similar Papers
No similar papers found.
💼 Related Jobs
