ELiRF-UPV@MedExACT 2026: Dynamic Section Conditioning for Medical Decision Span Detection in Discharge Summaries

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Health & Medical Research · Depth: Expert, medium

Summary

ELiRF-UPV proposed a system for the MedExACT track at ACL 2026, securing the 4th position in medical decision span detection from discharge summaries. This system addresses the challenge of heterogeneous electronic health record structures by first applying dynamic section conditioning to capture contextual dependencies. It then augments a transformer backbone with category- and section-aware layer mixing to integrate global document structure with fine-grained semantic cues. For enhanced robustness, the approach incorporates an ensemble of instruction-tuned large language models for automatic section extraction. A fairness-oriented model selection criterion ensures consistent performance across minority demographic subgroups. The system achieved a final score of 0.5806 on the held-out test set, demonstrating notable improvements over the baseline across all evaluated subpopulations.

Key takeaway

For NLP Engineers developing systems for medical information extraction from discharge summaries, consider integrating dynamic section conditioning and category-aware layer mixing into your transformer architectures. You should also explore using instruction-tuned large language model ensembles for robust automatic section extraction. Implementing a fairness-oriented model selection criterion is vital to ensure your system performs equitably across diverse demographic subgroups, improving reliability in clinical analytics applications.

Key insights

The system combines dynamic section conditioning, transformer layer mixing, and LLM ensembles for robust, fair medical decision extraction.

Principles

• Contextual dependencies are crucial for EHR processing.
• Fairness criteria improve model reliability across demographics.
• Ensemble LLMs enhance robustness in extraction tasks.

Method

The approach uses dynamic section conditioning, then a transformer with category- and section-aware layer mixing. It integrates an LLM ensemble for section extraction and a fairness-oriented model selection.

In practice

• Apply dynamic section conditioning to medical texts.
• Use instruction-tuned LLMs for automated sectioning.
• Implement fairness criteria in model selection.

Topics

• Medical Decision Extraction
• Discharge Summaries
• Transformer Models
• Large Language Models
• Fairness in AI
• Clinical NLP

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Measuring the sensitivity of LLM-based structured extraction to prompt, model, and schema choices in clinical discharge summaries — LLM-based clinical data extraction sensitivity varies significantly with model and schema, especially regarding "absence-versus-silence" distinctions.
• Severity-Aware Curriculum Learning with Multi-Model Response Selection for Medical Text Generation — A severity-aware multi-model framework using curriculum learning and response selection significantly improves medical text generation across case complexities.
• AI-assisted Protocol Information Extraction For Improved Accuracy and Efficiency in Clinical Trial Workflows — AI-assisted RAG significantly boosts clinical trial protocol data extraction accuracy and efficiency over standalone LLMs.
• Cross-Lingual Empirical Evaluation of Large Language Models for Arabic Medical Tasks — English-centric LLMs exhibit significant performance gaps and reliability issues in Arabic medical tasks.
• Noise reduction in BERT NER models for clinical entity extraction — A post-processing Noise Removal model significantly boosts clinical NER precision by analyzing probability distributions and contextual features.
• ClinEnv: An Interactive Multi-Stage Long Horizon EHR Environment for Agents — ClinEnv reveals LLMs struggle with sequential, uncertain clinical decisions and information gathering in a multi-stage environment.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.