ArgHiTZ at ArchEHR-QA 2025: A Two-Step Divide and Conquer Approach to Patient Question Answering for Top Factuality

2025-06-15

AI Summary

To address the factual consistency bottleneck in clinical question answering (QA) for the ArchEHR-QA 2025 shared task, this paper proposes a knowledge-free, two-stage decoupled framework. First, it retrieves salient sentence segments from electronic health records (EHRs) via semantic similarity matching, followed by BERT-based re-ranking for precise candidate selection. Second, it conditions an end-to-end generative model exclusively on the refined retrieved evidence to produce answers. Crucially, this work pioneers the deep integration of a re-ranker into the retrieval-generation pipeline—without external knowledge—thereby substantially improving answer factuality. In official evaluation, the system achieved top performance in factual consistency among 30 participating teams (Rank 1), with an overall score of 0.44 (Rank 8). This study establishes an interpretable, reproducible, and lightweight paradigm for high-assurance clinical QA, empirically validating the critical role of decoupled optimization of retrieval and generation in ensuring medical fact consistency.

Technology Category

• Natural Language Processing: Question Answering
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

This work presents three different approaches to address the ArchEHR-QA 2025 Shared Task on automated patient question answering. We introduce an end-to-end prompt-based baseline and two two-step methods to divide the task, without utilizing any external knowledge. Both two step approaches first extract essential sentences from the clinical text, by prompt or similarity ranking, and then generate the final answer from these notes. Results indicate that the re-ranker based two-step system performs best, highlighting the importance of selecting the right approach for each subtask. Our best run achieved an overall score of 0.44, ranking 8th out of 30 on the leaderboard, securing the top position in overall factuality.

Problem

• Automated patient question answering in clinical texts
• Extracting essential sentences without external knowledge
• Improving factuality in medical question answering

Innovation

• Two-step divide and conquer approach
• Prompt-based essential sentence extraction
• Re-ranker for optimal sentence selection