Retrieval-Augmented Generation for Clinical Question Answering in Portuguese Drug Leaflets: Benefits and Limitations

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Science & Research — Health & Medical Research, Mathematics & Computational Sciences · Depth: Advanced, quick

Summary

A study evaluated Retrieval-Augmented Generation (RAG) for clinical question answering in Portuguese, utilizing over 7,000 Brazilian regulatory drug leaflets and a clinical benchmark from national medical licensing examinations (Revalida and Fuvest). RAG significantly improved factual recall and clinical coherence for medication-specific queries, boosting F1 scores from 0.276 to 0.412. However, naive retrieval did not consistently enhance complex clinical reasoning and occasionally reduced accuracy compared to a parametric-only baseline. The research identified retrieval-induced anchoring bias, where partially relevant evidence led to clinically incorrect conclusions. Critique-based and adaptive retrieval strategies successfully mitigated this bias, achieving the highest clinical benchmark accuracy of 54.25%. The findings indicate RAG's effectiveness in regulatory contexts but highlight the need for adaptive control in higher-level clinical reasoning tasks.

Key takeaway

For AI Engineers developing clinical language models, recognize that while RAG enhances factual recall for medication queries, it can introduce anchoring bias in complex reasoning. You should prioritize implementing adaptive or critique-based retrieval mechanisms to improve accuracy and ensure clinical safety, especially when handling nuanced diagnostic or treatment-related questions. Evaluate your models using clinically grounded metrics beyond traditional NLP scores to identify safety-relevant differences.

Key insights

RAG improves factual recall in clinical Q&A but needs adaptive control for complex reasoning to avoid anchoring bias.

Principles

• Naive RAG can reduce accuracy in complex reasoning.
• Retrieval-induced anchoring bias is a significant risk.
• Adaptive retrieval mitigates anchoring bias.

Method

Controlled evaluation of RAG using Brazilian drug leaflets and medical licensing exam questions, comparing naive, critique-based, and adaptive retrieval.

In practice

• Use RAG for medication-specific factual queries.
• Implement adaptive retrieval for complex clinical reasoning.
• Evaluate RAG with clinically grounded metrics.

Topics

• Retrieval-Augmented Generation
• Clinical Question Answering
• Portuguese Drug Leaflets
• Clinical Reasoning
• Retrieval-Induced Anchoring Bias

Best for: AI Engineer, Machine Learning Engineer, AI Scientist, Research Scientist, NLP Engineer

Related on AIssential

• Benchmarking Retrieval Strategies for Biomedical Retrieval-Augmented Generation: A Controlled Empirical Study — Cross-Encoder Reranking significantly improves biomedical RAG performance by enhancing contextual precision.
• An Efficient Approach for Answering Not Readily Attainable Questions for RAG-based Applications — This paper proposes an efficient approach for RAG systems to answer questions not readily found in retrieved data.
• Structured Summaries for Retrieval-Augmented Generation in Portuguese-Language Consumer Complaints — Structured summaries significantly improve dense retrieval effectiveness in RAG systems for noisy, lengthy texts.
• Ablation Study of a Fairness Auditing Agentic System for Bias Mitigation in Early-Onset Colorectal Cancer Detection — Agentic AI systems with RAG significantly improve fairness auditing in clinical AI by grounding responses in external knowledge.
• When Cases Get Rare: A Retrieval Benchmark for Off-Guideline Clinical Question Answering — LLMs require retrieval augmentation and strong reasoning for reliable off-guideline medical question answering.
• NeocorRAG: Less Irrelevant Information, More Explicit Evidence, and More Effective Recall via Evidence Chains — Optimizing retrieval quality through "Evidence Chains" significantly improves RAG reasoning accuracy and efficiency.

Open in AIssential →

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