Towards Error-Free EHRs: Reasoning-Intensive Consistency Verification Between Clinical Notes and Structured Tables in Electronic Health Records

2026-05-25

AI Summary

This study addresses the deep semantic inconsistencies between clinical notes and structured tables in electronic health records (EHRs), which existing methods fail to capture due to their reliance on superficial matching rather than clinical reasoning, event relationships, and temporal dynamics. To tackle this, the authors introduce EHR-ReasonCon, a reasoning-intensive consistency verification benchmark built on MIMIC-III, featuring the first expert-guided, inference-level annotation protocol and a dedicated table exploration tool. They further propose EHR-Inspector, an LLM-driven framework that integrates clinical text segmentation, anchor entity and temporal reference extraction, structured querying, and an LLM-as-a-judge evaluation mechanism. Experiments demonstrate consistent and significant improvements over baselines across multiple model backbones, with robust performance under both stringent and lenient expert evaluations. Ablation studies confirm the efficacy of the design and highlight nuanced discrepancies with human judgments.

Abstract

Data consistency between unstructured clinical notes and structured tables in Electronic Health Records (EHRs) is essential for patient safety and clinical decision-making. However, existing work on note-table consistency verification mainly relies on surface-level matching of numeric values or simple events. Such approaches fail to capture the reasoning underlying real-world EHR documentation, including clinical interpretation, event relations, and temporal changes. To address this gap, we introduce EHR-ReasonCon, a reasoning-intensive benchmark for note-table consistency verification. Built on MIMIC-III with expert-guided annotations, it comprises 8,048 entities derived from clinical notes and provides high-quality ground-truth labels. The annotation protocol is supported by specialized table-exploration tools to ensure systematic evidence retrieval and reliable consistency assessment. We also propose EHR-Inspector, an LLM-based framework that segments notes, extracts anchor entities and temporal references, and uses table-exploration tools to verify consistency against structured tables. Evaluated using expert-validated LLM-as-a-judge metrics under harsh and lenient criteria, EHR-Inspector achieves state-of-the-art performance across multiple model backbones. Analyses further demonstrate the effectiveness of its components and highlight differences from human verification.

Problem

• EHR consistency
• clinical notes
• structured tables
• reasoning-intensive verification
• data inconsistency

Innovation

• reasoning-intensive
• consistency verification
• electronic health records
• LLM-based framework
• expert-guided annotation