AI-Enhanced Healthcare Data Quality Governance: An Integrated Approach for Anomaly Detection and Integrity Verification

Abstract

Healthcare data quality remains a critical challenge affecting clinical decision-making, patient safety, and operational efficiency across medical institutions. This paper presents an integrated approach for AI-enhanced healthcare data quality governance that combines rule-based anomaly detection, statistical scoring mechanisms, and temporal consistency verification. The proposed framework establishes hierarchical quality checkpoints across heterogeneous EHR tables and clinical documentation streams (and is extendable to multi-source settings), enabling real-time identification of data entry errors, logical conflicts, and distribution drift patterns. Through systematic evaluation on the MIMIC-III EHR dataset (53,423 ICU admissions; >50,000 ICU admission records) using proxy anomaly labels derived from rule violations and cross-field/temporal consistency checks (with controlled synthetic anomaly injections for robustness testing), our approach achieves 94.7% detection accuracy with a false-positive rate of 3.2%. The experimental results validate the effectiveness of the integrated governance methodology in maintaining data integrity across diverse clinical scenarios while providing interpretable evidence chains for healthcare practitioners.