A Multidimensional Framework for Data Quality Assessment in Heart Failure: Integrating IEEE 2801-2022 and Fairness Metrics

Abstract

Heart failure (HF) affects over 64 million people globally and poses complex diagnostic and therapeutic challenges. Reliable clinical research in HF hinges on high-quality data. This study presents a novel data quality assessment (DQA) framework tailored to retrospective HF datasets. It adapts the IEEE standard 2801-2022 criteria - originally for general medical data - to HF's clinical and multimodal structure and introduces a fairness-aware dimension to assess demographic representativeness. Applied to a real-world dataset of 6,039 patients and over 110,000 records across 11 clinical domains, the framework evaluates six dimensions: Completeness, Accuracy, Consistency, Compliance, Timeliness, and Fairness. Initial completeness was low (48.82%), but improved to 61.04% after cleaning via outlier correction, imputation, and schema normalization. Accuracy and compliance reached 100%, and consistency improved to 99.61%. Fairness, measured via JensenShannon Similarity across age, sex, and BMI, remained at 87.35%, highlighting demographic imbalance remained unresolved by technical cleaning. This is the first standards-aligned, domain-adapted, and fairness-extended DQA pipeline for HF, producing a robust dataset suitable for machine learning and clinical decision support.