Bidirectional data harmonization across the multiple chronic disease disparities research consortium: A pipeline description of opportunities and challenges.

Abstract

The Multiple Chronic Disease Disparities Research (MCD-DR) Consortium investigates ways to prevent and manage multiple chronic conditions in diverse populations. Data harmonization (DH) integrates and pools diverse datasets to address research questions in underrepresented populations and advance individualized care. This study outlines bidirectional DH pipelines for a large-scale research consortium focused on social determinants of health (SDOH). The DH pipeline involved developing common data elements (CDEs), forming working groups, addressing data security and privacy concerns, planning data transfer and management, and engaging partners. Bidirectional DH (retrospective for early-launch projects and prospective for later-launch projects) was conducted. Tools and methodologies, such as large language models (LLMs), were employed to improve efficiency in preprocessing and harmonizing CDEs. Three key challenges emerged during DH. First, balancing CDE comprehensiveness with participant burden required iterative discussions. Second, harmonizing heterogenous variables was complex and risked information loss across studies. Third, validating and modifying measures for diverse populations proved essential to ensure inclusivity. The Consortium's coordinated approach highlights strategies to balance comprehensiveness, manage variable heterogeneity, and validate measures across diverse groups. This study demonstrates innovative contributions to the field, including: 1) establishing comprehensive SDOH-related CDEs, 2) leveraging LLMs to enhance scalability and efficiency, 3) implementing bidirectional DH, and 4) engaging communities authentically to foster trust and ensure DH success. These strategies and insights provide guidance for researchers facing similar challenges in biomedical informatics, and the harmonized dataset provides valuable opportunities for identifying critical SDOH predictors, expanding future research possibilities.