1020: REAL-WORLD PROBLEMS WITH REAL-WORLD DATA: ADDRESSING DATA QUALITY IN THE ELECTRONIC HEALTH RECORD

Introduction: Real-world data (RWD) offers a valuable opportunity for clinicians and researchers to evaluate and explore the effectiveness of treatment strategies in real-world settings. The CURE Drug Repurposing Collaboratory, Johns Hopkins University, and SCCM Discovery are adapting an open science tool stack (the Edge Tool Suite) that facilitates harmonization of electronic health record (EHR) data into the Observational Medical Outcomes Partnership (OMOP) common data model, specifically designed to enable the systematic capture of data from the EHR. We present a process of iterative data quality assessments to support and expand an OMOP-based RWD registry of critically ill COVID-19 patients. Methods: Three healthcare sites utilized the Edge Tool Suite to harmonize data to OMOP following an extract, transform, and load (ETL) process, totaling 20,000+ cases. We applied the Data Quality Dashboard (DQD), a component of the suite, to evaluate evidence-based metrics of plausibility, conformance, and completeness. We developed profile scripts that generate summary statistics from the data (e.g., per patient frequency and distribution of laboratory tests and results, prevalence of comorbidities) to further elucidate potential data issues and explore whether these resulted from EHR challenges (e.g., documentation errors by clinical staff) or ETL issues (e.g., using the incorrect OMOP concept). Participating sites received guidance to explore and improve data quality. Results: Our process identified a wide range of data issues and facilitated rapid development of solutions. Some issues were readily identified and addressed through the DQD alone (e.g., plausibility rules identified a patient with a documented height of 3 inches). Completeness metrics helped identify an issue at one site where 99.7% of patients had a recorded PaO2. Through profile scripts, we found SpO2 had been incorrectly mapped to the concept for PaO2 during ETL. At another site the DQD showed excellent capture of oxygen devices, but the profile scripts revealed 98% of patients were documented as receiving oxygen by face tent. Conclusions: Our data quality process helped multiple sites improve data quality with a focus on being fit-for-purpose for critically ill COVID-19 patients and can be applied to other critical care populations and beyond.