(032) Race and Ethnicity Variables in Real-World Data: Implications for Reproducibility in Pharmacoepidemiology.

Background: Race and ethnicity variables are increasingly used in pharmacoepidemiological research, which relies on electronic health records and insurance claims data routinely generated during care that varies according to race and ethnicity. Yet, race and ethnicity variables’ roles in pharmacoepidemiology research reproducibility, has not been described.

Objectives: To identify and describe the influence of race and ethnicity variables in pharmacoepidemiological research reproducibility and potential strategies for improvement.

Methods: We analyzed pharmacoepidemiology with Goodman, Fanelli, and Ioannidis’ (2016) three-dimensional conceptualization of reproducibility: methods (reproducing the sample and parameter estimates), results (reproducing quantitative estimates in a new sample), and inferential (arriving at similar interpretations in the methods and results reproductions). We queried systematic reviews of epidemiological research for qualitative discussions of reproducibility, quantitative evidence on data quality (e.g., completeness), analysis (e.g., effect modification by race), reporting (e.g., definition and coding of race), and recommendations (use of DAGs to represent assumed causal relationships). We also evaluated pharmacoepidemiology reproducibility tools, The HARmonized Protocol Template to Enhance Reproducibility (HARPER) and The REporting of studies Conducted using Observational Routinely-collected Data statement for Pharmacoepidemiology (RECORD-PE) checklist, for race and ethnicity.

Results: Methods reproducibility is threatened by undocumented race and ethnicity variable definitions. Results reproducibility is uncertain when race and ethnicity data completeness are not reported in each study and compared between them, evaluating their impact on both selection bias and transportability of estimates. Inferential reproducibility is improved when authors describe the sociostructural factors that race and ethnicity variables are proxies for in causal structures of disease, treatments, outcomes, and their observability in real-world data, including the definition of the study population itself. Recent systematic reviews of race and ethnicity variables in epidemiology do not discuss research reproducibility, and no pharmacoepidemiology reproducibility guidance specifically addresses race and ethnicity.

Conclusions: Race and ethnicity variables are proxies for sociostructural processes fundamental to pharmacoepidemiology and thus its reproducibility. We propose RECORD-Race and Ethnicity to begin addressing the potentially wide-reaching influence of race and ethnicity variables in pharmacoepidemiology research reproducibility.