(123) Measuring Social Determinants of Health in Real-World Data: A Review of Pharmacoepidemiology Studies

Publication Number: 454

Background: Social determinants of health (SDOH) data are becoming increasingly important in studies using real-world data (RWD). However, it is unclear how disparities variables are operationalized and/or how these data are generated in insurance claims data.

Objectives: The aim of this study was to characterize the types of SDOH variables available in RWD, how these variables are generated, and whether the authors disclosed the data generation methods.

Methods: A structured search of published literature indexed in Google Scholar and PubMed was conducted. The specific search terms used were “disparities”, “United States”, “claims”, and “retrospective”. The literature search was restricted to articles that used real-world data sources in the United States, examined at least one SDOH variable, and were published between 2019 and 2021. Quality-control assessments verified summary data against the original articles to ensure accurate data extraction.

Results: We identified 35 articles that met the inclusion criteria. Race and ethnicity were the most frequent SDOH variables examined (n = 32), followed by socioeconomic status (n = 27) and urban/rural residence (n = 14). Among studies reporting race and ethnicity, 59% used a self-reported measure, 31% used an algorithm-derived measure, and 13% had an unknown method. Among studies examining urban/rural residence, only 1 study used self-reported data and 93% used an algorithm-derived measure most commonly based on location. Among studies with at least one socioeconomic status variable, use of proxy variables was the most common method (57%), followed by algorithm-derived measures (14%). Household income was the most common proxy for SES (43%), followed by Medicaid eligibility status (31%), education level (26%), and poverty level (17%). Discussion of data generation methods occurred least often for race and ethnicity, with only 34% disclosing how the variable’s data was generated, compared to 71% of papers with an urban/rural residence variable and 78% of papers with a socioeconomic variable.

Conclusions: There was significant heterogeneity in how data is generated for SDOH variables and inconsistent disclosure of data generation methods in the real-world data studies. While race and ethnicity were the most common social determinants examined, they had the least disclosure regarding how data is generated for the variable. Further discussion of data generation methods for SDOH variables in real-world databases is warranted.