Measuring Ethnicity and Race Using the Common Data Model: Implications for Precision in Health Disparities Research

Abstract

Research ObjectiveInconsistencies in patient health record data increase bias by marginalizing individuals whose self‐identified race or ethnicity do not map to politically designated categories. A solution to improving measurement of race and ethnicity in patient health records is to use common data models (CDM). CDMs use a standardized language to map patient data across several sources. Recently, the Veterans Health Administration (VHA) implemented a CDM that aggregates both self‐ and observer‐recorded race and ethnicity into single indicators. Demonstration that CDMs accurately reflect self‐reported race and ethnicity can improve identification of minority groups and foster enhanced care.The current study's main objective was as follows: (a) to examine the accuracy of race and ethnicity as indicated by the VHA’s CDM against self‐reported race and ethnicity; and (b) to examine the utility of a classification schema prioritizing ethnicity for disparities work.Study DesignRace and ethnicity data from VHA CDM and from a survey examining patient engagement were used for the main analysis (total N = 9176; Hispanic or Latinx (H/L) n = 1415; non‐Hispanic (NH) black n = 2455; non‐Hispanic (NH) white n = 5306). Data were compared using IF, THEN statements in three classification schemas in SAS v. 9.4. Values of 0.8 (ie, 80%) or above indicated that the VHA CDM was an excellent indicator of self‐reported race and ethnicity. Measures included sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and agreement.Population StudiedA national sample of Veterans from a larger study about patient engagement with VHA care (N = 9176).Principal FindingsIn Schema 1 (ie, H/L > NH‐black > NH‐white), all agreement values were above 80%. Schema 1 resulted in highest values across all indicators for H/L individuals (ie, values greater than 87.21%) and highest nonsensitivity values across all other indicators for NH‐black and NH‐white individuals, compared with Schema 2 (ie, NH‐black > H/L > NH‐white) and Schema 3 (ie, NH‐white > H/L > NH‐black). In Schema 3, VHA CDM only correctly predicted self‐reported H/L ethnicity 56% of the time. In both Schemas 2 and 3, VHA CDM only reliably categorized 11%‐12% of NH‐white and 7%‐13% of NH‐black individuals from self‐report data.ConclusionsThe VHA CDM accurately represents the racial and ethnic identity of a large sample of Veterans when self‐reported H/L ethnicity is prioritized over race. Data from CDMs can supplement self‐report data by replacing inconclusive responses. This is critical, as responses from individuals whose self‐identified race or ethnicity do not map to preset categories are generally unusable for health disparities analyses. Measuring health disparities across health systems can only be achieved by standardizing the language used in patient health records.Implications for Policy or PracticeThis study is one of the first to measure the accuracy of social data from a Common Data Model and can inform future efforts to improve these data across health systems.Primary Funding SourceDepartment of Veterans Affairs.