Misclassification of race/ethnicity in a Population-based Cancer Registry (United States)

Cancer registry data on race/ethnicity are vital for understanding cancer patterns in population subgroups, as they inform public health policies for allocating resources and form the bases of etiologic hypotheses. However, accuracy of cancer registry data on race/ethnicity has not been systematically evaluated. By comparing race/ethnicity in the Greater Bay Area Cancer Registry to self-reported race/ethnicity for patients from 14 racial/ethnic groups, we determined the accuracy of this variable and the patient and hospital characteristics associated with disagreement. The extent of misclassification (measured by sensitivity and predictive value positive (PV+)) varied across racial/ethnic groups (total n=11,676). Sensitivities and PV+’s were high (exceeding 90%) for non-Hispanic Whites and Blacks, moderate for Hispanics and some Asian subgroups (70–90%), and very low for American Indians (<20%). Overall, registry and interview race/ethnicity disagreed for 11% of the sample. In a multivariate model, disagreement was associated with non-White race/ethnicity, younger age, being married, being foreign-born but preferring to speak English, and diagnosis in a large hospital. Improving data quality for race/ethnicity will be most effectively attempted at the reporting source. We advocate a concerted effort to systematize collection of these patient data across all facilities, which may be more feasible given electronic medical admissions forms.