Decreasing missingness in race and ethnicity data by inclusion of preferred language for mapping to aggregate categories.

Background: Accurate and complete patient race and ethnicity data are essential for informing health care quality and patient safety initiatives. However, missing data remain a persistent issue. We aimed to explore the utility of preferred language to impute patient race and ethnicity. Methods: This was a retrospective analysis from 3 emergency departments in New York City, from June 1, 2023, to May 31, 2024. We leveraged a mapping schema for imputation of missing race and ethnicity data using preferred language for categorization into the Office of Management and Budget's 7 categories. We examined concordance between preferred language and predicted categories. Results: The proportion of patients with missing race and ethnicity data decreased from 9.7% to 8.6%, reducing missingness by 11.1%. The greatest proportion of change with the use of preferred language was for Middle Eastern and North African patients (14.7%). Conclusion: Our findings support that language-based imputation has the potential to reduce missing race and ethnicity data and may be a helpful tool in quality improvement and research efforts. For health systems where race and ethnicity fields may not be fully detailed or have a high rate of missing data, the use of language may serve as a valuable adjunct in improving the comprehensive picture of a population.