P3C202: Validation of Race, Ethnicity, and Language Data Collection in a Pediatric Emergency Department

Background: Stratifying institutional quality and performance metrics by race, ethnicity, and language (REaL) is a way for health systems to identify and address disparities in care. However, electronic medical record (EMR) data on patients’ race, ethnicity, and language is often incomplete, inaccurate, and not reported by the patient – the gold standard for collecting REaL data. We aimed to assess our own institution’s data validity by measuring concordance between patients’ self-reported and EMR-documented REaL data.

Methods: We conducted face-to-face surveys of 101 patients in our pediatric emergency department (PED) via a convenience sample. We used language disseminated by our health system to front line staff to survey patients on their self-identified race, ethnicity, and preferred language. Patients ≥ 18 years were surveyed directly. Parents of patients < 18 years provided their own preferred language but their child’s race and ethnicity data. We additionally collected EMR-documented REaL data for each patient. We used descriptive statistics to report aggregate patient demographics: race, ethnicity, language, and a combined race & ethnicity category. We then tested concordance between patients’ survey responses and EMR-documented REaL data using Cohen’s kappa. This study was approved by Columbia University’s Institutional Review Board.

Results: Of 101 surveyed patients, 35 (34.65%) reported their racial identity as Black, 14 (13.86%) White, 1 (0.99%) Asian/Pacific Islander (API), 1 (0.99%) American Indian/Alaskan Native (AIAN), and 50 (45.50%) other. Per the EMR, 22 (21.78%) identified as Black, 25 (24.75%) White, 44 (43.56%) other, and 10 (9.90%) declined to respond. There was 51% agreement between self-reported and EMR-reported race and a kappa score was 0.28 (p < 0.001). For ethnicity, 75 patients (74.26%) identified as Hispanic/Latino while 26 (25.74%) identified as non-Hispanic/Latino. In the EMR, 67 (67%) were Hispanic/Latino, 23 (23%) non-Hispanic-Latino, and 10 (10%) declined/unknown. There was 85% agreement; kappa 0.68 (p < 0.001). For language, 84 (83.17%) patients reported their preferred language as English, 16 (15.84%) as Spanish, and 1 (0.99%) French. In the EMR, 71 (70.30%) preferred English, 24 (23.76%) Spanish, 2 (1.98%) French, and 4 (3.96%) other. There was 81% agreement: kappa 0.50 (P < 0.001). Combining race/ethnicity into one category, 75 (74.26%) self-reported as Hispanic/Latino, 17 (16.83%) Black, 4 (3.96%) White, 1 (0.99%) API, and 4 (3.96%) other. In the EMR, 67 (70.53%) identified as Hispanic/Latino, 15 (15.79%) Black, 5 (5.26%) White, 2 (2.11%) other, and 6 (6.32%) other/unknown. There was 83% agreement, kappa 0.72 (p < 0.001).

Conclusion: Self-reported REaL data from patients in our PED showed varying levels of concordance with data in the EMR. Combining race and ethnicity into one category increased concordance over reporting race as a standalone metric. Future directions will focus on improving EMR data validity and ED workflows to ensure high quality data to measure and address healthcare disparities.