Validation of Self-Reported Race in a Canadian Provincial Renal Administrative Database

Abstract

Background: Administrative data are commonly used to study clinical outcomes in renal disease. Race is an important determinant of renal health delivery and outcomes in Canada but is not validated in most administrative data, and the correlation with census-based definitions of race is unknown. Objectives: Validation of self-reported race (SRR) in a Canadian provincial renal administrative database (Patient Records and Outcome Management Information System [PROMIS]) and comparison with the Canadian census categories of race. Design: Prospective patient survey study to validate SRR in PROMIS. Setting: British Columbia, Canada. Patients: Adult patients registered in PROMIS. Measurements: Survey SRR was used as gold standard to validate SRR in PROMIS. Self-reported race in PROMIS was compared with census race categories. Methods: This is a cross-sectional telephone survey of a random sample of all adults in PROMIS conducted between February 2016 and November 2016. Responders selected a race category from PROMIS and from the Canadian census. Sensitivity (Sn) and specificity (Sp) were calculated with 95% confidence intervals (CIs). Results: A total of 21 039 patients met inclusion criteria, 1677 were selected for the survey and 637 participated (38% response rate). There were no differences between the PROMIS, sampled, and responder populations. PROMIS SRR had an accuracy of 95.3% (95% CI: 94.2%-97.0%) when validated against the survey SRR with Sn and Sp ≥90% in all race groups except in Aboriginals (Sn 87.5%). The positive and negative predictive values were ≥95%, except in very low and high–prevalence groups, respectively. The Canadian census had an accuracy of 95.7% (95% CI: 94.4%-97.6%) when validated against PROMIS SRR with Sn and Sp ≥90%. The results did not differ in subgroups based on age, sex, birth outside Canada, or renal group (glomerulonephritis, chronic kidney disease, hemodialysis, peritoneal dialysis, transplant recipients, or live donors). Limitations: Analysis of minority groups and lower prevalence groups is limited by sample size. Results may not be generalizable to other administrative databases. Conclusions: We have shown high accuracy of PROMIS SRR that validates its use in the secondary analysis of administrative data for research. There is high correlation between PROMIS and census race categories which allows linkage with other data sources that use census-based definitions of race.