Demographic Variations in Diabetic Retinopathy Cohorts using Electronic Health Records-based Definitions in All of Us

Abstract

Purpose : Secondary use of electronic health record (EHR) data has accelerated the scale of retrospective data analyses. While billing codes such as the International Classification of Diseases (ICD) are often extracted from EHRs to select disease cohorts, variations exist in how investigators use these codes for cohort selection, which may be problematic when assessing generalizability and comparability of study findings. Using a selection of patients with diabetic retinopathy (DR) from the All of Us dataset as our use case, we aimed to assess variations in demographic characteristics between cohorts identified employing various codified definitions of DR.

Methods : We reviewed the literature and identified retrospective studies using ICD-10 codes to define cohorts of 1) incident DR at any severity and 2) DR with diabetic macular edema (DME). Using these definitions, we generated cohorts for each disease subset in All of Us and extracted data on patient gender, race, ethnicity, and age. Statistically significant differences between cohorts were were assessed using ANOVA or a Student’s T-test for continuous variables, and Pearson’s chi-squared tests for categorical variables, with p ≤ 0.05 thresholded for significance.

Results : Overall, 3 unique cohorts were identified for DR at any severity, and 2 unique cohorts were identified for the DME group (Table 1). For the 3 cohorts of DR at any severity, the varying definitions yielded variability in sample sizes: 3856, 3636, and 2241 patients. For DME, one cohort definition yielded almost double the sample size of the other (1214 vs.618). There were also significant variations in the distribution of racial groups among the cohort definitions for DR (p=0.004; Table 2).

Conclusions : Numerical and demographic variations exist between EHR-based cohorts defined for patients with DR in published literature. This has important implications regarding the variability of cohorts for diseases defined across various retrospective studies, which may affect study power, generalizability, reproducibility, and comparability of study conclusions. More work is needed to evaluate how cohorts for diseases such as DR can be defined in a standardized manner.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

Table 1. Patient cohorts defined by ICD-10 codes.

 

Table 2. Demographic characteristics of cohorts defined by ICD-10 codes.