(20) Representativeness in Clinical Trial Planning: An Epidemiologic Examination of Race and Ethnicity Leveraging Multiple Real-World Data (RWD) Sources in The United States

Publication Number: 694

Background: Clinical trials often enroll populations that do not reflect the diversity of the population who are potentially eligible to participate.

Objectives: To demonstrate a process that evaluates US data sources from which diversity targets for clinical trials can be generated to improve representation of patients, specifically race and ethnicity (R/E).

Methods: Data and literature were evaluated based on population representativeness, sample size, recency or study period, sensitivity/ability to identify the indication, granularity of R/E categories, completeness of data collected, known and potential biases. Each R/E category’s share of the total prevalent/incident case population was generated and expressed as a range of proportions, considering the potential limitations and/or biases of each RWD source specific to each of the 27 indications evaluated. Demographic distributions may be applied to US census data to estimate the real-world population, where appropriate.
The following sources were queried for R/E per indication of interest:
1. Longitudinal patient electronic medical records systems (EMRs); Integrated Health Network EMR and Ambulatory EMR; indications defined by ICD10 diagnosis codes and established coding algorithms
2. Government sources of publicly available electronic datasets including CDC’s cross-sectional household survey: National Health Interview Survey
3. Publications were graded and ranked using CDC Principles of Epidemiology in Public Health Practice, Strengthening the Reporting of Observational studies in Epidemiology (STROBE) Initiative, and Good Research for Comparative Effectiveness (GRACE) Principles

Results: The consistency of R/E distributions between RWD sources varied by each indication but demonstrated reliability based on the known data characteristics. For example, R/E distributions for acute events (eg, stroke) tended to be adequately captured in EMR, yet in some cases comprehensive estimates were available in literature and CDC data (eg, diabetes). Critical factors include inpatient/outpatient records, sample size and sample population, access and treatment disparities resulting in underrepresentation in healthcare system data, and differences in data capture (eg, categories and definitions).

Conclusions: RWD can be used to support clinical trial diversity goals, but there are strengths and limitations specific to the data source and indication under investigation. Decision-making relies on both quantitative and qualitative evaluations of R/E diversity that is indication specific. Estimates in the scientific literature are also important to establish background and contextualization of R/E estimates through a synthesis of multiple studies or cohorts.