(B-209) Defining Cancer Outcomes Using Primary Care versus Hospitalization Data: A Case Study of GLP-1 RAs and Endometrial Cancer Incidence

Background: Identifying cancer outcomes using primary care or hospitalization data is imperfect compared to cancer registries and may introduce misclassification. While this has been documented, further research is needed to assess its impact on effect estimates in comparative studies of cancer incidence.

Objectives: Using glucagon-like peptide-1 receptor agonists (GLP-1 RAs) use on the incidence of endometrial cancer as an example, we compared the number of events, incidence rates, and hazard ratios (HRs) when using primary care versus hospitalization data to identify the cancer outcome.

Methods: We used data from the United Kingdom Clinical Practice Research Datalink (CPRD) linked to the Hospital Episodes Statistics Admitted Patient Care (HES APC) to assemble an active comparator new-user cohort of female patients with type 2 diabetes newly prescribed GLP-1 RAs or sulfonylureas (2007-2021), with follow-up until 2022. Propensity score fine stratification weighting balanced the exposure groups on potential confounders. We estimated weighted HRs and 95% confidence intervals (CIs) using Cox proportional hazards models using an on-treatment exposure definition. Endometrial cancer was identified with CPRD alone and HES APC alone in separate analyses.

Results: The cohort included 9,239 new users of GLP-1 RAs (median follow-up: 1.8 years) and 80,086 new users of sulfonylureas (median follow-up: 2.7 years). Using CPRD alone, we identified 11 and 152 endometrial cancer events in GLP-1 RA and sulfonylurea users (crude incidence rates [IRs]: 66.8 and 73.8 per 100,000 person-years, respectively). Using HES APC alone, we identified 20 and 313 events (crude IRs: 134.1 and 137.5 per 100,000 person-years, respectively). Weighted HRs were 1.08 (95% CI 0.50-2.33) using CPRD-defined outcomes and 1.12 (95% CI 0.65-1.93) using HES APC-defined outcomes.

Conclusions: While HES APC-defined outcomes resulted in nearly twice the number of events and higher incidence rates compared to CPRD-defined outcomes, the HRs remained similar. This indicates that if the outcome definition is highly specific, non-differential misclassification should have a minimal impact on relative risk estimates.