Senior Director, Vaccines RWE Pfizer Durham, United States
Background: Algorithms are frequently used to identify immunocompromised (IC) patients in real-world data, but there is limited research on how algorithm definition impacts study results or sample sizes.
Objectives: To compare three claims-based definitions of IC status by evaluating differences in baseline characteristics and demonstrating impact of definition selection on estimated sample size for trial enrollment and incidence of outcomes, using herpes zoster (HZ) as a case example.
Methods: Using Optum’s de-identified Clinformatics® Data Mart Database, we identified adults aged 50-54 years, enrolled on January 1, 2019 (index), and continuously enrolled for 1 year prior to index (baseline). Patients were included in any of the three IC cohorts if the patient met the corresponding IC definition. Definition 1 was defined as having ≥ 1 claim for immunosuppressant (IS) medication or HIV/AIDs, hematologic malignancy, transplant, or primary immunodeficiency diagnoses. Definition 2 included intrinsic immune deficiency, CKD/ESRD/dialysis, autoimmune diseases, or solid malignancy diagnoses in addition to conditions in Definition 1. Definition 3 included the four diagnoses from Definition 1, but imposed different time windows and required IS medication use for specific conditions. To ascertain HZ outcomes, patients were followed from index until the earliest of HZ diagnosis; disenrollment; death or December 31, 2019. The unadjusted incidence rate (IR) of HZ and 95% confidence intervals (CIs) for 2019 were calculated for patients with no HZ diagnosis 90 days before index and reported as events per 1,000 person-years.
Results: We identified 36,973 patients as IC with Definition 1, 88,235 with Definition 2, and 23,892 with Definition 3. The most prevalent IC condition for Definition 1 was IS medication (85.1%); Definition 2 was autoimmune diseases (61.2%); and Definition 3 was solid malignancies with treatment (48.1%). Baseline demographics, healthcare utilization, and chronic conditions were comparable, except for prevalence of autoimmune diseases (Definition 1: 50.6%, Definition 2: 61.2%, Definition 3: 19.0%). The incidence of HZ was highest for Definition 1 (IR 14.29 (95% CI: 13.11-15.57)), followed by Definition 3 (IR 13.49 (95% CI: 12.08-15.07)). The rate was lowest for Definition 2 (IR 12.04 (95% CI: 11.34-12.80)).
Conclusions: Estimated sample sizes differed significantly based on the IC definition selected. The lower incidence rate for Definition 2 was likely attributed to inclusion of diagnoses without IS treatment. While the appropriate IC definition depends on the therapeutic area and the research question being asked, researchers should carefully assess the impact on sample size and outcome incidence when selecting IC definitions.
