(055) Measuring the Use of TNF-alpha Inhibitors in Patients with Ulcerative Colitis: Optimization of Washout Period Duration and Observability

Background: Investigating incident medication use in claims-based database studies entails excluding patients with prior exposure to reduce prevalent user and misclassification biases. However, the optimal washout period duration (time without evidence of prior exposure) and required observability (time of continuous health insurance coverage before an incident treatment episode) remain uncertain.

Objectives: To examine the variability in sample size and characteristics of three cohorts of patients with ulcerative colitis (UC) initiating TNF-α inhibitors (TNFi) by varying the washout duration and observability.

Methods: Using data from the U.S. Optum CDM, we identified three adult cohorts of TNFi incident users with UC at different lines of advanced treatment: first-line bio-naïve (not exposed to any prior biologic), second-line prior TNFi switchers (switched from another TNFi), and second-line prior non-TNFi switchers (switched from other biologics such as vedolizumab or ustekinumab). We compared sample sizes, patient characteristics and exposure misclassification for one-year and anytime washout durations with one year observability. Within the anytime washout cohorts, we examined sample sizes and characteristics for up to three years of observability. In the switcher cohorts, the eligible time for prior prescription fill varied with washout durations. Standardized differences >10% were deemed meaningful.

Results: With a one-year washout, we found 5,754 TNFi initiators compared to 5,390 with an anytime washout (4,726 vs 4,356 bio-naïve, 233 vs 313 TNFi switchers, 795 vs 721 non-TNFi switchers, respectively). A one-year washout misclassified 8% bio-naïve, 7% TNFi switchers, and 14% non-TNFi switchers with prior ineligible biologic use. Additionally, 5% of TNFi switchers and 31% of non-TNFi switchers were missed as incident users with a one-year washout. Prior TNFi switchers were younger than prior non-TNFi switchers or bio-naïve (45 vs 51 vs 49 years, respectively) and had higher prevalence of cardiovascular disease (17.5% vs 14.4% vs 13.1%), anemia (46.7% vs 35.7% vs 34.8%), and pancolitis (53.5% vs 37.8% vs 31.6%). Similar differences were found in cohorts with 2- or 3-year observability, despite reduced sample sizes of 25% and 42.5%, respectively. Patients with longer observability, mostly non-TNFi switchers and bio-naïve, were older with more comorbidities, hospitalizations, and Medicare coverage than their counterparts with only one year of observability.

Conclusions: Adopting a longer washout reduces exposure misclassification, and requiring long observability may impact generalizability. Optimizing washout and observability is crucial for accurate and generalizable results in studying advanced treatment of UC.