Pharmacoepidemiological studies of treatments for chronic inflammatory skin diseases
OA Version
Citation
Abstract
The rapid expansion of new immunomodulatory therapies for inflammatory skin diseases has intensified the need for timely, rigorous real-world safety evidence. While administrative claims data are a valuable source for generating rapid post-marketing evidence, pharmacoepidemiologic studies relying on these data face persistent methodological challenges, including limited sample sizes for newly approved therapies, unmeasured confounding by disease severity, and complex treatment patterns that complicate causal inference. This dissertation addresses these challenges through three interconnected aims that span study design methodology, bias quantification, and applied comparative safety research in psoriasis and atopic dermatitis.In the first aim, we compared the risk of infections and malignancy among patients with psoriasis initiating interleukin-17 inhibitors (IL-17i) versus tumor necrosis factor inhibitors (TNFi) using three pharmacoepidemiologic study designs: an active-comparator new user design, a hierarchical prevalent new user design, and a switcher new user design. No significant differences in serious infection, outpatient infection, or malignancy were observed across the three designs. The hierarchical prevalent new user design provided improved precision during the early post-approval years while yielding point estimates similar to the traditional new user design, supporting its use as a complementary approach for early post-marketing surveillance when new user sample sizes are limited.
The second aim employed a simulation-based quantitative bias analysis calibrated to the three Aim 1 cohorts to quantify the expected residual bias from unmeasured disease severity and to evaluate the effectiveness of proxy adjustment of unmeasured disease severity, such as use of non-biologic immunomodulatory medications at baseline. The proxy-confounder correlation emerged as the primary determinant of bias reduction, achieving up to 48% reduction in bias due to unmeasured disease severity at a proxy-confounder correlation of 0.7 but negligible improvement at 0.2. Under the most clinically plausible scenario, residual bias of 13–25% on the odds ratio scale persisted after proxy adjustment, providing practical benchmarks for interpreting claims-based comparative safety estimates that lack clinical severity data. The simulations also revealed that standard propensity score adjustment can occasionally increase bias through bias unmasking when measured and unmeasured confounders exert opposing effects.
The third aim applied these methodological insights to a comparative safety study of Janus kinase (JAK) inhibitors versus Th2 cytokine inhibitors among adults with atopic dermatitis, in relation to the risk of skin and soft tissue infections. JAK inhibitor use was associated with a 78% increased hazard of skin and soft tissue infections (HR = 1.78; 95% CI: 1.43–2.23), compared to Th2 cytokine inhibitors, with the excess risk emerging after 60 days of treatment. Young adults, males, and patients with greater baseline treatment burden were at highest risk. Marginal structural models with time-varying inverse probability of treatment weighting yielded a modestly higher estimate (HR = 1.98), suggesting that time-varying confounders partially masked the association.
Collectively, this dissertation contributes to methodological considerations in the design, analysis, and interpretation of claims-based comparative safety studies of immunomodulatory therapies for chronic inflammatory skin diseases and provides real-world evidence on the infection risk profile of JAK inhibitors in atopic dermatitis that may inform clinical monitoring strategies.
Description
2026