UCLA

Stepped wedge design (SWD) trials are cluster randomized trials that feature staggered, unidirectional cross-over between treatment conditions. Existing literature for SWDs focuses primarily on designs with two conditions, typically a control and an intervention condition, and a continuous outcome. The work for this dissertation is motivated by the NORVAX study, a SWD trial implemented at clinics in a safety-net health system to estimate the effectiveness of two interventions for promoting HPV vaccination among adolescents. The outcome for the NORVAX study is patient vaccination status, which is a multistate outcome (no doses, one dose, or two doses). This dissertation has two parts that make contributions to the literature regarding two salient features of the NORVAX study: the multiple interventions in a SWD and the multistate vaccination status outcome.

The first part of this dissertation develops methods for conducting power calculations for SWDs with multiple treatment conditions and a continuous outcome. We present a linear mixed model for such designs and derive standard errors of the intervention effect coefficients. Power for detecting intervention effects is calculated analytically assuming a normally distributed Wald test statistic under an alternative hypothesis. We apply the proposed method to both repeated cross-sectional and cohort designs. Design features, with a focus on treatment sequencing across periods, are examined to determine their impact on power. Simulations are used to verify results. %For factorial designs, power is maximized by allowing clusters to transition to several different treatment conditions. Power is maximized in multi-arm trials by ensuring each treatment condition is introduced early enough in the SWD.

The second part of this dissertation focuses on the vaccination status outcome and quantifying intervention effects for this outcome within the context of a SWD. A goal of the NORVAX study is to estimate intervention effects as changes in study population-level vaccination initiation and completion percentages, clinically meaningful outcomes. We propose a semi-Markov multistate cure model in which the number of doses of a vaccine received by the patient are the states. Sojourn times are assumed to be Weibull distributed. To account for individuals who will never receive their next required dose, we include cure proportions in the multistate model. Using the multistate cure model framework, population-level initiation and completion percentages are obtained by converting transition intensity estimates into transition probabilities. Intervention effects are quantified as changes in initiation and completion percentages attributable to interventions. We apply the model to both simulated and real-world data and highlight challenges of this modeling technique.