Bo Zhang

PhD

Fred Hutchinson Cancer Center

Assistent Professor of Biostatistics

PLCO (Learn more about this study)

PLCO-1336

Sep 28, 2023

Estimating and generalizing the complier average treatment effect in the PLCO trial

PLCO study shows that compared to the standard-of-care group, regular screening with flexible sigmoidoscopy could reduce the risk of colorectal cancer by 21%. The intention-to-treat (ITT) effect reported in the primary analysis was diluted by the noncompliance. The ITT can be translated into the complier average treatment effect by dividing it by the compliance rate: for instance, under an overall compliance rate of 80%, the treatment effect among those who complied with the treatment assignment would be 26%. The complier average treatment effect describes the actual treatment effect on a subgroup of study participant, rather than the intention-to-treat effect on the entire cohort. However, it's not clear whether the complier treatment effect can be generalized to the whole population, which has prevented it from being more widely used in the clinical trials settings.
The primary goal of this project is to estimate the treatment effect of regular sigmoidoscopy among compliers and investigate if this complier treatment effect can be generalized to the entire study cohort. To achieve this, we will formalize the clinical trial with noncompliance under an instrumental variable (IV) framework and develop a new statistical methodology to deal with noncompliance issues and extrapolate causal effects in randomized trials. We noticed that in the PLCO study, three screening centers switched from a dual consent process to a single consent process, which resulted in an improvement in compliance rate. We conceptualize this change in policy as strengthening the encouragement to adhere to the prescribed treatment. The treatment effect among participants who should have complied with the single consent process but didn't comply with the dual process may contain additional information for generalizability of the treatment effect. In this study, we formulated this problem in a formal causal framework. We defined the always-complier (the latent population that would comply in both consent processes) treatment effect and switcher (the latent population that would switch from non-complier under the dual consent process to complier under single consent process) treatment effect using principal stratification. We will prove identifiability for always complier treatment effect and switcher treatment effect under additional assumptions. We will also develop a new efficient estimation method and a new testing method for the homogeneity of the treatment effect among always complier and switchers. We plan to apply these methods on the PLCO data and estimate the actual treatment effect, rather than the intention-to-treat effect, of flexible sigmoidoscopy.