Attendance to screening and breast cancer mortality

Abstract

Mammographic screening is a method for detecting breast cancer at an early stage of the disease progression, and is thus considered a secondary prevention. The Norwegian Breast Cancer Screening Program invites all women aged 50-69 years to mammographic screening biennially. The benefit of attending screening is evaluated by estimates of reduced breast cancer mortality among women invited and women screened. We introduce causal inference to estimate the benefit of attending screening. As attendance to screening is optional, there is a possible selection bias when comparing women who never attend with women who regularly attend screening. We apply inverse probability weighting (IPW) to compensate for underlying differences between the comparison groups, under the assumption of no unmeasured confounders. A Cox proportional hazard model estimate the risk of breast cancer death to be 63% lower among the women regularly screened (HR=0.37, 95% CI: 0.32-0.43) compared to women never screened. The correction for selection bias by IPW does not have an impact on the estimated effect of mammographic screening, and we conclude that the covariates available do not introduce bias with respect to breast cancer mortality. Further, we are interested in the magnitude of selection bias. We consider an unknown factor representing any beneficial covariates associated with a reduced mortality. This factor needs to be 7 times stronger in women regularly screened compared to women never screened in order to eliminate the estimated effect of mammographic screening on breast cancer mortality. The aim of this thesis is to present a methodological framework in which we can estimate the effect of mammographic screening on breast cancer mortality including a possible effect of the women s screening history. We describe one possible way to achieve the true effect of screening and present the underlying assumptions. The assumption of no unmeasured confounders is violated in the currently available data, and we are thus not able to find the true effect of mammographic screening on breast cancer mortality in this thesis.