学术文献库

The attributable fraction among the exposed (\textbf{AF}$_e$), also known as the attributable risk or excess fraction among the exposed, is the proportion of disease cases among the exposed that could be avoided by eliminating the exposure. Understanding the \textbf{AF}$_e$ for different exposures helps guide public health interventions. The conventional approach to inference for the \textbf{AF}$_e$ assumes no unmeasured confounding and could be sensitive to hidden bias from unobserved covariates. In this paper, we propose a new approach to reduce sensitivity to hidden bias for conducting statistical inference on the \textbf{AF}$_e$ by leveraging case description information. Case description information is information that describes the case, e.g., the subtype of cancer. The exposure may have more of an effect on some types of cases than other types. We explore how leveraging case description information can reduce sensitivity to bias from unmeasured confounding through an asymptotic tool, design sensitivity, and simulation studies. We allow for the possibility that leveraging case definition information may introduce additional selection bias through an additional sensitivity parameter. The proposed methodology is illustrated by re-examining alcohol consumption and the risk of postmenopausal invasive breast cancer using case description information on the subtype of cancer (hormone-sensitive or insensitive) using data from the Women's Health Initiative (WHI) Observational Study (OS).