On Estimation of Time-varying Population Attributable Fractions for Population-based Case-control Studies

Abstract

The population attributable fraction (PAF) is an important measure in evaluating the contribution of risk factors to disease burden or mortality. It is a useful tool for planning of prevention actions. Recent development of PAF has extended the classic static measure to a more dynamic time-varying functional form, which may provide more details of the underlying quantity. This dissertation focuses on estimation of the time-varying PAF for population-based case-control studies. We consider both situations with or without adjustment for confounders. The underlying model is assumed to be the Cox model with possibly time-varying covariates. We propose three kernel-type estimators of the time-varying PAF for case-control studies. We study large sample properties of these estimators, and prove strong uniform consistency and asymptotic normality. We also study the finite sample properties of these estimators, and derive variance estimators with corrections for the finite sampling. We also propose a resampling based approach for calculating the simultaneous confidence bands. Simulation studies show that all the point and variance estimators perform well under all the designed scenarios. We apply our proposed methods the Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO) smoking data to understand the risk of colorectal cancer attributable to smoking.