Methods for time-varying exposures

Abstract

Much research has been conducted into methods for estimation of time-varying exposure effects. However, it is yet unknown to what extent these methods have found their way into current research practice and how large their impact is, in terms of more valid and more accurate exposure effect estimates. Furthermore, although IPW is the most frequently used of the above mentioned methods to deal with time-varying confounding affected by prior exposure, clear recommendations on how to handle large IPW weights, application of IPW in studies of continuous exposures, and IPW model selection (e.g., how to deal with rare exposures and/or nonpositivity) are limited. In addition, it is unknown to what extent features of the estimated IPW (e.g., its range or variance) are informative about the quality of the weights and their ability to control for confounding. This thesis aims to address these knowledge gaps. Therefore, the two main objectives of this thesis are: 1) To evaluate methods to estimate causal effects of time-varying exposures; 2) To evaluate methods to deal with time-varying confounding in observational studies of timevarying exposures.