Description
- Abstract:
- Estimating the average treatment effect (ATE) in the target population is often a main interest of epidemiologic studies. I consider three different ways to estimate ATE and I deal with a remaining issue of each approach. Three ways are: 1) Transport inferences of randomized trials to the target population directly when there is no evidence of heterogeneity of treatment effects (HTE), 2) Transport inferences of randomized trials to the target population adjusting different distribution of effect modifiers between trial population and the target population, 3) Estimate causal effects in the target population using observational studies. Assessing HTE is necessary to carry out the first approach and we need to understand different ways of HTE assessment. In the first chapter, we propose that one way of assessing HTE (outcome score methods) cannot find HTE or classify population by treatment benefit properly when outcome score and true conditional average treatment effects are independent. The outcome score methods fail to detect HTE even if there are significant amount of heterogeneity exists in population. We hypothesize another approach (effect score methods) does not have problems under the same condition. I demonstrate our hypothesis by extensive simulation studies and applied the methods to a randomized trial (ALLHAT) for an empirical application. Missing data is a common issue in epidemiologic studies and this is a main topic of chapter 2 and 3. In chapter 2, I review the study design of transporting inferences of randomized trials to the target population of all trial-eligible individuals and propose how to identify ATE when some covariates have missingness using inverse probability weighting. I focused on the situation when treatment and outcome of non-randomized participants are not observable and discuss identification of ATE and assumptions of missing data mechanism. I illustrate the method using the CASS study. In chapter 3, I review missing data methods, causal inference methods and how to combine them to estimate ATE in observational studies from a comprehensive viewpoint. In simulation studies, I compare finite sample performance of various estimators that are derived by combining missing data methods and causal inference methods. I also address up-to-date issues of the topic: comparison of Across and Within approach of multiple imputation and implementation of augmented inverse probability weighting estimators. The CASS study is used to illustrate the methods.
- Notes:
- Thesis (Ph. D.)--Brown University, 2021
Citation
Kim, Hongseok,
"Topics in Causal Inference in Epidemiological Studies: Assessing Heterogeneity of Treatment Effects & Addressing Missing Covariates"
(2021).
Epidemiology Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://repository.library.brown.edu/studio/item/bdr:exy3spzh/
