Causal Inference In Cluster-Randomized Trials and Observational Studies With Partial Interference

Vaccine effects or other health-related treatments are important to the field of public health. Causal effects can go beyond simple association to determine whether a treatment is effective in reducing a disease, for example. In infectious diseases, one person's treatment may affect another individual's outcome. This is known as interference. Causal inference with interference can be a powerful tool in the benefits of vaccines or other treatments. This work considers methods for drawing inference about causal effects in cluster-randomized trials and observational studies in the presence of interference. Cluster-randomized trials are often conducted to assess vaccine effects. Defining estimands of interest before conducting a trial is integral to the alignment between a study's objectives and the data to be collected and analyzed. The first paper considers estimands and estimators for overall, indirect, and total vaccine effects in trials where clusters of individuals are randomized to vaccine or control. The scenario is considered where individuals self-select whether to participate in the trial and the outcome of interest is measured on all individuals in each cluster. Unlike the overall, indirect, and total effects, the direct effect of vaccination is shown in general not to be estimable without further assumptions, such as no unmeasured confounding. An illustrative example motivated by a cluster-randomized typhoid vaccine trial is provided. In the setting of observational studies with partial interference, inverse probability weighted estimators have previously been developed. Unfortunately, these estimators are not well suited for studies with large clusters. Therefore, in the second paper, the parametric g-formula is extended to allow for partial interference. G-formula estimators are proposed of overall effects, spillover effects when treated, and spillover effects when untreated. The proposed estimators can accommodate large clusters and do not suffer from the g-null paradox that may occur in the absence of interference. The large sample properties of the proposed estimators are derived, and simulation studies are presented demonstrating the finite-sample performance of the proposed estimators. The Demographic and Health Survey from the Democratic Republic of the Congo is then analyzed using the proposed g-formula estimators to assess the overall and spillover effects of bed net use on malaria. In the third paper, g-estimation is extended to the case of partial interference where different treatment policies are of interest. This partial interference setting means that individuals within a cluster may interfere with one another, but they cannot interfere with individuals in other clusters. In this setting, prior work has focused on inverse probability weighting and the parametric g-formula. However, inverse probability weighting does not handle large cluster sizes well. The parametric g-formula relies upon a correctly specified outcome model. G-estimation is able to handle larger clusters and is not subject to the g-null paradox, providing an alternative method for this setting. Additionally, g-estimation is doubly robust and is thus more robust to model misspecification than the parametric g-formula. G-estimators of overall effects, spillover effects when treated, and spillover effects when untreated are considered. The large sample properties of the proposed estimators are derived using estimating equation theory. A set of simulation studies are presented to demonstrate the finite-sample performance of the proposed estimators. The 2013-14 Demographic and Health Survey in the Democratic Republic of the Congo is analyzed to determine the causal effect of bed net use on malaria.Doctor of Philosoph