UC Berkeley

In epidemiology, causal effects are often of interest within a post-exposure subgroup andthe exposure may affect membership into this subgroup. In the presence of censoring, loss to follow-up, non-compliance in randomized trials, or when exposure affects a precondition necessary for the outcome of interest to occur, subgroup effects may be particularly relevant for formulating causal parameters that answer meaningful questions about the specific mechanisms by which an exposure drives changes in an outcome. In this dissertation, we use the Causal Roadmap to introduce and illustrate the application of counterfactual strata effects in these types of settings. In perinatal studies, where an exposure may affect whether a live birth occurs and infant outcomes are only defined among live births, these types of counterfactual strata effects can be used to understand 1) the effect of an exposure on infant outcomes among live births if everyone were to receive the treatment as compared to the control (i.e., a counterfactual strata total effect) and 2) the effect of an exposure on infant outcomes among live births absent any exposure-induced changes to the composition of people who had live births and the timing of births (i.e., a counterfactual stratum direct effect).

Using the SEARCH Study as a motivating example, we first discuss the challenges of formulatingcausal effects in perinatal settings and then apply counterfactual strata effects to understand the impact of the SEARCH intervention on perinatal HIV transmission among women living with HIV at study baseline. Lastly, we describe how counterfactual strata effects can be more generally formulated to ask and answer questions when effects are of interest within post-exposure strata in simple settings with a single mediator of subgroup membership, as well as in more complex settings with multiple, possibly longitudinal mediators.