Primary Submission Category: Generalizability/Transportability
Transporting average causal effects with positivity violations
Authors: Paul Zivich, Jessie Edwards, Bonnie Shook-Sa, Eric Lofgren, Justin Lessler, Stephen Cole,
Presenting Author: Paul Zivich*
Transportability methods can be used to estimate causal effects from a biased sample of the target population. Transportability relies on a positivity assumption, such that all relevant covariate patterns in the target population also occur in the secondary population from which the sample was selected. Strict eligibility criteria, particularly in the context of randomized trials, can lead to violations of this assumption. Common methods to address nonpositivity are to restrict the target population, restrict the adjustment set, or extrapolate from a statistical model. Instead of these approaches, which all have concerning limitations, we propose a synthesis of statistical (e.g., g-methods) and mathematical (e.g., mechanistic) models. Briefly, a statistical model is fit for the regions of the parameter space where positivity holds, and a mathematical model is used to fill-in the nonpositive regions. For estimation, we propose two novel augmented inverse probability weighting estimators; one based on a marginal structural model, and the other based on the conditional average causal effect. The proposed methods are applied to estimate the effect of antiretroviral therapy on CD4 cell count among women with HIV. The synthesis approach addresses positivity violations when transporting and may be extended for other applications in causal inference more generally.