Primary Submission Category: Longitudinal causal inference
Semi-parametric G-computation for Longitudinal Analysis of Antiretroviral Therapy
Authors: Andrew Spieker, Bryan Shepherd,
Presenting Author: Andrew Spieker*
The g-formula is a longitudinal generalization of standardization designed for settings in which there is time-varying treatment subject to time-dependent confounding. While the associated g-computation algorithm is straightforward and conceptually intuitive, its dependence upon parametric models is not ideal in practical circumstances in which variables possess distributions that are difficult to specify. We discuss the utility of cumulative probability models for use in g-computation as a way to relax certain forms of distributional assumptions. Simulations show this approach to be robust and feasible to implement in the real world. We illustrate the utility of this methodology through a study of core and ancillary agents comprising longitudinal antiretroviral therapy regimens and their effects on weight gain in a large cohort of persons living with HIV. Specifically, we hypothesize that modern integrase strand transfer inhibitors and tenofovir alafenamide are associated with greater mean weight gain as compared to other core and ancillary agents.