G Computation for Estimating the Per Protocol Effect of Aspirin on Pregnancy Outcomes
While randomized trials are often considered to be the "gold standard" for estimating treatment effects, they are subject to complications when participants do not comply with assigned treatment categories. Non-compliance with study protocol is particularly challenging to deal with in pragmatic trials, when treatment consists of a series of actions over a long period. Standard "as treated" or "per protocol" approaches can be inconsistent due to post-randomization confounding and selection bias. Here, we show how the g computation algorithm (Robins 1986 Mathematical Modelling; 7:1393-1512) can be used to estimate per protocol effects in the presence of time-dependent post-randomization confounding using data from a trial of the effect of low dose aspirin on pregnancy outcomes.