Primary Submission Category: Generalizability/Transportability
Recovering target causal effects from post-exposure selection induced by missing outcome data
Authors: Johan de Aguas, Johan Pensar, Tomás Varnet, Guido Biele,
Presenting Author: Johan de Aguas*
Two significant challenges to the validity of causal claims are confounding bias and selection bias. The latter can arise through informative missingness, where partial information about units in the population is missing, censored, or coarsened due to factors related to the exposure, the outcome, or their consequences. We extend existing graphical criteria to address selection bias induced by missing outcome data by leveraging post-exposure variables. We introduce the generalized adjustment criteria with post-exposure variables (GACPE), which supports a recovered causal effect based on sequential regressions. A refined estimator is further developed by applying targeted minimum loss estimation (TMLE). Under certain regularity conditions, this estimator is multiply-robust, ensuring consistency even in scenarios where inverse probability weighting (IPW) and the sequential regressions approach fall short. A simulation study with various scenarios contrasts the relative robustness and efficiency of the two proposed solutions against other classical estimators. As a motivating application case, we study the effects of pharmacological treatment for attention-deficit/hyperactivity disorder (ADHD) upon the scores of national tests taken by diagnosed Norwegian schoolchildren. Findings support the accumulated clinical evidence affirming a positive but small effect of stimulant medication on school performance.