The exposure potential restriction rule revisited

Abstract

There are people who cannot receive certain treatments or experience certain exposures. For example, people without a uterine cervix cannot receive an intrauterine device. This lack of exposure potential in some persons instigated an interesting discussion in the 1980s regarding whether such persons should be included in case-control studies. A recommendation to exclude them was named the exposure potential restriction rule. We consider this rule in the context of current modern epidemiology and causal inference including clearly defining which causal questions can be answered with which assumptions, how exposure potential relates to the positivity assumption, how the exposure potential restriction rule may amplify uncontrolled confounding when the reason for a lack of exposure potential is an instrumental variable and the complementary idea of exposure compulsion. Using a simple simulation, we demonstrate that both restricting and not restricting on a variable that defines lack of exposure may induce bias depending on the causal structure. Therefore, careful thought must be used when deciding whether to remove participants who have no potential to be exposed or no potential to be unexposed.

Keywords: bias amplification; confounder adjustment; exposure potential; instrumental variables; positivity.