Methods and Biology in an Epidemiology of Consequence
In this issue of the Journal, Kinlaw et al.1 consider issues relevant to studies of the relationship between cigarette smoking and risk of preeclampsia. Noting that simulation studies are limited by the realism of the underlying assumptions, Kinlaw et al.1 revisit the findings of a previously published simulation study by Lisonkova and Joseph,2 focusing on the biologic uncertainty about the relation between abnormal placentation and preeclampsia. To address this issue, the authors reconsider the influence of collider stratification via their own simulation study and use a similar assumed causal structure but with a broader parameterization based on published literature. The new simulations indicate that the magnitude of the bias is less than previously suggested. The authors conclude that increased risk of pregnancy loss related to smoking is not sufficient to explain the observed protective association of smoking with preeclampsia risk. This conclusion begs the question, what does explain it: true biologic effects or other, unidentified biases? The purpose of this commentary is to explore questions raised by this conclusion and the related discussion.