Directed acyclic graph showing the relation between placental abruption (X) and perinatal mortality (Y) with preterm delivery as the intermediary (mediating) covariate (M), US singleton births, 1995–2002. “C” is a set of confounders observed at baseline (common causes for placental abruption, preterm delivery, and perinatal mortality), and “U” is a set of unobserved confounders. We represent the relation between X and U as a dashed line because we do not consider this confounding scenario; consequently, our sensitivity analysis assumes that this pathway does not exist, and that we believe this confounding is weak in comparison.