Kramer et al. respond to "Composite outcomes and paradoxes".

We thank Hernán et al. (1) for their thoughtful comments about the causal relationships among exposure, birth, and perinatal death; how selection bias can distort those relationships; and the role of composite outcomes. In our article (2), we do not claim composite outcomes to be a solution (“escape”) from the selection bias caused by conditioning on birth when analyzing the effect of pregnancy exposures on perinatal death (1). Rather, we discuss composites as an approach to analysis of competing risks. Composites and competing risks relate to the numerator events in risk expressions. As we explain in our article, the selection bias causing the crossover paradox is a problem of denominators, not numerators. Our 3 major points are these: