Invited commentary: composite outcomes as an attempt to escape from selection bias and related paradoxes

This commentary reviews the recent history of explanations to crossover paradoxes such as the birth weight and gestational age paradoxes, with a special emphasis on the current proposal by Kramer et al. in this issue of the Journal (Am J Epidemiol. 2014;179(3):361-367). We contend that the causal structure of these paradoxes is essentially identical to that of several well-known selection biases. We then consider the pros and cons of using composite outcomes to circumvent these selection biases.