Formulating and Answering High-Impact Causal Questions in Physiologic Childbirth Science: Concepts and Assumptions

In this article, we conclude our 3-part series by focusing on several concepts that have proven useful for formulating causal questions and inferring causal effects. The process of causal inference is of key importance for physiologic childbirth science, so each concept is grounded in content related to women at low risk for perinatal complications. A prerequisite to causal inference is determining that the question of interest is causal rather than descriptive or predictive. Another critical step in defining a high-impact causal question is assessing the state of existing research for evidence of causality. We introduce 2 causal frameworks that are useful for this undertaking, Hill's causal considerations and the sufficient-component cause model. We then provide 3 steps to aid perinatal researchers in inferring causal effects in a given study. First, the researcher should formulate a rigorous and clear causal question. We introduce an example of epidural analgesia and labor progression to demonstrate this process, including the central role of temporality. Next, the researcher should assess the suitability of the given data set to answer this causal question. In randomized controlled trials, data are collected with the express purpose of answering the causal question. Investigators using observational data should also ensure that their chosen causal question is answerable with the available data. Finally, investigators should design an analysis plan that targets the causal question of interest. Some data structures (eg, time-dependent confounding by labor progress when estimating the effect of epidural analgesia on postpartum hemorrhage) require specific analytical tools to control for bias and estimate causal effects. The assumptions of consistency, exchangeability, and positivity may be especially useful in carrying out these steps. Drawing on appropriate causal concepts and considering relevant assumptions strengthens our confidence that research has reduced the likelihood of alternative explanations (eg bias, chance) and estimated a causal effect.