How do we learn what works? A two-step algorithm for causal inference from observational data

Making decisions among several courses of action requires knowledge about their causal effects on health outcomes. Randomized experiments are the preferred method to quantify those causal effects. When randomized experiments are not feasible or available, causal effects are often estimated from observational databases. Therefore, causal inference from observational data can be viewed as an attempt to emulate a hypothetical randomized experiment—the target trial—that would quantify the causal effect of interest. Through several examples, this talk outlines a general algorithm for causal inference using observational databases that makes the target trial explicit. This causal framework channels counterfactual theory for comparing the effects of sustained treatment strategies, organizes analytic approaches, provides a structured process for the criticism of observational analyses, and helps avoid common methodologic pitfalls.