Causal Inference in Anesthesia and Perioperative Observational Studies

Purpose of ReviewObservational studies are of great importance to anesthesia and perioperative care research, as they reflect routine clinical practice. However, because observational data are nonexperimental, assigning causality to identified relationships has a significant risk of bias. After describing the pros and cons of observational studies, we provide an overview of the different methods used to make causal inferences. Of these, we focus on the propensity score analysis, which achieves an increasing popularity in anesthesia and perioperative literature.Recent FindingsSeveral methods are proposed for estimating treatment effects in observational studies. Although multivariable regression has traditionally been used to infer causal effects by adjusting for confounding variables, the reported result mainly depends on the model specification that fits the researcher’s hypothesis. Preprocessing observational data can reduce this model dependence, by balancing confounders across the treatment groups like in experimental studies. In particular, the propensity score analysis approximates the randomized controlled trial.SummaryCompared to randomized experiments, observational studies are low-cost sources of “real-life” data, but they are exposed to bias. Treatment effects can be estimated by using appropriate methods, such as the propensity score analysis, which limits confounding and model-dependence bias. We provide an illustrative example of propensity score analysis using a recently published study, which assessed the outcomes after hip fracture surgery compared with elective total hip replacement.