Abstract:In clinical research, the gold standard level of evidence is the randomized controlled trial (RCT). The availability of nonrandomized retrospective data is growing; however, a primary concern of analyzing such data is comparability of the treatment groups with respect to confounding variables. Propensity score matching (PSM) aims to equate treatment groups with respect to measured baseline covariates to achieve a comparison with reduced selection bias. It is a valuable statistical methodology that mimics the RCT, and it may create an "apples to apples" comparison while reducing bias due to confounding. PSM can improve the quality of anesthesia research and broaden the range of research opportunities. PSM is not necessarily a magic bullet for poor-quality data, but rather may allow the researcher to achieve balanced treatment groups similar to a RCT when high-quality observational data are available. PSM may be more appealing than the common approach of including confounders in a regression model because it allows for a more intuitive analysis of a treatment effect between 2 comparable groups.We present 5 steps that anesthesiologists can use to successfully implement PSM in their research with an example from the 2015 Pediatric National Surgical Quality Improvement Program: a validated, annually updated surgery and anesthesia pediatric database. The first step of PSM is to identify its feasibility with regard to the data at hand and ensure availability of data on any potential confounders. The second step is to obtain the set of propensity scores from a logistic regression model with treatment group as the outcome and the balancing factors as predictors. The third step is to match patients in the 2 treatment groups with similar propensity scores, balancing all factors. The fourth step is to assess the success of the matching with balance diagnostics, graphically or analytically. The fifth step is to apply appropriate statistical methodology using the propensity-matched data to compare outcomes among treatment groups.PSM is becoming an increasingly more popular statistical methodology in medical research. It often allows for improved evaluation of a treatment effect that may otherwise be invalid due to a lack of balance between the 2 treatment groups with regard to confounding variables. PSM may increase the level of evidence of a study and in turn increases the strength and generalizability of its results. Our step-by-step approach provides a useful strategy for anesthesiologists to implement PSM in their future research.

On Principles for Modeling Propensity Scores in Medical Research

Propensity Score Method: a Non-Parametric Technique to Reduce Model Dependence

Using Propensity Scores to Help Design Observational Studies: Application to the Tobacco Litigation

Using Propensity Scores to Estimate Effects of Treatment Initiation Decisions: State of the Science

Characterizing the Effect of Matching Using Linear Propensity Score Methods with Normal-Distributions

Use of Propensity Scoring and Its Application to Real‐World Data: Advantages, Disadvantages, and Methodological Objectives Explained to Researchers Without Using Mathematical Equations

High-dimensional propensity scores for empirical covariate selection in secondary database studies: Planning, implementation, and reporting

Estimating the Propensity Score

Estimating and Using Propensity Scores with Partially Missing Data

Taking Causality Seriously: Propensity Score Methodology Applied to Estimate the Effects of Marketing Interventions

Propensity score matching: a tool for consumer risk modeling and portfolio underwriting

Should Observational Studies Be Designed to Allow Lack of Balance in Covariate Distributions Across Treatment Groups?

Propensity score analysis for time-dependent exposure

Matching Using Estimated Propensity Scores: Relating Theory to Practice

High-Dimensional Propensity Score and Its Machine Learning Extensions in Residual Confounding Control

Five Steps to Successfully Implement and Evaluate Propensity Score Matching in Clinical Research Studies

Estimating Causal Effects from Large Data Sets Using Propensity Scores

Bayesian propensity scores for high-dimensional causal inference: A comparison of drug-eluting to bare-metal coronary stents

Applying Propensity Score Methods in Clinical Research in Neurology

Propensity scores using missingness pattern information: a practical guide

Propensity score‐based methods for causal inference and external data leveraging in regulatory settings: From basic ideas to implementation