Propensity scores as a basis for equating groups: Basic principles and application in clinical treatment outcome research.

Abstract

A propensity score is the probability that a participant is assigned to the treatment group based on a set of baseline covariates. Propensity scores provide an excellent basis for equating treatment groups on a large set of covariates when randomization is not possible. This article provides a nontechnical introduction to propensity scores for clinical researchers. If all important covariates are measured, then methods that equate on propensity scores can achieve balance on a large set of covariates that mimics that achieved by a randomized experiment. We present an illustration of the steps in the construction and checking of propensity scores in a study of the effectiveness of a health coach versus treatment as usual on the well-being of seriously ill individuals. We then consider alternative methods of equating groups on propensity scores and estimating treatment effects including matching, stratification, weighting, and analysis of covariance. We illustrate a sensitivity analysis that can probe for the potential effects of omitted covariates on the estimate of the causal effect. Finally, we briefly consider several practical and theoretical issues in the use of propensity scores in applied settings. Propensity score methods have advantages over alternative approaches to equating groups particularly when the treatment and control groups do not fully overlap, and there are nonlinear relationships between covariates and the outcome.

Neutrophil to lymphocyte ratio (NLR) for prediction of distant metastasis-free survival (DMFS) in early breast cancer: a propensity score-matched analysis

“…The overall χ 2 balance test was not significant ( p=1.0000), and the λ1 measure was larger in the unmatched (0.540) than in the matched sample (0.478), indicating improved overall balance with matching. 26 In addition, before matching, the propensity score was 0.53±0.09 and 0.56±0.08, in the low and high NLR group ( p=0.002), respectively. After matching, the difference between the two groups was not significant (0.55±0.08 and 0.55±0.09, respectively, p=1.000).…”

Testing the Developmental Origins of Health and Disease Hypothesis for Psychopathology Using Family‐Based, Quasi‐Experimental Designs

“…There are certainly other quasi-experimental approaches (Rutter et al, 2001), such as Mendelian randomization (Wiebe, Fang, Johnson, James, & Espy, 2014) and natural experiments (e.g., across pregnancy; Class et al, 2014), as well as advances in the use of statistical covariates (e.g., propensity scores; West et al, 2014), that can help provide insight into causal mechanisms through which early risk factors influence later psychopathology because the approaches have different assumptions and limitations than those discussed here. We stress that the existing research using quasi-experimental designs as a whole clearly illustrates that such approaches are absolutely critical for studying early risk factors (Academy of Medical Sciences Working Group, 2007).…”

The Association of Low Parental Monitoring With Early Substance Use in European American and African American Adolescent Girls

“…Next, consistent with earlier work (e.g., Waldron et al, 2014Waldron et al, , 2015, PSA was conducted to compare adolescents who reported low parental monitoring (lowest 25% within racial/ethnic group) and those who reported higher parental monitoring (upper 75%) based on their predicted probability of experiencing low parental monitoring. PSA is a statistical technique that can be used to reduce bias from confounding variables by matching groups on a range of highly correlated risk factors presumed to predate exposure (Green & Stuart, 2014;Rosenbaum, 2010;Rosenbaum & Rubin, 1983;West et al, 2014), in this case, to low parental monitoring. The predicted probability of low parental monitoring was estimated using logistic regression in STATA and categorized into quintiles.…”

“…The strategy implemented in the current study, propensity score analysis (PSA; Green & Stuart, 2014;Rosenbaum, 2010;Rosenbaum & Rubin, 1983;West et al, 2014), achieves this aim by creating subsamples matched on a liability indicator (i.e., propensity) to initiate alcohol, tobacco, or cannabis at an early age, using as predictors familial and neighborhood factors (i.e., background characteristics) associated with low parental monitoring. PSA was developed to allow tests of causal hypotheses without dependence on strong linear statistical model assumptions, for situations where randomized experimentation was not possible.…”