Propensity weighting: how to minimise comparative bias in non-randomised studies?

Abstract

Non-randomised study designs are frequently used by researchers in cardiovascular nursing and allied professions. Baseline differences between the groups to be compared may introduce bias in the results. Methods for causal inference address this issue. One such method is propensity weighting, in which two or more treatments/exposure groups are weighted to make the groups as comparable as possible. As such, it mimics a randomised controlled trial design. In this article, the Twang package is presented for propensity weighting, and its use is exemplified in a study on smoking and cannabis consumption in adults with congenital heart disease.