Confounding adjustment using propensity score methods

This chapter introduces the need for confounding adjustment in comparative effectiveness studies that are based on real-world data. Propensity score matching, stratification, weighting, and regression adjustment are presented as methods to adjust for confounding bias. The fundamental principles and assumptions of propensity score methods are described. This chapter provides recommendations on covariate selection, propensity score estimation, implementation of propensity score matching, stratification, weighting, and regression adjustment, assessment of covariate balance and overlap, and conducting sensitivity analyses. It also provides an overview of methods to combine propensity score matching or weighting with multiple imputation when there are missing data. While the core of this chapter focuses on the comparative effectiveness of two treatments, it ends with a brief discussion on considerations for propensity score analyses with multiple treatments. A case study in multiple sclerosis is used to illustrate the steps, with example R code of good practice provided as supplementary information.