The lay abstract featured today (for the Tutorial in Biostatistics So Many Choices: A Guide to Selecting Among Methods to Adjust for Observed Confounders by Luke Keele and Richard Grieve) is from Statistics in Medicine with the full Open Access article now available to read here.

How to cite

Keele, L. and Grieve, R. (2025), So Many Choices: A Guide to Selecting Among Methods to Adjust for Observed Confounders. Statistics in Medicine, 44: e10336. https://doi.org/10.1002/sim.10336

Lay Abstract

In a randomised controlled trial (RCT) participants are randomly allocated to  alternative treatment strategies. This randomisation can ensure that similar groups are compared. In many situations it is not feasible or ethical to randomise patients to different treatments. Instead ‘non-randomised studies’ can generate evidence on comparative effectiveness. These studies do not randomise patients, but instead they compare outcomes across groups who had different treatments in routine practice.

A major challenge that often arises in these non-randomised studies is that there are systematic differences between the comparison groups in the characteristics of the patients and setting. Unless the study addresses these differences between the comparison groups, the  assessment of the comparative effectiveness of the treatments is likely to provide biased results.

Over the last 20-30 years many statistical approaches have been developed that attempt to reduce the risk of non-randomised studies providing biased evidence about treatment effectiveness, and there are now many statistical approaches to choose from. However, there is little guidance for analysts about how to choose the best statistical approach to reduce the risk of bias when assessing comparative effectiveness in these studies.

The aim of this paper is to outline alternative statistical methods for reducing a major risk of bias when using non-randomised study to assess comparative effectiveness. The paper outlines the key assumptions that are behind the main groups of methods. The authors provide a framework to help analysts choose from amongst the alternative methods. The paper use the example of an evaluation of a monitoring device in intensive care to illustrate the application of alternative methods in practice. The paper concludes with recommendations about how to select statistical models to help future non-randomised studies provide more accurate comparative effectiveness evidence.