Disclaimer – I’m “computationally challenged” so I use words.
It seems like the propensity score has been increasingly used in prominent research these days. Propensity scores are a way to account for confounding in observational studies so we can have more confidence that an outcome is truly associated with an exposure. If that’s all you care to know, just skip to the bottom. For an example, read on.
The randomized controlled trial (RCT) is the king of study designs, because it’s the best way to account for confounding and prove causation. But many studies simply can’t use a RCT methodology for various reasons, e.g. you can’t ethically randomize someone to the “smoking group” to look at the effects of smoking. Observational data can show association not causation, but the association may not be true if a confounding variable is contributing to the outcome. Say you want to measure whether cricoid pressure makes laryngoscopic view of the glottis worse. The problem is, other variables may also make the glottic view worse, like obesity. So obesity may be a confounder. A propensity score would look at the independent effect of obesity on predicting a difficult glottic view and use statistical techniques to score that variable. Then the researcher would match patients with the same propensity score who got the treatment (cricoid pressure) and who didn’t to see if there was still an association with the outcome (worsened glottic view). It’s a way of making observational data a little more like RCT data when a RCT is not possible or feasible.
Propensity scores are a way to account for confounding in observational studies so we can have more confidence that an outcome is truly associated with an exposure. If you want to read more, this is a good place to start. Now my brain hurts.
Emerg Med J. 2017 Mar;34(3):129-131. doi: 10.1136/emermed-2016-206542. Epub 2017 Jan 31.
1 Design, Trials and Statistics, ScHARR, University of Sheffield, Sheffield, UK email@example.com.
Propensity score methodology is being increasingly used to try and make inferences about treatments when randomised trials are either impossible or not conducted and the only data are from observational studies. This paper reviews the basis of propensity scores and the current state of knowledge about them. It uses and critiques a current paper in the Emergency Medicine Journal to illustrate the methodology.
© Article author(s) (or their employer(s) unless otherwise stated in the text of the article) 2017. All rights reserved. No commercial use is permitted unless otherwise expressly granted.