Skip to main navigationSkip to main content
The University of Southampton
Southampton Statistical Sciences Research Institute

Covariate adjustment and prediction of mean response in randomised trials Seminar

14:00 - 15:00
26 October 2017
University of Southampton Building 54, Room 8033 (8B) Highfield Campus Southampton SO17 1BJ

For more information regarding this seminar, please email Dr Helen Ogden at .

Event details


A key quantity which is almost always reported from a randomised trial is the mean outcome in each treatment group. When baseline covariates are collected, these can be used to adjust these means to account for imbalance in the baseline covariates between groups, thereby resulting in a more precise estimate. Qu and Luo (DOI: 10.1002/pst.165​8) recently described an approach for estimating baseline adjusted treatment group means which, when the outcome model is non-linear (e.g. logistic regression), is more appropriate than the conventional approach which predicts the mean outcome for each treatment group, setting the baseline covariates to their mean values. In this talk I will first emphasize that when the outcome model is non-linear, the aforementioned `conventional’ approach estimates a different quantity than the unadjusted group means and `Qu and Luo’ estimator. I will then describe how for many commonly used outcome model types, the Qu and Luo estimates are unbiased even when the outcome model is misspecified. Qu and Luo described how standard errors and confidence intervals can be calculated for these estimates, but treated the baseline covariates as fixed constants. When, as is usually the case in trials, the baseline covariates of patients would not be fixed in repeated sampling, I show that these standard errors are too small. I will describe a simple modification to their approach which provides valid standard errors and confidence intervals. I will also describe an estimator of the marginal means which exploits baseline covariate yet remains always consistent under misspecificatio​n of a working model for the outcome. I will discuss the impact of stratified randomisation and missing outcomes on the preceding results, and give suggestions for when baseline adjusted means may or may not be preferred to unadjusted means. The analytical results will be illustrated through simulations and application to a recently conducted trial with recurrent events analysed using negative binomial regression.

Speaker information

Jonathan Bartlett, AstraZeneca. Statistical Science Director in The Statistical Innovation Group

We use cookies to ensure that we give you the best experience on our website. If you continue without changing your settings, we will assume that you are happy to receive cookies on the University of Southampton website.