Randomisation covariates

These models can be written in a more precise form by defining a binary indicator to denote treatment. Let z = 0 for patients randomised to the placebo group and let z = 1 for patients randomised to the test treatment. The model with a single covariate and assuming a common slope can then be written ... 

Covariates affected by treatment allocation. Variables measured after randomisation (e.g. compliance, duration of treatment) should not be used as covariates in a model for evaluation of the treatment effect as these may be influenced by the treatment received. A similar issue concerns late baselines , that is covariate measures that are based on data captured after randomisation. The term time-dependent covariate is sometimes used in relation to each of the examples above. 

It is not advisable to adjust the main analyses for covariates measured after randomisation because they may be affected by the treatments. ... 

Remember however that variables used to stratify the randomisation should be included. It is also not usually appropriate to select covariates within ANCOVA models using stepwise (or indeed any other) techniques. The main purpose of the analysis is to compare the treatment groups not to select covariates. 

In Chapter 6 we covered methods for adjusted analyses and analysis of covariance in relation to continuous (ANOVA and ANCOVA) and binary and ordinal data (CMH tests and logistic regression). Similar methods exist for survival data. As with these earlier methods, particularly in relation to binary and ordinal data, there are numerous advantages in accounting for such factors in the analysis. If the randomisation has been stratified, then such factors should be incorporated into the analysis in order to preserve the properties of the resultant p-values. 

---