Clinical Trials

Clinical Trials: A Practical Guide ■❚❙❘Avoiding imbalances and planning an adjusted analysisat the design stageIf a baseline variable has little or no impact on the primary outcome variable, thenany imbalances between treatments are usually unimportant. On the other hand,if a baseline variable is strongly associated with the primary outcome, then even amodest level of imbalance can have an important influence on the treatmentcomparison. By identifying the possible baseline variables that might have asubstantial effect on the primary endpoint, an ‘adjustment’ can be incorporatedinto the trial analysis, regardless of whether there are serious imbalances or not.Two types of adjustment are often used at the design stage. The first strategy is toperform a stratified randomization, to ensure a reasonable balance acrosstreatment groups in a limited number of baseline factors known to be strongpredictors. This method of adjustment can be extremely useful when a singlebaseline predictor has a small number of groups (such as age groups) or a smallnumber of prognostic factors. However, if there are several influential predictors,the number of strata needed will be large and this can lead to over-stratification(see Chapter 7).The second strategy is to prespecify in the protocol which baseline covariates willbe adjusted for and then present the results from the adjusted analysis. In manycases it will be possible to identify important prognostic variables before the startof the trial, by, for example, looking at previous studies. This strategy has theadvantage of overcoming the problem of subjectively selecting predictors in anad hoc manner in the analysis. The US Food and Drug Administration (FDA)and the International Conference on Harmonisation of Technical Requirementsfor Registration of Pharmaceuticals for Human Use (ICH) guidelines for clinicalreports require that the selection of and adjustment for any covariates should bean integral part of the planned analysis, and hence should be set out in theprotocol and explained in the reports [10].ConclusionIn most clinical trials, estimates of treatment effects unadjusted for baselinecovariates are produced and reported. The validity of an unadjusted analysis relieson the assumption that there are no important imbalances involving measuredand unmeasured baseline covariates across treatment groups. When imbalancesoccur on measured predictors of outcome variables, adjusted analyses can beperformed to account for this.293