Most triallists who present such tests, about half in the study by Pocock et al. [6], do not do so because they suspect that their own trial has been subverted. They do it because they wish to reassure themselves and their readers that chance has not led to an imbalance for which they need to adjust, or to identify variables for which adjustment should be made. But this comparison is irrelevant to the decision to adjust for baseline variables. If a baseline variable is prognostic for the outcome, researchers should adjust for it whether the groups differ or not, as adjustment will improve the estimate of the treatment effect. If a baseline variable is not prognostic for the outcome, researchers do not need to adjust for it even when the groups differ, as adjustment will not improve the estimate of the treatment effect. The triallist should adjust for known prognostic variables and should be able to plan the adjustment before a patient has been recruited.
