Crossing controls to treatment in repeated-measures trials

For repeated-measures trials with continuous outcomes in which one group is designated as control and the other treatment, it is sometimes advantageous to add the treatment to the control group in the last periods of the design rather than to continue with control conditions throughout the repeated measures. Many circumstances preclude crossing patients who were initially administered a test treatment back to control conditions so that a full crossover design is not possible. Maintaining the treatment in the treatment group while crossing controls over to treatment in the last periods can result in greater efficiency in estimating a treatment effect that is constant over measures than continuing with control conditions. In contrast, crossing controls is less efficient for estimating a slope parameter for the treatment effect. Equations are provided for both treatment-effect models to determine loss or gain in efficiency from crossing controls. The information from the measures in which controls are crossed to treatment decreases variances of other estimators of interest, such as subgroup-by-treatment interaction effects, stepped or quadratic changes in treatment effect, and interpatient random treatment-effect variance and covariances. Exploration of various models for changes in treatment effect over measures showed that the efficiency of crossing controls depended on the model. This sensitivity to model means that the practical and statistical advantages of crossing controls over to treatment only sometimes outweigh those of the traditional continued-control design.