Prospective individual matching: covariate balance and power in a comparative study

In phase II to phase IV studies, randomization has gained widespread acceptance as a methodologic tool for the allocation of patients to treatment. However, randomization is not always feasible. At times, the treatment intervention occurs universally throughout one or more units (for example, a hospital unit), while the control therapy is the only intervention provided in other units. Patients may arrive randomly at a unit, based solely on availability of the unit to accept new subjects. Thus, the treatment assignment process is out of the investigator's control and not subject to selection bias. We describe a prospective individual matching procedure through which one can achieve balanced allocation of subjects to treatment groups in this comparative study setting. In this paper, we compare balance of baseline covariates and power for this design, in which the subject is selected at random and assigned to a treatment group, and the traditional randomized block design, in which the treatment is chosen at random and assigned to a subject. We show that the prospective individual matching procedure compares favourably to the traditional randomized blocked design with respect to both baseline covariate comparability and statistical power.