Although PCTs do not necessarily require a particular statistical design approach, both the kinds of questions PCTs are designed to answer and the settings in which they take place may tend to favor certain approaches, such as cluster randomization. For example, the nature of the interventions they seek to test, which may involve healthcare delivery changes, might be better implemented through randomization at the practice, clinic, or even hospital level.
In the following sections, we will examine key considerations in statistical study design and analysis for PCTs.
Unit of Randomization
A key design factor for PCTs is the choice of unit of randomization. For “traditional” RCTs, the unit of randomization is generally the individual trial participant or patient, and each individual is randomized to receive an intervention versus a comparator therapy or a placebo. For cluster-randomized trials (CRTs), the unit of randomization can take place at the level of the physician, practice, hospital, health system, city block, or other unit that comprises multiple patients/participants.
Case Example: Unit of Randomization vs Unit of Measurement
The NIH Collaboratory’s “Time to Reduce Mortality in End-Stage Renal Disease” (TiME) study provides an example of the difference between the unit of randomization and the unit of measure typical of cluster-randomized trial.
In the TiME trial, participating dialysis clinics providing care to patients with end-stage renal disease were randomly assigned to provide one of two interventions: an “extended” period of hemodialysis for a minimum of 4.25 hours, versus standard care. The trial was designed to evaluate whether the extended period of dialysis would be associated with better survival and quality-of-life outcomes. Thus, the unit of randomization for TiME was the dialysis clinic, but the measurements of interest were the outcomes of individual patients.