Cluster-Randomized Trials

Experimental Designs and Randomization Schemes

Section 3

Cluster-Randomized Trials


Elizabeth R. DeLong, PhD

For the NIH Health Care Systems Collaboratory Biostatistics and Study Design Core


Contributing Editor

Jonathan McCall, MS

Cluster randomized trials (CRTs) differ from individually randomized RCTs in that the unit of randomization is something other than the individual participant or patient. CRTs are in common use in areas such as education or public health research; they are particularly well-suited to testing differences in a method or approach to patient care (as opposed to evaluating the physiological effects of a specific intervention).

Why Choose Cluster Randomization?

There are several reasons why CRT designs might be preferred to or more suitable than a traditional RCT. First, a CRT might be preferred when the target of the intervention is a collective or system rather than a particular person, such as a patient.  For instance, while a traditional RCT may be better suited to discriminating whether or not a novel therapy works in patients with a given disease or condition, a CRT is better able to evaluate whether a new standard of care, guideline recommendation, or other practice-wide, hospital-wide, or system-wide change is affecting patient outcomes. Second, a CRT might be preferred when there is a significant potential for contamination in the study. Contamination occurs when aspects of an intervention are adopted by members of the group that was randomized to not receive that intervention (please see also "What Is Contamination, and Why Does it Matter?" immediately below).

There are also compelling practical reasons for randomizing clusters rather than individuals (Cook et al. 2016). For example: in a trial comparing 12-hour nursing shifts to 8-hour shifts, implementing these protocols on a patient-specific level would be nearly impossible. In this case, randomizing wards or floors would be much more practical and would also accommodate the need to avoid contamination.

What Is Contamination, and Why Does it Matter?

The most compelling reason to randomize at the cluster level rather than at the individual level is the potential for contamination, whereby participants within a cluster are likely to be treated similarly and hence exhibit similar outcomes.

When contamination occurs during a clinical trial, it will dilute the observed differences between comparators and can affect the reliability and validity of the study.

Case Examples: Contamination
  • Example 1: Participants who share the same provider in a trial comparing different weight-loss strategies may meet each other in the waiting room and communicate about their respective strategies, or the provider might not be able to adapt to coaching differently depending on the randomization. Some participants in each group might even adopt elements of both strategies, and neither group would demonstrate the impact of its intended strategy. Randomization at the provider level, with each provider coaching only one of the strategies, would reduce the risk of contamination.
  • Example 2: A trial evaluating a campaign designed to reduce nosocomial infections by encouraging better staff handwashing practices might include posters in each of the rooms. Staff generally cover several rooms on a floor and would be exposed to the posters, which would likely change their behavior if the posters were actually effective. Not only would it be infeasible to randomize at the provider or patient level, doing so would minimize the difference between groups due to the contamination. The campaign might then be declared unsuccessful despite actually having had a positive effect. The solution would be to randomize different areas of the hospital (taking care to consider potential confounding as described in the coming sections) with only half of the areas receiving the posters.

Although avoidance of contamination is one of the most important reasons for using CRT designs, pragmatic concerns can dominate the need for cluster randomization when it is practically impossible to randomize at an individual level.



A brief history of the cluster randomized trial design

Provides a historical overview of the development and application of CRTs in research, along with key references

Designing Multi-center Cluster Randomized Trials: An Introductory Toolkit

Provides an overview of considerations for designing and implementing CRTs in health care systems



back to top

Cook AJ, Delong E, Murray DM, Vollmer WM, Heagerty PJ. 2016. Statistical lessons learned for designing cluster randomized pragmatic clinical trials from the NIH Health Care Systems Collaboratory Biostatistics and Design Core. Clin Trials. 13:504-512. doi:10.1177/1740774516646578. PMID: 27179253.


DeLong ER. Experimental Designs and Randomization Schemes: Cluster-Randomized Trials. In: Rethinking Clinical Trials: A Living Textbook of Pragmatic Clinical Trials. Bethesda, MD: NIH Health Care Systems Research Collaboratory. Available at: Updated August 24, 2017.