Statistics in Brief: The Cluster Randomized Controlled Trial—What Is It and Why Is It Relevant to Research in Surgery?

Background

Research studies in orthopaedics help guide best-care practices for patients. Randomized controlled trials (RCTs), once considered the gold standard for research in orthopaedics, have evolved to address real-world challenges with their implementation and interpretation. A newer research study design is the cluster RCT. Cluster RCTs have key differences compared with traditional RCTs in terms of analysis and application of their findings.

Questions

What Is a Cluster Randomized Control Trial?

The cluster RCT is an alternative trial design in which the study participants are randomized as entire groups or “clusters” rather than as individuals, as is the case with conventional RCTs. Cluster RCTs can be used to measure the community effect of an intervention, to review survival analysis of a population-based treatment, or to evaluate an educational intervention. The number of clusters required for the study is determined by sample size calculations designed specifically for cluster RCTs. By changing the unit or level of randomization from an individual to a group, the statistical analysis must subsequently be adapted.

In a cluster RCT, the researchers may assign clusters of participants to treatments in either parallel or stepped-wedge designs. In a parallel cluster RCT, each cluster is designated via randomization to treatment with either the studied intervention or the control, which is very similar to the allocation of subjects in an individual RCT. The designation of each cluster remains the same throughout the study period, unlike a stepped-wedge RCT design. In a stepped-wedge design, each cluster is randomized at a predetermined point throughout the study period to either the intervention or to the control group. As the trial progresses, more clusters become exposed to the intervention (Fig. 1). Within the stepped-wedge design, each cluster assigned to the control or intervention may change or “crossover” during the study period. These crossovers may occur at different times for each cluster during the trial.

Fig. 1.

A-B In a cluster randomized control trial (RCT), study participants are randomized as entire groups or clusters instead of individuals. There are two types of cluster RCT designs: parallel and stepped-wedge. (A) In a parallel cluster RCT, each cluster is randomly assigned to the control group (light gray boxes) or the intervention (dark gray boxes); the cluster remains the same for the duration of the study. (B) In a stepped-wedge cluster RCT, again, each cluster is randomized to either the intervention (light gray boxes) or the control group (dark gray boxes); however, these clusters randomly change or “crossover” throughout the trial until all clusters have been exposed to the intervention.

There are several advantages to a stepped-wedge cluster RCT. A stepped-wedge trial design requires fewer clusters than a parallel trial because each cluster acts as its own control group. Stepped-wedge cluster RCTs allow for a graduated delivery of the intervention and may be the superior trial design in scenarios where the intervention is costly, in low supply, requires specific instrumentation, or cannot be implemented at all participating sites because of logistical issues.

When Should Cluster Randomized Controlled Trials be Considered an Option in Surgical Research Studies?

There are several circumstances in which cluster randomized trials are the superior design in surgical research. A cluster RCT offers an alternative to addressing issues with traditional individual RCTs, such as patient recruitment, provider (surgeon) bias, and contamination between study groups (Table 1). A cluster RCT is particularly useful when the outcome of a randomized trial aims to measure the effect of an intervention in an entire community. Standard RCTs require randomization before determining treatment allocation for each patient; this makes study enrollment, consent, and clinical flow burdensome. Because cluster RCTs randomize patients in naturally occurring groups (like hospital wards, community clinics, or regional centers of care), study flow is made easier. Given that each cluster’s treatment allocation is determined at the start of the study, and that the cluster remains the same as the study progresses, clinical care is further simplified even as research is being conducted. For example, a cluster RCT can randomize one hospital’s patients to the intervention group, while another hospital’s patients serve as the control group. Defining each cluster based on naturally occurring groups allows for flexibility and more normal clinical care. Group or cluster allocation can ease participant recruitment, observation, and follow-up.

Table 1.

Comparison of randomized control trial and cluster randomized control trials

Randomized control trial	Cluster randomized control trial
• Individuals are randomized to control or treatment groups in a study • Sample size required to ensure appropriate statistical power to observe effect of intervention • Limits bias by randomization • Can be challenging to implement at a single center with patient crossover • Follows CONSORT guidelines	• Patients are grouped together and randomized as “clusters” to control or treatment groups in a study • Larger sample size required than RCT for appropriate study power • Limits bias by randomization • Real-world effectiveness of conclusions in large populations or groups • Requires complex planning and statistical analysis and adjustment • Follows CONSORT guidelines

In addition, in a standard surgical RCT, it can be difficult to separate the effect of a surgical technique being studied from the skill of the providers administering it. Surgical techniques vary among surgeons depending on training, experience, and practice location. One review of randomized trials suggested that this problem is more severe in highly specialized fields, such as orthopaedic surgery [4]. Cluster RCTs can take advantage of the natural abilities and preferences of surgeons or groups of surgeons, allowing them to deliver care using the approaches they are most experienced and comfortable with. Surgeons practicing within a geographical area who have been performing a specialized procedure with similar training and experience can be clustered according to their patients’ roster in a cluster RCT. Defining a cluster based on factors that influence the technique and skill of a surgeon is a practical application of this study design in surgical trials.

RCTs are designed to evaluate the application of an intervention on an individual level in a controlled setting; however, some interventions by nature are designed to be applied to populations [3]. Observing the effect of an intervention on a group or population as a whole is a notable benefit of cluster RCT design. In cluster RCTs, differences between the intervention and control groups are measured at the individual level; however, the comparison of the control and study intervention is at the cluster level. Evaluating the effectiveness of regional health promotion programs such as fall prevention strategies in an urban region or intimate partner violence education programs to healthcare workers at tertiary trauma centers can be addressed with this design type. A cluster RCT design permits flexibility to define the clusters or group within a study to permit observations of the intervention on a specified population.

Cluster designation within a cluster RCT can also be designed to observe geographical regions with socioeconomic similarities. Because groups of individuals who share socioeconomic environments may also have similar health challenges, standard surgical trials done in regions with limited socioeconomic means (including developing countries) present unique challenges in their generalizability of study findings. Groups of individuals residing in a rural center compared with an urban center may have different barriers to resources for fracture care. Villages in rural India may have different economic limitations for trauma care compared with villages in rural Pakistan. Specialized healthcare interventions within a particular population can be observed with a cluster RCT design that evaluates real-world settings within populations and generate study findings that can be generalized to a wider group of individuals.

Finally, cluster RCTs may help mitigate a problem of conventional RCTs known as contamination. Contamination occurs when the observed responses of a study treatment are affected by other study participants in the research trial [4]. Contamination may occur if patients in a surgical RCT compare their experiences while meeting in a physician’s office or as inpatients in a hospital ward. For example, imagine a study of accelerated physiotherapy after an ankle ligamentous injury. Two patients undergo operative care, with one patient following a standard physiotherapy program and the other doing an accelerated physiotherapy program. If these two individuals meet and decide they will both then use some of the exercises in the accelerated physiotherapy program, this would be a form of contamination. By having clusters in a study separated by a naturally occurring geographical area (such as hospitals or regions), the risk of interaction between study participants is decreased, therefore helping to preserve the effects of the treatments being studied.

Do Cluster RCTs Have Important Disadvantages?

Like all study designs, they do. First, they are methodologically demanding, and they can be expensive; cluster RCTs are not a good choice if a more straightforward and less resource-intensive design, such as a conventional RCT, will answer the research question. Cluster RCTs are usually larger than individually randomized control trials and consequently, may require increased financial resources and grant funding. Evaluating an intervention with multiple clusters may be costly, and a cluster RCT design may not be as cost efficient as a conventional RCT.

Cluster RCTs should not be used in situations where randomizing individuals is feasible and can answer the desired research question. Choosing a cluster RCT as a study design is best suited to address research questions involving preexisting groups, such as a classrooms, villages, or regions. Randomizing individuals in preexisting groups to a study intervention directed at observing differences at a group level may not be practical to address a research question. For example, a research question designed to observe the effects of a drug treatment on individual patients is best addressed with a conventional RCT design with individual randomization. A research question to observe the benefits of a drug safety education program within hospitals in a large region would be suited with group randomization, with each hospital defined as a cluster for randomization in a cluster RCT study.

Another disadvantage arises from how consent is obtained in cluster RCTs. In a conventional RCT, consent is obtained from each individual participant in the study. The process for obtaining informed consent in a cluster RCT is complex as the study participants are groups of individuals. Furthermore, when to obtain informed consent of the individuals in a cluster can be problematic. Clusters may be randomized to the control or treatment group before members of a cluster can be approached for consent. Obtaining informed consent from participants in a cluster randomized trial should occur before data collection or the study treatment begins. Involving key stakeholders and institutional decision makers in mitigating concerns of informed consent in the research ethics board approval process is paramount in a cluster RCT study. Several ethical guidelines, including the Ottawa Statement on the Ethical Design and Conduct of Cluster Randomized Control Trials [8], provide recommendations for the process of obtaining informed consent in this trial design.

Although cluster RCTs require complex planning and special methodological expertise, they are worth the trouble. When applied to the correct clinical problem, the cluster RCT design can be implemented to address many of the challenges present in conventional surgical RCT design (Table 2).

Table 2.

Challenges of RCT in surgery and benefits of cluster RCT design

Challenges of RCT in surgery

Cluster RCT solution/benefit

• Individual randomization is not always feasible for all surgical trials • Surgeon expertise or preference of treatment may vary • Interaction between participants in different study groups may cause contamination • Application of research findings to population as a whole

• Randomize group of study participants within a naturally occurring clinical setting–hospital, ward, regional health center • Cluster allocation, according to physician practice or hospital-based setting, can ease recruitment and physician standardized treatment to all patients • Limited patient interaction with natural geographical boundaries • Observing differences in naturally occurring groups can allow for ease of application of study findings in real-world setting to large populations

Myths and Misperceptions

Myth 1: There Is No Contamination Risk in a Cluster RCT Because the Groups Will Not Interact with Each Other

In fact, cluster RCTs still may experience contamination, although the hope is that it will be a smaller problem than it is in conventional randomized trials in which patients are randomized to an intervention performed by a group of surgeons at a particular care center. Cluster segregation tactics, such as geographic distancing, reduce the risk of study participants interacting with each other. Although the interactions of study participants may be decreased in a cluster RCT design due to natural physical or geographical boundaries, the risk of interaction cannot be eliminated completely.

Myth 2: Statistical Analysis and Sample Size Calculations in Cluster RCTs Are Similar to Those in Standard RCTs

Researchers who conduct cluster RCTs must explicitly account for clustering at every stage of the design and analysis. A cluster RCT design requires additional attention to the sample size calculation, data analysis, and loss to follow-up. It is imperative for researchers adapting a cluster RCT design to ensure that a correct and appropriate sample size calculation is performed before starting the study. In accordance with the current Consolidated Standards of Reporting Trials (CONSORT) guidelines for cluster RCTs [2], the reasoning for determining a cluster must be stated in the study protocol, as well as in the supporting documentation. Without employing specific statistical methods to account for variability in each cluster, the differences between each group will be underestimated, which may invalidate the study results. The differences of observations within a cluster and between clusters requires specific statistical tactics to account for these variances. One approach for handling this is the intracluster correlation coefficient, which reflects the fact that observations about individuals within the same cluster may be more similar when compared with the observations about individuals in a different cluster [1].

Determining the sample size within a cluster RCT is also different compared with a conventional RCT. In a cluster RCT, a sample size calculation must consider the differences between clusters. To account for these differences, the sample size calculation of a cluster RCT must incorporate the design effect. The design effect is an adjustment that indicates how much the sample size of a cluster RCT must be increased to obtain the same study power as a conventional RCT or individually randomized trial. The design effect of a cluster RCT is correlated to the intracluster coefficient. The larger the intracluster correlation coefficient, the larger the design effect, resulting in a larger sample size required for a cluster RCT design.

Although some participants may be lost to follow-up in conventional RCTs, the loss of a cluster can more severely impact a study’s conclusion because it represents not the loss of a patient, but rather the loss of an entire treatment group. This can occur, for example, if a surgeon’s office practice closes or if a trauma hospital in a developing country is destroyed by flooding. The statistical allocation for a cluster RCT can be challenging and requires appropriate resource planning, including a statistician as an essential part of the study team.

Misperception: Cluster RCTs Have Too Many Challenges to Properly Execute in Orthopaedic Research

Cluster RCTs have been done in orthopaedic research, and some of these studies have been influential [3, 5, 6, 7]. This trial design has continued to see wider use with several large trials in injury prevention, arthroplasty, use of opiate analgesia, and trauma-based care.

A recent large cluster RCT in England evaluated healthcare education as a means to prevent falls and associated fractures in the community setting [5]. Clusters were established according to physician practices and randomized to two different treatment groups for fall prevention: advice received via mail and a multifactorial advice program consisting of in-person assessments, caregiver involvement, motivational interviewing, and telephone follow-up. The study found that neither multifactorial nor mail advice prevented falls or subsequent fractures [5]. Because the researchers used a cluster RCT design that took advantage of preexisting natural groups (practice settings), contamination in this study was limited and recruitment less challenging.

Another cluster RCT examined the effectiveness of blood salvage in primary THA and TKA [7]. Twenty-one participating hospitals were stratified by geographic location to limit study contamination of clusters defined by regional preferences. The investigators identified concerns with interaction and contamination risks of the study participants, and therefore chose a cluster RCT design to decrease these risks. To further decrease these risks, the cluster control hospitals were not contacted during the intervention period, and a single individual at each identified site was available for study communication. This study observed the outcomes of a therapeutic intervention in separate surgical environments in this geographical region with a limited risk of study contamination. Blood salvage and erythropoietin use did not decrease, despite the utilization of an educational program to discourage their use in this study [7].

Another cluster randomized trial of 235 participants examined three modalities of opiate-sparing analgesia after elective hip surgery. Recruitment in this study was simple: Each patient under each participating surgeon’s practice was allocated to one of three study designed clusters [3]. Each surgeon acted as a “guardian” and allowed each of the low-risk study interventions to be implemented as standard of care for the allotted period [3]. The study concluded that multimodal analgesia (antiinflammatories and anticonvulsants) decreases the amount of opiate use after elective hip surgery.

Another set of cluster RCTs are underway, known as the PREP-IT trials (Program of Randomized trials to Evaluate Pre-operative antiseptic skin solutions in Orthopedic Trauma [6]). These pragmatic cluster RCTs will compare the effects of two solutions on surgical site infections and unplanned fracture-related reoperations in trauma patients [6]. These studies chose a center-cluster randomized design for feasibility; each participating hospital will use one intervention.

Conclusion

Both standard RCTs and cluster RCTs have advantages and disadvantages, but there are some instances in orthopaedic research where the advantages of the cluster RCT are compelling. In particular, cluster RCTs allow surgeons to complete procedures with which they are most familiar (and most comfortable recommending to their patients), they facilitate study flow because they do not require randomization and blinding on a patient-by-patient basis, and they may minimize the study contamination. However, to perform them correctly, cluster RCTs require special statistical and methodological expertise, and it is recommended that an experienced statistician be on the research team. The findings of cluster RCTs may be especially applicable at the population level since they often are performed at the population level.