Cluster randomized trial data capture by clinic, school or community

In a cluster randomized trial the unit of randomization is a group (a clinic, a school, a ward, a village) rather than an individual, and outcomes are usually measured on the individuals within. Researchers choose this design when the intervention is delivered at the group level, when contamination between individuals in the same setting would be a problem, or when the intervention is a service change. Because people within a cluster tend to be more alike than people in different clusters, the effective sample size is smaller than the head count. The usual way of expressing this is the intracluster correlation coefficient and the design effect, which is commonly given as 1 plus (average cluster size minus 1) times the ICC.

The CONSORT extension for cluster trials reflects this. It asks authors to say what the unit of randomization was, whether allocation concealment applied at the cluster or individual level, the cluster sizes and an ICC for each primary outcome, and whether results are reported at the individual or cluster level. Several of those items depend on data being captured consistently: the cluster identifier on every record, the number of eligible and enrolled people per cluster, and which participants were recruited before and which after the cluster knew its allocation. Capturing this when it happens is much easier than reconstructing it for a flow diagram.

What Capture contributes to this design is the multi-site machinery. Participants are enrolled under a site, via QR-code enrollment (site-specific or study-specific, with no app install) or by a site coordinator in the separate portal, and receive site-level numbering such as 001-0042. Data can be filtered by site and exported by site. If your clusters are clinics, schools or villages, a cluster can be modelled as a site. Capture does not offer cluster-level allocation or ICC estimation tools; allocate clusters according to your statistician’s plan, record the arm for each cluster in the study documents or on a form, and run cluster-aware analysis in your statistics software on the export.

The first practical question is how many clusters you need and how many participants per cluster. Adding participants within a cluster gives diminishing returns when the ICC is not tiny, so adding clusters is usually more valuable than enlarging existing ones. Published ICCs for your outcome in a similar setting are a better input than guesses, though they are often imprecise and few trials report them, so plan sensitivity analyses around the value you choose.

The tool below is the live cluster sample size calculator. Enter an effect size, power, the average cluster size and an ICC and compare the result with an individually randomized design. It is a first estimate to bring to your statistician, not a replacement for one. The dedicated cluster randomized sample size calculator page has more explanation, and the dropout-adjusted sample size calculator helps if you expect loss within clusters.

Consent in cluster trials is its own question: sometimes participants consent to data collection while the intervention is delivered at the cluster level, sometimes a gatekeeper agrees on behalf of the group first. Follow your ethics committee’s ruling and record whatever applies. eConsent supports on-screen signatures verified through an email one-time code, with staff countersignature and optional witness or legally authorized representative signatures; see eConsent for multi-site trials for the multi-site pattern. If a cluster consents at the group level, keep that decision as a documented field or in the study documents.

Recruitment bias is a known weakness when people are recruited after clusters know their arm. Capture the sequence: the date the cluster was allocated, the date each participant was approached, enrolled and measured, so that an analysis can compare early and late recruits. Use the screening forms to record how each participant was identified, and edit checks to flag enrolment dates before the allocation date where your protocol forbids them.

Pragmatic cluster designs often run inside routine services. See pragmatic clinical trial software and the pragmatic trial glossary entry for the design vocabulary, EDC for independent trial sites for a clinic-led model and multi-site clinical trial management for operations. For export to your statistician, export EDC data to R, SAS and SPSS is the practical guide.