Optimal design of cluster randomised trials with continuous recruitment and prospective baseline period

Why it matters

Trials are one of the most robust ways of measuring the real effects of a treatment or service.  But in complex improvement work in health services, some of the traditional designs may need tweaking to capture real-world changes and effects.  A cluster randomised trial involves participants from the same group, for example patients from a specific geographical region or who attend the same general practice. These participants are randomised to receive the same healthcare intervention.

There may be advantages to assessing the outcomes of cluster randomised trials from a baseline period of data collection before the cluster is randomised, in order to control for cluster differences and give more precise results.

Approach

In this study we looked at trials where clusters were recruited for future study, or participants were identified as part of a continuous process over a given calendar period.

We asked whether researchers should collect baseline data as part of the trial, from a period where participants from all clusters receive routine care, followed by a more conventional trial scheme in which half of the clusters cross over to intervention and a new series of participants is recruited from each cluster. We also investigated the optimal length of time for a baseline period.

What we found

From studying this type of trial, we found that in some circumstances it is best not to include a baseline at all, while others benefit from an optimal duration for the baseline period. We discovered that researchers could achieve close to precision by choosing either to have no baseline period at all, or a baseline period which takes up half of the trial period.

The circumstances where it is preferable not to include a baseline period are studies with a smaller recruitment rate, groups are less like each other or become less like each other or longer time is needed for assessment as service changes.  Where there is a transition period between recruiting or identifying participants to receive an intervention, the benefit of having data available from the control group during the transition period may only be modest.

It would be interesting to extend these investigations to cluster randomised trial designs, including stepped wedge designs. This might make it easier for future researchers to evaluate complex service changes as they happen.