Evidence Synthesis of Pragmatic Clinical Trial Methodology

This project conducted reviews of statistical methods that has been developed to address two aspects of pragmatic trials: (1) accounting for unequal numbers of participants in clusters in a cluster-randomized trial, and (2) combining trial data with information from outside the trial to obtain more precise answers. These reviews will help trial designers more easily find the information needed to design their trials as well as identifying when new methods need to be developed. One new method was developed during this project.

In a cluster-randomized trial, participants are assigned to receive a treatment in groups, instead of individually. For example, if the trial is about testing a new way of providing care in a hospital, then all the participants (patients) within one hospital will receive the same type of care while the type of care (usual care vs new way of care) will vary across different hospitals. For this type of design, the calculations for how many participants are needed and the correct way to analyze the data is complicated. We conducted a review of literature on methods for doing these calculations for different types of cluster-randomized trials.

Real-world trials often involve comparisons of interventions to routine care or to interventions that have already been tested previously. This means that often there is knowledge about how well the interventions being compared work even before the trial is conducted. We conducted a review of literature on Bayesian methods for combining existing knowledge with trial data to get more precise answers.

The new method developed in this project showed how one can increase the precision of the treatment effect from a stepped-wedge cluster-randomized trial by taking into account outside information on the changes in outcome rate over time using Bayesian methods.