Researchers with PRIM-ER, an NIH Collaboratory Trial, published 2 innovative statistical techniques for evaluating intervention effects in stepped-wedge, cluster randomized trials. The new models, which use Bayesian methods, outperformed traditional analytic methods and other Bayesian approaches in simulations and real-world applications.

The article was published online in Statistics in Medicine.

In cluster randomized trials with stepped-wedge designs, the clusters are randomized into several groups, and all groups start the trial in the control condition. Groups of clusters cross over to the intervention condition on a staggered timeline, and all groups receive the intervention before the end of the trial.

Stepped-wedge designs can be advantageous when simultaneous rollout of the intervention to all clusters is infeasible, or when withholding the intervention from any cluster would be unethical, or when there is a risk of contamination between intervention subjects and control subjects. However, stepped-wedge designs can also introduce confounding by time, as the intervention is rolled out to clusters in waves. Temporal trends during the study can influence the study’s outcomes.

(Learn more about stepped-wedge designs in the Living Textbook.)

The PRIM-ER researchers tested 2 new Bayesian hierarchical penalized spline models to improve the estimation of intervention effects in stepped-wedge trials. The first model focuses on immediate intervention effects and accounts for large numbers of clusters and time periods. The second model extends the first by accounting for time-varying intervention effects. The researchers applied both models to data from PRIM-ER.

Read the full report.

PRIM-ER tested a multidisciplinary primary palliative care intervention in a diverse mix of emergency departments in the United States to improve the delivery of goal-directed emergency care of older adults. The study was supported by the National Institute on Aging. Learn more about PRIM-ER.