power for crossover design

[Madu Abuchi]

Hi

I am trying to calculate sample size and power for single-group mixed-methods crossover design.

The assumption is that 40 people were recruited and trialed for treatment A (with pre and post observation) and then treatment B (with pre and post observation)

What is the best approach to calculate power to be able to detect a priori mean difference between the two treatments given that each treatment has pre and post observations?

Is the power pairedmeans command of stat appropraite in this scenario.

Suggestions are welcomes.

Thanks, Madu

[Clyde Schechter]

Assuming that your outcome measure is a continuous variable for which a paired t-test would be a suitable analysis, then, yes -power pairedmeans- could be an appropriate approach to calculating statistical power for this design.

However, in crossover studies, where possible, it is preferred to also randomize the order in which participants experience treatments A and B so that treatment itself is not confounded with the effect of the order in which the treatments are given.

If you do that, however, then the full analysis of the data is not a paired t-test but a within-person regression model in which treatment, order (1st vs 2nd) and treatment#order are the predictors. I don't think any of the built-in power calculations in official Stata will handle this. And while this design is, in most respects, a cluster-randomized study, the hypothesis you want to test here is not the usual hypothesis test for which -clustersampsi- calculates power/sample size. So I don't know of any alternative to simulation here.

[Dave Airey]

There is a repeated measures power tool in Stata 15 that supports one between factor and one within factor. All subjects get both A and B so this is a within subjects factor with levels A and B. By randomization about half are going to have order AB and half order BA, and this is a between subjects factor. If the pre and post are made to be differences, this tool might be worth a try.

[Madu Abuchi]

Clyde and Dave,

Thank you both for your wonderful advice