Description:
When comparing two means in a cluster-randomized design, the samples for each group consist of a number of clusters for each group. For example, patients at a hospital would be a cluster, with a number of hospitals assigned to each of two treatments.
The Tests for Two Means in a Cluster-Randomized Design procedure in PASS is used to determine the needed number of clusters, the needed cluster size, or the power associated with a stated number of clusters and cluster sizes.
In this video, we will focus on solving for the number of clusters needed to properly power the study.
The test will be a two-sided T-test, which is based on the number of subjects, as described in recent publications.
The power is set to 0.90 and alpha is set to 0.05.
We will enter a variety of cluster sizes, to determine the effect of cluster-size on the needed number of clusters.
We would like to have the number of clusters and the cluster sizes be the same in both groups.
We will also examine a variety of values for the coefficient of variation.
We would like to base the calculations on a true mean difference of 16.2 and standard deviations of 20, 30, 40, and 50.
Based on other similar studies, we use a value of 0.027 for the intra-cluster correlation.
The Calculate button is pressed to generate the estimated number of clusters needed.
Each line represents a different combination of parameters. There are 80 lines, since we chose 5 cluster sizes, 4 coefficient of variation values, and 4 standard deviations.
The first line shows that 8 clusters are needed in each group to achieve 90% power for the case of a cluster size of 5, a coefficient of variation of 0.4, a standard deviation of 20, and so forth.
If we examine the column of powers produced, we notice that many are considerably higher than 0.9. This is due to the number of individuals that are added when adding a single cluster to each group to achieve the desired power. Since the power jumps with each cluster, the resulting values are often well above the specified power.
A variety of plots are given at the end of the report. Each plot shows the needed number of clusters for various combinations of varied parameters.
The final plot shows number of clusters needed for all 80 parameter combinations.
