Sample size calculation for a cluster randomized controlled trail with three arms

[Teketo Tegegne]

Dear All,
How can I calculate sample size determination for a cluster randomized controlled trail which had three arms using STATA?

I have three trial arms:
Control group - will receive a currently used treatment
Intervention group 1 - will receive new treatment 1
Intervention group 2 - will receive new treatment 2
With regard to the nature of clusters: It will be a two stage cluster.
Stage one = districts will be randomly selected
Stage two = One primary health care facility/Health Center per sampled district will will be selected
Thus, all clients, who will fulfill the inclusion criteria of the study, coming for the specified health care service at the sampled health facility/Health Center will be included.
I have two types of outcomes; binary and count.

The sample size I want to determine will take into consideration the following issues;

1. Number of clusters
2. Cluster size (varying cluster size)
3. Coefficient of variation
4. Intracluster correlation coefficient / rho and
5. Effect size

How can I determine the sample size for the three groups; is the Bonferroni or Tukey-Kramer correction appropriate for it or is it possible to use the two population formula and then allocate for the three groups or is there any correction assumption other than this that STATA will consider?

I was looking at STATA but I didn't see option for cluster data sample size calculation and for multiple group too. Which option should I use?

http://www.statalist.org/forums/help#adviceextras

Regards

Teketo

[Clyde Schechter]

I am not aware of any software that will do this for a three armed study, nor with varying cluster sizes.

The user written -clustersampsi- is the closest thing that I know of. You can get it at -net sj 15-2 st0286_2- (Follow the -net install- and -net get- instructions there.)

It may be that there is something more flexible out there that I don't know about, and I, too, would be eager to learn of it if somebody else responds.

[Jeph Herrin]

For complex study designs, I typically use simulation to estimate power. The idea is to generate fake datasets with known effect sizes, simulating the noise, and determine in what proportion the effect is detected (P < 0.05) under a suitable test. The proportion will be the power. This article may get you started http://www.stata-journal.com/sjpdf.h...iclenum=st0010


hth,
Jeph

[Clyde Schechter]

Jeph Herrin . Thanks. Actually, that's what I generally do, too. I was hoping, along with Teketo, that somebody has come up with something easier: something similar to -sampsi-. These simulations can be time consuming to code, and even more so to run, if, as is often the case, you need to do sensitivity analysis around several uncertain parameters.

[Jeph Herrin]

Clyde - There is a formula for calculating sample size for a cluster randomized trial, at least in the two arm case - one adjusts for the variance inflation factor (VIF) = (1+(m-1)*ICC); this works fine unders some assumptions about uniform cluster size. See http://www.jclinepi.com/article/S089...169-2/fulltext

And I've just remembered this as well -findit clustersampsi-

Jeph

[Clyde Schechter]

Yes, -clustersampsi- is quite useful for the simple case of uniform cluster size, as is the VIF formula. Unfortunately, in my line of work, cluster sizes usually vary quite a bit, so these approaches don't do the trick.

I have an intuition that when the cluster sizes vary, one might be able to get the right results, or at least a decent approximation to them, by calculating the harmonic mean of the cluster sizes and using that as if it were a uniform cluster size study. But I've never been able to prove that, nor find it anywhere in the literature.