Sample size and power calculation- Stepped Wedged Design

[Kennedy Mutai]

Hello Listers,

I am looking for an advice on sample size and power calculation for an intervention study. The study uses a stepped wedged design.

The study aims to evaluate the effectiveness of case managers (intervention) in improving linkage and retention rates to HIV care among newly diagnosed HIV-infected persons. The study has two arms i.e., the control arm and the intervention arm. The setup is a hospital facility with 12 departments. We will randomize 6 departments to begin intervention in the 3^rd month and the remaining 6 to commence intervention in the 6^th month. A previous study done in the facility showed that 38% of the patients are linked to care. We anticipate that using the case-managers (intervention) will increase the linkage rate by approximately 20%. In the past 9 months, 1137 patients tested HIV positive. I have tried to follow Hussey and Hughes paper in the link below but it wasn’t promising. Equally other papers have not yielded fruits for me. http://www.google.co.ke/url?sa=t&rct...76247554,d.c2E

My questions are:-

1. I would like to have a Stata code to calculate a sample size for this study, given the above information.
2. Given any arbitrary (naïve) sample size of say 960, from the information provided above, is there a code in Stata that I can calculate the power of the study?
3. In addition to that, can I get a code to generate the power curves?

Please bear with my long email.
Thanks in advance.

[Clyde Schechter]

There is a package st0341 from the Stata journal that does this. I have not personally used it, so I can't guide you on how to use it, or comment on it.

-net install st0341.pkg-

[Kennedy Mutai]

Thanks Clyde. I will try the package!

[Kennedy Mutai]

The package worked perfectly well.
The code is written as

steppedwedge, binomial power complete(1) p1(0.38) p2(0.58) m(80) k(6) rho(0.01) alpha(0.05) steps(2)

Where:-
power specify a power calculation
binomial specify a comparison of proportions
alpha(#) the significance level of the test; default is alpha(0.05)
rho(#) intracluster correlation coefficient
k(#) number of clusters randomized at each step
m(#) average cluster (cell) size
complete(1) complete design (fully specifiable by k(), m(), and t())
p1(#) proportion for sample 1
p2(#) proportion for sample 2

[Rachel Jewkes]

I have just tried to find this software and didn't succeed. Does anyone know why?

[Stephen Jenkins]

Rachel: welcome to the Forum. Please read the FAQ (hit the black bar at the top of the page) from top to bottom, and note the remarks about how to post in a manner likely to raise the chances of a helpful answer. Reporting that you tried something and that it didn't work provides no useful information at all for people to help. How did you search for the package? What were the results, etc. etc. After your reading, type the following at your Stata prompt:

search steppedwedge, all

But, in any case, Clyde Schechter had already indicated how to get the software (post #2). Did you try that?

[Rachel Jewkes]

Stephen: I am sorry I imagined that as I added to this feed it would be taken as read that I had followed the advice given here. That is what I have done and for some reason it isn't working. I have Stata 13. It says: proxy host not found
http://www.stata.com/ either
1) is not a valid URL, or
2) could not be contacted, or
3) is not a Stata download site (has no stata.toc file).
r(660);

[Ramadhani Abdul]

This was very helpful...Thanks @Kennedy