Power analysis - reliable change score

[Bellinda]

Hello,

I am trying to do a power analysis for a grant. I have a fixed sample
size (secondary data analysis) of 3,400, and want to know the what
size of an effect I can detect given 80% and a p-value of .05. There
will be two measurements of cognitive functioning based on a battery
of cognitive tests. We are looking to see whether the change in MPOD
and serum carotenoids at Wave 3 relative to Wave 1 reflects cognitive
change over time. We are planning to look at change using the reliable
change score. There is very little literature to know what constitutes
a meaningful change, does anyone know the best way to approach this?
Any thoughts would be greatly appreciated.

Warm regards,
Bellinda

[Steve Simon, P.Mean Consulting]

Since you've already chosen your sample size, you don't need the minimum
clinically important difference. The sample size will tell you what size
difference you will be able to detect and then you will have to make a
judgment about whether that difference about right.

What you do need is the variability of your outcome measure. Surely
there is something in your literature review about MPOD and serum
carotenoids. Pull out one of those references and see what the
variability is. It is a bit trickier when you look at change from
baseline but getting a standard deviation for MPOD and/or for serum
carotenoids is a start.

Steve Simon, n...@pmean.com, Standard Disclaimer.
Sign up for the Monthly Mean, the newsletter that
dares to call itself average at www.pmean.com/news

[Bellinda]

Hi Steve,

I have some test retest reliabilities on MPOD, so I think this will help, right?

Thank you for your response.

Bellinda

> --
> To post a new thread to MedStats, send email to MedS...@googlegroups.com .
> MedStats' home page is http://groups.google.com/group/MedStats .
> Rules: http://groups.google.com/group/MedStats/web/medstats-rules



--
Dream as if you’ll live forever. Live as if you’ll die today — James Dean

[Steve Simon, P.Mean Consulting]

On 5/19/2012 4:17 AM, Bellinda wrote:

> I have some test retest reliabilities on MPOD, so I think this will
> help, right?

You can use the test-retest reliability as a possible measure of the
correlation between the baseline and the end of study measurement. It
will be a less than perfect measure, but almost all input into a power
calculation is less than perfect.

The key thing to avoid is using the variation from a test-retest
reliability study as that is often going to be inappropriate. The people
in a test-retest reliability study are not likely to be representative
of the patients that you see in your study. Look for a standard
deviation from a study that has a similar demographic profile and a
similar prognosis to your patients.