Power Calculation using Stata

Discussion in 'Scientific Statistics Math' started by ZQ, Nov 1, 2011.

  1. ZQ

    ZQ Guest

    Hi,

    I am involved in a clinical study where I need to ascertain that the
    study has adequate power. The study aims at understanding the effect
    of a program on patients. We are required to have a control and a
    treatment sample. I believe I can recruit about 22 individuals for the
    study in each group. We think a repeated measures approach
    (observation at time =0, time =3months, time =6 months) would be best.
    Could you please help me understand how I can do a power calculation
    using STATA for such a project? I have heard about "sampsi" but am not
    sure if that would help. Please advise.
     
    ZQ, Nov 1, 2011
    #1
    1. Advertisements

  2. ZQ

    Rich Ulrich Guest


    You can probably figure out for yourself how to use STATA if you
    figure out what the simple version of your hypothesis is.

    WHY do you have two followups? I would figure, usually, that
    you must expect continued change in the improved group, or
    else you expect immediate change and hope to demonstrate
    that the improvement perseveres.

    The "simple test" for these two circumstances will ignore
    either the middle period or the end period, for the sake of
    the power analysis. For continued improvement, you might
    want to compute your formal test as a comparison of the
    two linear trend lines, but that is more complicated than you
    need to (or want to) go into for the sake of a power analysis.
    Remember, this is an approximation, since you are having to
    estimate the variances and changes anyway.


    Either of those becomes a comparison of two change scores, or
    of two regressed-change scores. The correlation of pre and post
    tells you how much you can reduce the variance of sum of Pre
    and Post, if you do the eventual power analysis as the diffference
    in scores between two groups ... just about the simplest, eventual
    power analysis, after the preliminary steps.

    - If you are selecting your samples as a "down" point in the cycle
    of disease, then you are stuck with improvement being likely in
    both groups, which means you need a treatment change that is
    still large when added to the no-treatment change. - This is
    classically a problem for designs in psychiatry, where I have my
    own experience. At "hospital admission", virtually everyone is
    expected to improve quite a bit in the first couple of weeks.
    Because that is what happens, long-term followups are very
    often designed to start with Day 0 being a week or two later.
     
    Rich Ulrich, Nov 1, 2011
    #2
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.