Discussion: View Thread

  • 1.  Appropriate test for Tumor Recurrence With Historical Control

    Posted 10-17-2012 16:57
    I have a study design that uses a historical control group where the subject level information is not available for the control.  The primary endpoint is recurrence at 5 and 10 years.  What is the best and most efficient test for time point specific (5-year and 10-year) comparisons?

    -------------------------------------------
    Kathy Warner
    Consultant Services

    -------------------------------------------


  • 2.  RE:Appropriate test for Tumor Recurrence With Historical Control

    Posted 10-17-2012 17:05
    Is that one primary endpoint or two?

    -------------------------------------------
    Eric Siegel
    Biostatistician
    Univ of Arkansas for Medical Sciences
    -------------------------------------------





  • 3.  RE:Appropriate test for Tumor Recurrence With Historical Control

    Posted 10-17-2012 17:09

    It is one primary endpoint evaluated at two time points.
    -------------------------------------------
    Kathy Warner
    Consultant Services

    -------------------------------------------








  • 4.  RE:Appropriate test for Tumor Recurrence With Historical Control

    Posted 10-17-2012 18:53

    Kathy:

      I presume that you have an estimate of the 5-yr and the 10-yr recurrence rates in your historical control patients.   In other words, I presume that you do have what might be considered as population reference values.

      So, you have a 1-sample design and you wish to size your new study to test your future estimates of the 5-yr and the 10-yr recurrence rates against their respective population reference values.

      If all that is correct, then you can determine the required sample size (separately) for each endpoint to be tested using the "One Arm Survival" program at the following website:
    http://www.crab.org/Resources/StatisticalTools.aspx

      They cite a methodolgy reference there too, which you can cite in your stat'l design writeup if you need to.  

    Please note that the Normal theory assumption for this program yields smaller N's than will result from using the related "One Arm Nonparametric Survival" program also available there.

      I hope the above suggestion proves both on target for your study design problem, and helpful to solving it.   Beware that you are embarking on a study that will require long-term follow-up.

    -------------------------------------------
    Lance Heilbrun
    Karmanos Cancer Institute Biostatistics Core
    -------------------------------------------








  • 5.  RE:Appropriate test for Tumor Recurrence With Historical Control

    Posted 10-17-2012 19:19

    Please forgive my initial post.  I don't believe I defined the problem adequately.  My treatment arm is prospectively enrolling with enrollment to be completed in two years.  My control arm has multiple, historical studies with aggregate data extracted from study reports.  I have a reference from Statistics in Medicine, 2008 titled "Meta-analysis of summary survival curve data."  The article details a method using a multivariate random-effects framework with SAS Proc MIXED.  The method makes sense, but I would appreciate any feedback from anyone who has implemented this or any other method for time point specific comparisons where control groups have only summary data.

    -------------------------------------------
    Kathy Warner
    Consultant Services

    -------------------------------------------








  • 6.  RE:Appropriate test for Tumor Recurrence With Historical Control

    Posted 10-17-2012 19:36
    Okay, I can see why you say it's only one primary endpoint.  My past practice has been to tell my clients that, if they evaluate recurrence at two different timepoints like that, then it's really two primary endpoints...and then my past practice has been to ask them which timepoint (or endpoint) is the one that's more important to them for evaluating treatment success.  The reason why I do that is so that I can declare one timepoint to be the sole primary endpoint, so I can allocate all my alpha onto that one endpoint & thereby avoid doing a Bonferroni adjustment if I'm calculating power and sample size. 

    Just curious: is this a study that is being proposed? or a study that has already been conducted?

    -------------------------------------------
    Eric Siegel
    Biostatistician
    Univ of Arkansas for Medical Sciences
    -------------------------------------------





  • 7.  RE:Appropriate test for Tumor Recurrence With Historical Control

    Posted 10-17-2012 20:07
    Thanks for your feedback.  This is a study proposal.

    -------------------------------------------
    Kathy Warner
    Consultant Services

    -------------------------------------------