Discussion: View Thread

  • 1.  Sample Size Calculation for a Superiority Trial

    Posted 01-18-2011 11:51
    Dear Colleagues,

    I'm interested in knowing if there's a way to compute the sample size required for designing a superiority trial using PASS and nQuery Advisor?  The manuals for both packages are not very clear on computing this for a superiority trial.  The indexes for the manuals also do not specify 'superiority' in their list of key words to be able to find easily.

    Design of the Trial:  There are three treatment groups - one is current standard of care (SOC), and the other two groups are two different doses of the new treatment.  The end point being measured is binary and higher rates indicate success and therefore, is desired.

    Current Assumptions: Desired Power = 0.8, Type I Error Rate = 5%, Proportion of Success in SOC arm = 0.9, Proportion of Success in New Treatment arm = 0.95 (regardless of dose).

    Any insights would be greatly appreciated.

    Thank you,

    Sincerely,
    -------------------------------------------
    Mamtha Balasubramaniam
    Sr. Biostatistician
    Beaumont Hospitals
    -------------------------------------------


  • 2.  RE:Sample Size Calculation for a Superiority Trial

    Posted 01-18-2011 12:03
    A program like nQuery specifies noninferiority and equivalence because they are special and require the use of margins.  Superiority is just current lingo particularly used at FDA to describe the standard one sided hypothesis test.  All these packages do superiority.

    -------------------------------------------
    Michael Chernick
    Director of Biostatistical Services
    Lankenau Institute for Medical Research
    -------------------------------------------








  • 3.  RE:Sample Size Calculation for a Superiority Trial

    Posted 01-18-2011 12:13
    Hi,

    So, wouldn't a simple 1-tailed test do the trick, i.e., double the alpha in a test for inequality (unless there is a specific 1-tailed option)?

    -------------------------------------------
    Stephan Arndt
    Professor
    University of Iowa, Iowa Consortium
    -------------------------------------------






    Original Message


  • 4.  RE:Sample Size Calculation for a Superiority Trial

    Posted 01-18-2011 12:28

    Exactly, I thought it was obvious from what I said.
    -------------------------------------------
    Michael Chernick
    Director of Biostatistical Services
    Lankenau Institute for Medical Research
    -------------------------------------------






    Original Message


  • 5.  RE:Sample Size Calculation for a Superiority Trial

    Posted 01-18-2011 13:31


    -------------------------------------------
    Patricia Fox
    Research/Consultant
    Fox Statistical Consulting & Applied Research
    -------------------------------------------
    It would be most appropriate to use a binomial or poisson model approach since the probability of
    superiority is assumed to be 0.9... which would still be a simple calculation unless the programs he mentioned use only the normal approximation.





    Original Message


  • 6.  RE:Sample Size Calculation for a Superiority Trial

    Posted 01-18-2011 14:16

    Take a look at my paper in the American Statistician in 2002 with Christine Liu.  We compared several packages with respect to how they handled the saw-toothed behavior.  At that time nQuery which is Janet Elashoff's product ignored the issue by only allowing you to specify sample size and it would calculate power.  Since there is not a unique crossing point of the power curve it is not clear what the best was is to determine sample size given power.Agresti and others have recommendations and you could of course use trial and error to construct the power curve and make a decision.  StatXact, Power and Precision, PASS and SAS are much more explicit about the problem.  I also cover this issue in Chapter 16 of my biostatistics book. When using a normal approximation the issue goes away and all packages can give you a sample size based on specifying power.  The saw-toothed issue is a problem with any discrete distribution model.  I just am not familiar with how they handle it.  Also I do not use nQuery any more since I have SAS which since 2002 developed a power procedure and a glmpower procedure based on Ralph O'Brien's software that was in my 2002 comparison.
    -------------------------------------------
    Michael Chernick
    Director of Biostatistical Services
    Lankenau Institute for Medical Research
    -------------------------------------------






    Original Message