Discussion: View Thread

  • 1.  Two primary endpoints

    Posted 09-03-2015 15:24

    Hi

    I am wondering if anyone has experience in having designed a study with two primary endpoints. 

    Can you call the study successful if only one endpoint reaches statistical significance and the other does not? 

    Thanks

    Shelley

    ------------------------------
    Shelley-Ann Walters
    3M
    ------------------------------



  • 2.  RE: Two primary endpoints

    Posted 09-03-2015 15:36

    If you keep in mind that the objective of any study involving hypothesis testing is not to find statistical significance but to test the hypothesis(es) in a valid way, then a well design study is successful whether one rejects or does not reject the null hypothesis.  The question is: was the study designed to have power to answer both questions?  If so, you found the answers for the hypotheses you were testing,

    Marcia

    ------------------------------
    Marcia Ciol
    Research Associate Professor
    University of Washington
    ------------------------------




  • 3.  RE: Two primary endpoints

    Posted 09-03-2015 15:58

    Shelley,

    This is a common problem in pharma.  It depends on whether you define overall success as "at least one" or "both" -- one is a union-intersection test (composite alternative is a union & composite null is intersection) and the other is intersection-union assuming the individual tests are one-sided. The former requires adjustments to maintain the familywise error rate. Either way, this must be done in the protocol stage and not post hoc. There is a nice tutorial in Statistics in Medicine by Dmitrienko and D'Agostino (2013) that discusses several multiple testing procedures for union-intersection tests.

    Chris

    ------------------------------
    Christopher Vahl
    Assistant Professor of Statistics
    Kansas State University
    ------------------------------




  • 4.  RE: Two primary endpoints

    Posted 09-03-2015 16:45

    Chris has provided an excellent response.  I am writing to highlight his first point because it reflects a problem that I find to be common in this and similar forums -- terms without a unique accepted meaning should be defined.  In my opinion an experiment is a statistical success if it is designed, performed and interpreted correctly -- regardless of the outcome.  If you have a non-statistical concept of success than the statistician should not be making the decision.  



    ------------------------------
    Charles Mann
    Charles R Mann Associates, Inc.
    ------------------------------


    Original Message


  • 5.  RE: Two primary endpoints

    Posted 09-08-2015 14:09

    Colleagues in Japan have led the development of a recent (brief) book on sizing trials with multiple endpoints:

     

    http://link.springer.com/book/10.1007/978-3-319-22005-5

     

    Thank you.

     

    Scott

     



    ------Original Message------

    Shelley,

    This is a common problem in pharma.  It depends on whether you define overall success as "at least one" or "both" -- one is a union-intersection test (composite alternative is a union & composite null is intersection) and the other is intersection-union assuming the individual tests are one-sided. The former requires adjustments to maintain the familywise error rate. Either way, this must be done in the protocol stage and not post hoc. There is a nice tutorial in Statistics in Medicine by Dmitrienko and D'Agostino (2013) that discusses several multiple testing procedures for union-intersection tests.

    Chris

    ------------------------------
    Christopher Vahl
    Assistant Professor of Statistics
    Kansas State University
    ------------------------------


    Original Message