Discussion: View Thread

  • 1.  Type I Error

    Posted 04-01-2013 22:58

    Hi,

    I have a question regarding interim analysis and the rules for multiple looks.  I have two studies both using the same treatment methodology and measuring the same outcome.  One study is at the end of the planned five-year follow-up, and the other is at the four-year mark of a planned five-year follow-up.  The sponsor wants to do a manuscript combining the two studies with the data in its current state.  The primary endpoint is a survival endpoint.  I need to know what are the implications of doing the analysis.  Since one study will not be complete at the time of the planned analysis, is this a situation where you would need to split the planned alpha?

    Thank you,

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


  • 2.  RE:Type I Error

    Posted 04-01-2013 23:44
    Kathy, In this situation I dont think there is an "alpha" penalty for this -- there is no "penalty for looking at data" there is a penalty when the interim analysis can result in a stop, claim, or "victory" it is actions or possible actions, that cause penalties... there is nothing in it that "looking" at data causes a penalty of some sort.  Now, concerns could be raised on what was done at this interim... would they have claimed significance if it were reached?? Can you present a p-value in the paper and say "significant"... this becomes a possible issue -- but not insurmountable.  You need to report that the primary time hasn't been reached and the primary of the second trial not yet conducted. This is in part why typically researchers are hypersensitive about just not looking.  

    What a concern here can be is "operational bias" -- that by disclosing some data -- by investigators, patients, etc, seeing information about outcomes affects behavior of the clinicians and subjects (maybe that drop from the study, maybe seek other treatments... ) It is likely more worry than actual problem -- and given you are 4 years in a 5 year study hard to imagine it could impact things... but be prepared to defend this issue.  The kind of endpoint, the type of treatment, etc could make this a non-issue, or a possible issue.

    -------------------------------------------
    Scott Berry
    Berry Consultants
    -------------------------------------------








  • 3.  RE:Type I Error

    Posted 04-02-2013 00:11
    Kathy, some additional questions -  is this a blinded study? What kind of treatments and procedures are being evaluated?  One  important question is what is the potential benefit to  the patients by carrying out the  analysis, if you discover that one treatment works remarkably well?   What kind of decisions could be made once you have analyzed the data- is it possible for  patients to have an option  for the new treatment or procedure if one of the treatments works very well ? By splitting alpha, do you mean an "unplanned" interim?  If so, that can be accommodated using group sequential methods.

    -------------------------------------------
    Chris Barker, Ph.D.

    www,barkerstats.com

    ---
    "In composition you have all the time you want to decide what to say in 15 seconds, in improvisation you have 15 seconds."
    -Steve Lacy
    -------------------------------------------








  • 4.  RE:Type I Error

    Posted 04-02-2013 10:06

    Thanks to everyone that responded.  I believe the answer is that there is no penalty in this type of look at the data.  This is a situation where I thought I knew the answer based on everything I've ever learned.  However, I was having a difficult time convincing the lead investigator on the study.  His position was so fervent, that I started to doubt myself.  Thanks again for taking the time to consider my question.

    Regards,
    -------------------------------------------
    Kathy Warner
    Consultant Services
    Statistical Services
    -------------------------------------------