Discussion: View Thread

  • 1.  Confidence bands on Survival Curves

    Posted 02-08-2015 20:05

    Hello fellow consultants,

    I am working on a survival analysis with nine covariates (mostly control variables) and a class variable with the groups I need to compare. I would like to get 95% confidence bands on the survival curve so I can estimate the median survival time. Does anyone know if SAS can do this and what the code is?

    Thank you for any help you can give me.
    -------------------------------------------
    Teri Peterson
    Idaho State Univ
    -------------------------------------------


  • 2.  RE: Confidence bands on Survival Curves

    Posted 02-09-2015 14:19
    Do you want 95% confidence bands on the observed survival curve of each group, in which the values of the nine covariates are ignored? Or do you want 95% confidence bands on predicted survival curves conditional on the covariates being set equal to specified values such as zero?

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




  • 3.  RE: Confidence bands on Survival Curves

    Posted 02-09-2015 15:24
    Eric,
    Thank you for asking. I want 95% confidence bands on predicted survival curves conditional on the covariates set to specific values and group membership.

    -------------------------------------------
    Teri Peterson
    Idaho State Univ
    -------------------------------------------




  • 4.  RE: Confidence bands on Survival Curves

    Posted 02-09-2015 15:55
    And will this be so you can obtain 95% confidence limits on the median survival of each group?

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




  • 5.  RE: Confidence bands on Survival Curves

    Posted 02-09-2015 16:53
    Teri,

    I can't give you code, but I can suggest you look at the SAS documentation for the PHReg Procedure, and look specifically for (1) the Baseline statement, and (2) the "Survival Curves" example . Together, they will describe how to go about generating the hypothetical "baseline" survival curve when all the covariates are set equal to zero (or some other set of values). Note that group membership is considered a covariate, so if you want a separate "baseline" curve for each of your groups, you'll have to use a By statement. You can ask the baseline survival curve to be accompanied by (vertical) 95% confidence limits, thus providing the desired confidence band. The time values at which the (vertical) 95% confidence limits on survivals cross the median will provide the Brookmeyer-Crowley (horizontal) 95% confidence limits on the median survival time. 

    There are undoubtedly some trial-and-error sessions you'll have to go through, but I'm confident that you can eventually coax SAS into giving you what you want using the above approach. Good luck.  

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