Discussion: View Thread

  • 1.  kaplan meier confidence intervals for failure

    Posted 12-23-2022 09:37
    I'm seeking advice on how to get the confidence limits for kaplan meier failure times out of SAS?  I'm using proc lifetest for a simple kaplan meier.  The outsurv option gives for each time point the probability of survival, probability of failure, standard error of survival and the confidence limits for survival.

    Since failure is just 1 - survival, would it be acceptable to use 1 - confidence limit for survival as my confidence limits for failure?
    Or, can I use failure +- 1.96(standard error of survival)?

    Any input is greatly appreciated.  

    Nancy Buderer, MS
    Nancy Buderer Consulting, LLC
    Biostatistician, Program Evaluator, Research Consultant

  • 2.  RE: kaplan meier confidence intervals for failure

    Posted 12-23-2022 10:12
    I think it depends on what you are trying to do here, but using 1-survival to get estimates of the "probability of failure before time t" sounds good (along with 1-confidence limits, but note the lower and upper limits will switch... you'll see what I mean). Just make sure that is what you are looking for! The SAS documentation overview of lifetest goes a bit into the relationship between the S(t) and F(t) = 1-S(t), and how to interpret.

    On the off chance that you are actually interested in the probability of failure at time t, you want to look into the hazard function instead. 

    As an aside, I'm not sure if calculating the confidence error via the standard error of survival is that straightforward. If you are using the default options, a loglog transformation is occurring somewhere in the calculation. Of course, you could test this in SAS by trying to use the SE(Survival) to replicate the confidence limits they provide (I loaded up a quick proc lifetest, but the output doesn't seem to include the SE(Survival) for me).

  • 3.  RE: kaplan meier confidence intervals for failure

    Posted 12-23-2022 10:40
    Hi Nancy

    If you are unable to get SAS to do confidence intervals, you may wish to use Stata or R for your figures. 

    With Stata for example, it is just a matter of adding ci to the code.

    I found samples of SAS code that do confidence intervals:

    7.1. Graphing Kaplan-Meier survival function estimates to assess proportional hazards for categorical covariates

    Do post back on how you get on....


    Dr Arun Chind 
    Proshen Health & Risk Consulting
    Wakefield, England. 


  • 4.  RE: kaplan meier confidence intervals for failure

    Posted 12-23-2022 15:54

    • Should you need an excellent reference book about KM, I used and cited (cit below)  a book by Klein & Moeschberger.  There are certainly other excellent texts. You can refer to the original  Kaplan Meier paper and they refer to the Greenwood calculation.
     I wrote a paper which does not -directly- answer your question but -might- help. We discovered that there was a difference in how SAS vs other packages (R, SPLUS , etc.) computed the mean of the Kaplan meier.  In our application (in health economics) the impact on the estimation was substantial
    I reference the KM confidence intervals settings in the software for SPLUS, SAS, R, in the paper. 

     I don't have a SAS manual and I'll assume you mean you're getting these from the ODS output. And there might be a "switch" to turn on to get the quantities you're interested in. SAS manual should have excellent documentation about the calculation.
    And SAS has confidence limits and confidence regions/bands.

    Chris Barker, Ph.D.
    2022 Statistical Consulting Section
    Consultant and
    Adjunct Associate Professor of Biostatistics

    "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