Proc LIFETEST - Kernel smoothing for Kaplan Meier

[mmhxc5]

Hi all,

We are using Kaplan Meier method to model bridge deterioration. What I exactly want is to get smoothed curve output for survival and deterioration as an output in order to prepare graphs in graphing software. I could not generate nice survival plots using SAS. I know smoothed curves for other methods in proc LIFETEST are available, but not available for Kaplan Meier.

Could anyone give me a link to a macro or write me a macro to output the Kernel smoothed curve for Kaplan Meier data points?

Thanks,

[Reeza]

Can you use the output from LIFETEST and pass that to another proc to generate a smooth curve? Though I wonder if you shouldn't be using LIFEREG if you need a smooth curve. It's more normal to use a step curve for survival graphs than smoothed graphs. SAS can generate pretty complicated graphics these days so I'm a bit surprised you couldn't get what you need though it's definitely more work than a point and click software. Though the benefit of code is always, if it needs to be repeated you save the time :).

[Rick_SAS]

Please provide the code you are using. If you can provide sample data, that's even better.

[mmhxc5]

Thank you @Reeza and @Rick_SAS for your time and interest to help.

The following is the code i am using.

title;
proc LIFETEST method=km data=xx.WI_nbi_psc_super_ticr plots=s(cb=ep)OUTSURV=xx.PSC_Survival;* (atrisk);
TIME count*CENSOR(0);
strata SUPERSTRUCTURE_COND_059;
label count="Time in Condition Rating (Year)";
run;

The data file I used in the above code is attached. The code output and graphs are attached too. I want to smooth the step curves using Kernel smoothing method and have it as an output similar to the code output and possibly overlay it on the graph.

Thanks,

[Rick_SAS]

Although it is possible to apply a smoother to the points in the K-M graph, I don't recommend it. Smoothers are based on nonparametric regression methods that assume a relationship between the X and Y variables. The predicted values and confidence intervals for a smoother are based on assumptions that probably do not hold for the K-M points.

If you decide to proceed anyway, you can use the LOESS or PBSPLINE statements in PROC SGPLOT to overlay a smoother on the data. For example, here is a PBSPLINE smoother:

proc lifetest method=km data=sashelp.BMT plots=s(cb=ep) OUTSURV=PSC_Survival;
   time T * Status(0);
   strata Group;
run;

proc sort data=PSC_Survival;
by Group T;
run;

proc sgplot data=PSC_Survival;
band x=T lower=SDF_LCL upper=SDF_UCL / group=Group transparency=0.6;
pbspline x=T y=Survival / group=Group;  /* or use LOESS */
run;