R help - Feb 2012 - Reporting Kaplan-Meier / Cox-Proportional Hazard Standard Error, km.coxph.plot, survfit.object

What is the best way to report the standard error when publishing
Kaplan-Meier plots?  In my field (Vascular Surgery), practitioners
loosely refer to the "10% error" cutoff as the point at which to stop
drawing the KM curve.  I am interpreting this as the *standard error
of the cumulative hazard*, although I'm having a difficult time
finding some guidelines about this (perhaps I am not searching the
correct terms or references).  My KM figures contain typically two
curves that I am comparing using the logrank test.  Inspecting the
?survfit.object yields the std.err field that gives the standard error
for each timepoint on the curve.

Is it recommended that I just name the timepoint at which the standard
error exceeds 0.1 in the figure legend?  For example, "The standard
error exceeds 10% at time points beyond 394 days."  I have seen this
strategy in other publications.

What is your approach?

Thanks for your help,
PCJ

The 10% rule does not provide a unique answer.  Should it apply to the
cumulative probability, its logarithm, or log-log (log hazard scale)?  Many
studies are too small to achieve 10% at any time point.  I think it is more
traditional (but not without bias) to stop where fewer than 10 subjects are
still being followed.  There's room for many other choices though. 
Sometimes I think that the curve should go to the max but be accompanied by
confidence bands.
Frank

Paul Johnston wrote
> 
> What is the best way to report the standard error when publishing
> Kaplan-Meier plots?  In my field (Vascular Surgery), practitioners
> loosely refer to the "10% error" cutoff as the point at which to
stop
> drawing the KM curve.  I am interpreting this as the *standard error
> of the cumulative hazard*, although I'm having a difficult time
> finding some guidelines about this (perhaps I am not searching the
> correct terms or references).  My KM figures contain typically two
> curves that I am comparing using the logrank test.  Inspecting the
> ?survfit.object yields the std.err field that gives the standard error
> for each timepoint on the curve.
> 
> Is it recommended that I just name the timepoint at which the standard
> error exceeds 0.1 in the figure legend?  For example, "The standard
> error exceeds 10% at time points beyond 394 days."  I have seen this
> strategy in other publications.
> 
> What is your approach?
> 
> Thanks for your help,
> PCJ
> 
> ______________________________________________
> R-help@ mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 

-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context:
http://r.789695.n4.nabble.com/Reporting-Kaplan-Meier-Cox-Proportional-Hazard-Standard-Error-km-coxph-plot-survfit-object-tp4403045p4404200.html
Sent from the R help mailing list archive at Nabble.com.