Dear Statalist,
especially programmers knowledgeable in survival analysis
Last September, Todd Alonzo wrote:
> Has anyone developed a Stata program that estimates and plots cumulative
> incidence for the estimation of failure probabilities in the presence of
> competing risks? See, for example, Gooley et al. (1999) Statistics in
> Medicine 18: 695-706 for a description of cumulative incidence and an
> illustration of why the complement of the Kaplan-Meier estimate is not
> appropriate when there are competing risks.
And Bill Dieterich replied:
Vim van Putten's -sby- generates a table with survival statistics. It
includes an option for calculating failure probabilities in the presence of
competing risks (what Marubini & Valsecchi refer to as the
crude cumulative incidence). The user specifies the time points at which
the probabilities are calculated and displayed. At a recent Stata Users
Conference, an abstract was submitted describing a similar program called
-mtable- by Mohamed Ali. Producing graphs from the results of -sby- is
feasible for small samples. I hope this helps somewhat. I would be
interested to know if you discover anything more promising.
It seems that -mtable- has not been available yet?
I would second the request to have an -st- option or an ado file that
would calculate and plot cumulative survival instead of the Kaplan-Meier
estimate. I have to analyse a dataset of ca. 300 persons who have a
follow-up of up to 25 years, and naturally there is a significant number
of deaths from competing risks and, hence, the Kaplan-Meier plot is
seriously flawed. It would be tedious to use -sby- for this.
The cumulative incidence plot in this context is finding more and more
advocates. A chapter in Handbook of Statistics in Clinical
Oncology, 2001, is very provocative on this issue: "The wide availability
of statistical software packages capable of calculating KM estimates but
which do not calculate cumulative incidence (CI) estimates undoubtedly
contributes to the frequent misuse of 1-KM estimate", and "Although we
have not seen the CI estimate offered commercially in any software
package, its calculation is not computationally difficult and programs
that accomplish this are reasonably straightforward to write".
I also noticed that he has made available a -cmprsk- module for Splus this
Summer that accomplishes cumulative incidence analysis:
cuminc Cumulative Incidence Analysis
plot.cuminc Create Labeled Cumulative Incidence Plots
print.cuminc Print cuminc objects
(http://cran.r-project.org/src/contrib/PACKAGES.html#cmprsk)
Could we have this type cumulative incidence analysis also in Stata,
which is otherwise so well versed in survival statistics?
With anticipation,
T.Kivela
Univ.of Helsinki
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
