s-news
[Top] [All Lists]

Plotting parametric survival curves

To: "'s-news@lists.biostat.wustl.edu'" <s-news@lists.biostat.wustl.edu>
Subject: Plotting parametric survival curves
From: "Paul, David A" <paulda@BATTELLE.ORG>
Date: Tue, 18 Jun 2002 18:13:29 -0400
Apologies in advance for the large number of questions I have been posting
lately.

First, Bill Dunlap [bill@insightful.com] was able to answer my question
about using the 
summary( ) command to summarize a censorReg object making use of the strata(
)
option.  He wrote:

"Try editing print.summary.censorReg and adding ... to the end
of the argument list so it starts with the line function(x, digits 
= max(options()$digits - 4, 3), quote = T, prefix = "", ...)
Does that fix it?  (If not, executing traceback() will show
the offending function and it will need the trailing ellipsis.)
This has been fixed in Splus6.1."

This worked very well.  Second, now that I can properly summarize a
parametric censorReg
object with strata, I would like to be able to plot the estimated curves and
see some 
regression diagnostics.  On page 467 of "Guide to Statistics, Vol. II" for
Splus 6.0, the 
command

plot(fit1)

is recommended.  Apparently this only works when the object stored in "fit1"
is a simple
censorReg object (the strata( ) option creates a list of censorReg objects,
one for each
stratum).  Passing the results into plot( ) of censorReg with the strata( )
option gives 
the error message

"Problem in censor(rr, cc, type = tc, inCodes = "1-4"): Need "time", 
"time2" and "censor.codes" for interval data. Use traceback() to see the 
call stack"

I subsetted my data.frame to include only those members of one stratum.
After creating
a censor object with the censor( ) command, I ran censorReg( ) and plot( ).
This yielded
the error message

"Problem in plotfun1(fits, Residuals, censor.codes, outCodes = ocd, fit.lab,

add.censored, ..: There is no uncensored observation to be plotted.  Use 
traceback() to see the call stack."

It is a fact that my data are all either right censored or interval censored
due to the way that
the subjects were monitored during the course of the study.  Is there a way
to plot the best
fit Weibull curve(s) to the strata in my data.frame given the censoring in
my data?  Since the
subjects were monitored every 8 hours, I could simplify matters by assuming
that an event
occurred at the beginning of a period in which an event was "noticed", but
would rather not.

 David Paul, Ph.D.
  Battelle Memorial Institute
  505 King Avenue
  Columbus, OH  43201
  614.424.3176


<Prev in Thread] Current Thread [Next in Thread>
  • Plotting parametric survival curves, Paul, David A <=