Dear R gurus,
for a simulation concerning study effects and historical controls
in survival analysis, i would like to experiment with a gaussian
frailty model.
The simulated scenario consists of a randomized trial
(treatment and placebo) and historical controls (only placebo).
So the simulated data frames consist of four columns
$time, $cens, $study, $treat.
$time, $cens are the usual survival data.
For the binary thretment indicator we have
$treat == 0 or 1, if $study == 1,
$treat == 1 if $study > 1
Typical parameters for my simulations are:
sample sizes (per arm): between 100 and 200
number of historical studies: between 7 and 15
hazard ratio treatment/placebo: between 0.7 and 1
variance of the study effekt: between 0 and 0.3
Depending on the sample sizes, the following call sometimes leads to
a segmentation fault:
coxph(Surv(time,cens) ~
as.factor(treatment) + frailty(study, distribution="gaussian"),
data=data)
I noticed, that this segmentation fault occures most frequently, if the
number of randomized treatment patients is higher than the number of
randomized placebo patients, and the number of historical studies is
large.
There seems to be no problem, if there are at least as many randomized
placebo patients as treated patients. Unfortunately, this is not the
situation i want to investigate (historical controls should be used
to decrease the number of treated patients).
Is there a way to circumwent this problem?
Christian
P.S.
Is it allowed, to attach gzipped sample data sets in this mailing list?
We really need a reproducible example to find segmentation faults. Can you make one? -thomas On Wed, 28 Jul 2004, Christian Lederer wrote:> > Dear R gurus, > > for a simulation concerning study effects and historical controls > in survival analysis, i would like to experiment with a gaussian > frailty model. > > The simulated scenario consists of a randomized trial > (treatment and placebo) and historical controls (only placebo). > > So the simulated data frames consist of four columns > $time, $cens, $study, $treat. > $time, $cens are the usual survival data. > For the binary thretment indicator we have > $treat == 0 or 1, if $study == 1, > $treat == 1 if $study > 1 > > Typical parameters for my simulations are: > sample sizes (per arm): between 100 and 200 > number of historical studies: between 7 and 15 > hazard ratio treatment/placebo: between 0.7 and 1 > variance of the study effekt: between 0 and 0.3 > > Depending on the sample sizes, the following call sometimes leads to > a segmentation fault: > > coxph(Surv(time,cens) ~ > as.factor(treatment) + frailty(study, distribution="gaussian"), > data=data) > > I noticed, that this segmentation fault occures most frequently, if the > number of randomized treatment patients is higher than the number of > randomized placebo patients, and the number of historical studies is > large. > There seems to be no problem, if there are at least as many randomized > placebo patients as treated patients. Unfortunately, this is not the > situation i want to investigate (historical controls should be used > to decrease the number of treated patients). > > Is there a way to circumwent this problem? > > Christian > > P.S. > Is it allowed, to attach gzipped sample data sets in this mailing list? > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://www.stat.math.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html >Thomas Lumley Assoc. Professor, Biostatistics tlumley at u.washington.edu University of Washington, Seattle
Dear Thomas, attached you find a data frame which produces the error. I am using survival 2.11-5 under R 1.9.1-1 and 1.9.0-1. By the way, if i randomly omit 50% of the data, i usually get no crash, but a warning message like this: Inner loop failed to coverge for iterations 1 2 3 in: coxpenal.fit(X, Y, strats, offset, init = init, control, weights = weights, Maybe, the model is not appropriate for this kind of data. But on the other hand, as soon the treatment group (study == 1, treatment == 1) is smaller than the randomized placebo group (study == 1, treatment == 0), the warnings disappear. and the model gives reasonable results in my first simulations with normally distributed study effects. Christian Thomas Lumley wrote:> We really need a reproducible example to find segmentation faults. Can > you make one? > > -thomas > > > On Wed, 28 Jul 2004, Christian Lederer wrote: > > >>Dear R gurus, >> >>for a simulation concerning study effects and historical controls >>in survival analysis, i would like to experiment with a gaussian >>frailty model. >> >>The simulated scenario consists of a randomized trial >>(treatment and placebo) and historical controls (only placebo). >> >>So the simulated data frames consist of four columns >>$time, $cens, $study, $treat. >>$time, $cens are the usual survival data. >>For the binary thretment indicator we have >>$treat == 0 or 1, if $study == 1, >>$treat == 1 if $study > 1 >> >>Typical parameters for my simulations are: >>sample sizes (per arm): between 100 and 200 >>number of historical studies: between 7 and 15 >>hazard ratio treatment/placebo: between 0.7 and 1 >>variance of the study effekt: between 0 and 0.3 >> >>Depending on the sample sizes, the following call sometimes leads to >>a segmentation fault: >> >>coxph(Surv(time,cens) ~ >> as.factor(treatment) + frailty(study, distribution="gaussian"), >> data=data) >> >>I noticed, that this segmentation fault occures most frequently, if the >>number of randomized treatment patients is higher than the number of >>randomized placebo patients, and the number of historical studies is >>large. >>There seems to be no problem, if there are at least as many randomized >>placebo patients as treated patients. Unfortunately, this is not the >>situation i want to investigate (historical controls should be used >>to decrease the number of treated patients). >> >>Is there a way to circumwent this problem? >> >>Christian >> >>P.S. >>Is it allowed, to attach gzipped sample data sets in this mailing list? >> >>______________________________________________ >>R-help at stat.math.ethz.ch mailing list >>https://www.stat.math.ethz.ch/mailman/listinfo/r-help >>PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html >> > > > Thomas Lumley Assoc. Professor, Biostatistics > tlumley at u.washington.edu University of Washington, Seattle > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://www.stat.math.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html > >