Multilevel Survival Analysis - Cox PH Model

Hello all, thanks for your time and patience.

I'm looking for a method in R to analyse the following data:

Time to waking after anaesthetic for medical procedures repeated on the same
individual.
str(mysurv)
labelled [1:740, 1:2] 20  20  15  20  30+ 40+ 50  30  15  10  ...
 - attr(*, "dimnames")=List of 2
  ..$ : NULL
  ..$ : chr [1:2] "time" "status"
 - attr(*, "type")= chr "right"
 - attr(*, "units")= chr "Day"
 - attr(*, "time.label")= chr "ORIENTATION"
 - attr(*, "event.label")= chr "FullyOrientated"

mysurv is constructed from the following data:

head(data.frame(MRN, ORIENTATION, FullyOrientated))

       MRN ORIENTATION FullyOrientated
1 0008291           20               2
2 0008469           20               2
3 0008469           15               2
4 0010188           20               2
5 0013664           30               1
6 0014217           40               1

I had planned to use a Cox PH model to analyse time to waking (ORIENTATION =
10, 15, 20 mins ....... 50 mins) and whether or not people (MRN) are fully
awake within an hour (FullyOrientated). I've put  GENDER, etc. into the
model but I have the following bias:

The procedure is repeated weekly on each individual (MRN), so each
individual has 5-9 cases associated with them. Currently I am including
these in the model as if they were independent. 

Is there a way to account for the non-independence of these waking times?

I'm thinking of something similar to the NLMER package and Multilevel /
Mixed Effects analysis as described in Pinheiro and Bates.

I'd be appreciative of any help at all?

Thanks again,

R

--
View this message in context: http://r.789695.n4.nabble.com/Multilevel-Survival-Analysis-Cox-PH-Model-tp3638278p3638278.html
Sent from the R help mailing list archive at Nabble.com.

Hello all, thanks for your time and patience.

I'm looking for a method in R to analyse the following data:

Time to waking after anaesthetic for medical procedures repeated on  
the same
individual.

str(mysurv)
labelled [1:740, 1:2] 20  20  15  20  30+ 40+ 50  30  15  10  ...
- attr(*, "dimnames")=List of 2
 ..$ : NULL
 ..$ : chr [1:2] "time" "status"
- attr(*, "type")= chr "right"
- attr(*, "units")= chr "Day"
- attr(*, "time.label")= chr "ORIENTATION"
- attr(*, "event.label")= chr "FullyOrientated"

mysurv is constructed from the following data:

head(data.frame(MRN, ORIENTATION, FullyOrientated))

      MRN ORIENTATION FullyOrientated
1 0008291           20               2
2 0008469           20               2
3 0008469           15               2
4 0010188           20               2
5 0013664           30               1
6 0014217           40               1

I had planned to use a Cox PH model to analyse time to waking  
(ORIENTATION =
10, 15, 20 mins ....... 50 mins) and whether or not people (MRN) are  
fully
awake within an hour (FullyOrientated). I've put  GENDER, etc. into  
the
model but I have the following bias:

The procedure is repeated weekly on each individual (MRN), so each
individual has 5-9 cases associated with them. Currently I am  
including
these in the model as if they were independent.

Is there a way to account for the non-independence of these waking  
times?

I'm thinking of something similar to the NLMER package and  
Multilevel /
Mixed Effects analysis as described in Pinheiro and Bates.
Have you looked at the coxme package?
David Winsemius, MD
West Hartford, CT

On Jul 1, 2011, at 10:10 AM, dunner wrote:

Hello all, thanks for your time and patience.

I'm looking for a method in R to analyse the following data:

Time to waking after anaesthetic for medical procedures repeated on  
the same
individual.

str(mysurv)
labelled [1:740, 1:2] 20  20  15  20  30+ 40+ 50  30  15  10  ...
- attr(*, "dimnames")=List of 2
..$ : NULL
..$ : chr [1:2] "time" "status"
- attr(*, "type")= chr "right"
- attr(*, "units")= chr "Day"
- attr(*, "time.label")= chr "ORIENTATION"
- attr(*, "event.label")= chr "FullyOrientated"

mysurv is constructed from the following data:

head(data.frame(MRN, ORIENTATION, FullyOrientated))

     MRN ORIENTATION FullyOrientated
1 0008291           20               2
2 0008469           20               2
3 0008469           15               2
4 0010188           20               2
5 0013664           30               1
6 0014217           40               1

I had planned to use a Cox PH model to analyse time to waking  
(ORIENTATION =
10, 15, 20 mins ....... 50 mins) and whether or not people (MRN)  
are fully
awake within an hour (FullyOrientated). I've put  GENDER, etc. into  
the
model but I have the following bias:

The procedure is repeated weekly on each individual (MRN), so each
individual has 5-9 cases associated with them. Currently I am  
including
these in the model as if they were independent.

Is there a way to account for the non-independence of these waking  
times?

I'm thinking of something similar to the NLMER package and  
Multilevel /
Mixed Effects analysis as described in Pinheiro and Bates.
Have you looked at the coxme package?
As an initial strata()-gem, as it were, perhaps just adding  
strata(MRN) may parcel out the intra-individual variability and  
degrees of freedom, so that they are not inappropriately included in  
the IV's. My initial suggestion of coxme may be overkill.
-- 

David Winsemius, MD
West Hartford, CT

______________________________________________
R-help at r-project.org 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.
David Winsemius, MD
West Hartford, CT
Is there any right censoring?

If not, then plain old lme, lmer, gam (in mgcv), ... etc. would seem
to me do just fine for time to waking = ORIENTATION as a response --
or are you thinking of this as interval-censored data, which it would
appear to be since you've binned the response?  I strongly suspect
that the simpler approach would work pretty well, anyway, but ...

For FULLY ORIENTED, which is binary, glmm, glmer, etc. would seem to
do if there's no censoring.

I'd be happy doing this even for a "small" amount of right censoring
(uh-oh) and the usual hack cheats --replacing the censoring time with
something slightly bigger than the censoring time, for example. But
here I'd be getting by knuckles slapped -- whack whack! -- for
statistical hackery, probably justifiably, so you probably need to
retreat to the original approach if that's the case.

I would welcome criticism/correction on this suggestion if I'm all wet
-- or even just damp.

Cheers,
Bert
On Jul 1, 2011, at 10:22 AM, David Winsemius wrote:

On Jul 1, 2011, at 10:10 AM, dunner wrote:

Hello all, thanks for your time and patience.

I'm looking for a method in R to analyse the following data:

Time to waking after anaesthetic for medical procedures repeated on the
same
individual.

str(mysurv)
labelled [1:740, 1:2] 20 ?20 ?15 ?20 ?30+ 40+ 50 ?30 ?15 ?10 ?...
- attr(*, "dimnames")=List of 2
..$ : NULL
..$ : chr [1:2] "time" "status"
- attr(*, "type")= chr "right"
- attr(*, "units")= chr "Day"
- attr(*, "time.label")= chr "ORIENTATION"
- attr(*, "event.label")= chr "FullyOrientated"

mysurv is constructed from the following data:

head(data.frame(MRN, ORIENTATION, FullyOrientated))

? ? MRN ORIENTATION FullyOrientated
1 0008291 ? ? ? ? ? 20 ? ? ? ? ? ? ? 2
2 0008469 ? ? ? ? ? 20 ? ? ? ? ? ? ? 2
3 0008469 ? ? ? ? ? 15 ? ? ? ? ? ? ? 2
4 0010188 ? ? ? ? ? 20 ? ? ? ? ? ? ? 2
5 0013664 ? ? ? ? ? 30 ? ? ? ? ? ? ? 1
6 0014217 ? ? ? ? ? 40 ? ? ? ? ? ? ? 1

I had planned to use a Cox PH model to analyse time to waking
(ORIENTATION =
10, 15, 20 mins ....... 50 mins) and whether or not people (MRN) are
fully
awake within an hour (FullyOrientated). I've put ?GENDER, etc. into the
model but I have the following bias:

The procedure is repeated weekly on each individual (MRN), so each
individual has 5-9 cases associated with them. Currently I am including
these in the model as if they were independent.

Is there a way to account for the non-independence of these waking times?

I'm thinking of something similar to the NLMER package and Multilevel /
Mixed Effects analysis as described in Pinheiro and Bates.
Have you looked at the coxme package?
As an initial strata()-gem, as it were, perhaps just adding strata(MRN) may
parcel out the intra-individual variability and degrees of freedom, so that
they are not inappropriately included in the IV's. My initial suggestion of
coxme may be overkill.

--

David Winsemius, MD
West Hartford, CT

______________________________________________
R-help at r-project.org 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.
David Winsemius, MD
West Hartford, CT

______________________________________________
R-help at r-project.org 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.

"Men by nature long to get on to the ultimate truths, and will often
be impatient with elementary studies or fight shy of them. If it were
possible to reach the ultimate truths without the elementary studies
usually prefixed to them, these would not be preparatory studies but
superfluous diversions."

-- Maimonides (1135-1204)

Bert Gunter
Genentech Nonclinical Biostatistics
467-7374
http://pharmadevelopment.roche.com/index/pdb/pdb-functional-groups/pdb-biostatistics/pdb-ncb-home.htm
There is indeed right censoring, but I obviously didn't explain it very well. 

Patients are either fully oriented or not (1 or 2) after an hour. If they're
not, then the data is right censored.

However, I don't feel that "coxme" is overkill at all, as I may also have to
account for repeated COURSES of treatments on the same individiual, so the
data would be  structured as follows: TREATMENTS repeated within COURSES
repeated within PATIENTS.

However, I may need a little help later specifying the model technically in
R.

What would the random effects bit of the formula look like? At   the moment
I have

mycoxme<-coxme(mysurv~SEX + (1|MRN))

for example.

would a correct specification for the nested data be:

mycoxme<-coxme(mysurv~SEX + (1|MRN|COURSE))

Then of course we have the added issue that treatments are ordered.... so is
there a "Frailty" model that can account for that I wonder? 

Thanks for all your help.

Ross

--
View this message in context: http://r.789695.n4.nabble.com/Multilevel-Survival-Analysis-Cox-PH-Model-tp3638278p3640352.html
Sent from the R help mailing list archive at Nabble.com.