Re: [R] plotting spline term when frailty term is included

[Therneau, Terry M., Ph.D.]

On 03/02/2016 05:00 AM, r-help-requ...@r-project.org wrote:

I'd very much appreciate your help in resolving a problem that I'm having with 
plotting a spline term.

I have a Cox PH model including a smoothing spline and a frailty term as 
follows:

fit<-coxph(Surv(start,end,exit) ~ x + pspline(z) + frailty(a))

When I run a model without a frailty term, I use the following in order to plot 
the splines:

termplot(fit, term = 2, se = TRUE, ylab = "Log Hazard", rug=TRUE, xlab = 
"z_name")

However, when the frailty term is included, it gives this error:

Error in pred[, first] : subscript out of bounds

What am I doing wrong here? Or is it the case that termplot does not work when 
splines and frailty are included?


There are 3 parts to the answer.
 1. The first is a warning: wrt to mixed effects Cox models, I shifted my energy to coxme 
over 10 years ago.  The "penalized add on to coxph" approach of the frailty function was 
an ok first pass, but is just too limited for any but the simplest models.  I'm unlikely 
fix issues, since there are others much higher on my priority list.


 2. As Dave W. pointed out, the key issue is that predict(type='terms') does not work 
with for a model with two penalized terms, when one is frailty and the other pspline. 
Termplot depends on predict.


 3. Again, as Dave W pointed out, the whole issue of what the "correct" answer would be 
gets much more complicated when one adds random effects to the mix; some things have not 
done because it is not clear where to go.  (Survival curves for a mixed effects model only 
recently got added to my "todo" list, even though it has been on the wish list forever, 
because I finally have a notion of what a good approach would be.)


In your case I'd advise an end run: fit the model using ns() instead of pspline.  I like 
smoothing splines better than regression splines, but the fact is that for most data sets 
they result in nearly identical answers.


Terry T

__
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
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.

Re: [R] plotting spline term when frailty term is included

[David Winsemius]

> On Mar 1, 2016, at 2:59 AM, Conor Edward Little  wrote:
> 
> Dear colleagues,
> 
> I'd very much appreciate your help in resolving a problem that I'm having 
> with plotting a spline term.
> 
> I have a Cox PH model including a smoothing spline and a frailty term as 
> follows:
> 
>   fit<-coxph(Surv(start,end,exit) ~ x + pspline(z) + frailty(a))
> 
> When I run a model without a frailty term, I use the following in order to 
> plot the splines:
> 
>   termplot(fit, term = 2, se = TRUE, ylab = "Log Hazard", rug=TRUE, xlab = 
> "z_name")
> 
> However, when the frailty term is included, it gives this error:
> 
>   Error in pred[, first] : subscript out of bounds

I think this is due to the inability of `predict` to cope with the frailty 
model when type="terms". Using the example on the ?frailty page as teh basis 
for a reproducible example:

> predict(coxph(Surv(time, status) ~ age + frailty(inst, df=4), lung), 
> type="terms")
Error in predict.coxph.penal(coxph(Surv(time, status) ~ age + frailty(inst,  : 
  length of 'dimnames' [2] not equal to array extent

> 

> What am I doing wrong here? Or is it the case that termplot does not work 
> when splines and frailty are included?
> 
> A similar question was asked a number of years ago - as far as I can see it 
> didn't receive an answer (online, at least).
> http://r.789695.n4.nabble.com/a-question-in-coxph-td908035.html


The `?termplot` help-page says that there must be a `predict` method for the 
object that accepts a "terms"- argument. Therneau advises the use of the 
'coxme'-package for mixed effects survival modeling and the `predict.coxme` 
function does not have a "terms" argument. 

Prediction with frailty: I think that the issue of prediction with frailty 
models may be more complex than you understand. My level of understanding is 
inadequate as well, so I tried searching for earlier answers. If Thomas Lumley 
says it is "hard" then I take him at my word. Appears you would need to do 
calculation separately for each value of the frailty "term" if you stay with 
the coxph/frailty environment. Therneau has written:  "Residuals methods for 
coxme would be an
important addition and is on my to-do list. (But as my wife would pointout, so 
is a bathroom remodel and she isn't holding her breath.)".  So it appears that 
the problem is more than just a simple two line bit of code.

Lumley's approach from 10+ years ago: 

http://markmail.org/search/?q=list%3Aorg.r-project.r-help+coxph+frailty+prediction+#query:list%3Aorg.r-project.r-help%20coxph%20frailty%20prediction%20+page:1+mid:qy7nxom3uknj2oop+state:results

http://markmail.org/search/?q=list%3Aorg.r-project.r-help+coxph+frailty+prediction+#query:list%3Aorg.r-project.r-help%20coxph%20frailty%20prediction%20+page:1+mid:nbcdjwfs3jhqjsji+state:results


> 
> Best,
> Conor Little
> 
> 
> 
> Conor Little
> Postdoctoral Researcher
> Department of Political Science, University of Copenhagen
> 
> 
>   [[alternative HTML version deleted]]
> 
> __
> R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
> 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
Alameda, CA, USA

__
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
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.

[R] plotting spline term when frailty term is included

[Conor Edward Little]

Dear colleagues,

I'd very much appreciate your help in resolving a problem that I'm having with 
plotting a spline term.

I have a Cox PH model including a smoothing spline and a frailty term as 
follows:

   fit<-coxph(Surv(start,end,exit) ~ x + pspline(z) + frailty(a))

When I run a model without a frailty term, I use the following in order to plot 
the splines:

   termplot(fit, term = 2, se = TRUE, ylab = "Log Hazard", rug=TRUE, xlab = 
"z_name")

However, when the frailty term is included, it gives this error:

   Error in pred[, first] : subscript out of bounds

What am I doing wrong here? Or is it the case that termplot does not work when 
splines and frailty are included?

A similar question was asked a number of years ago - as far as I can see it 
didn't receive an answer (online, at least).
http://r.789695.n4.nabble.com/a-question-in-coxph-td908035.html

Best,
Conor Little



Conor Little
Postdoctoral Researcher
Department of Political Science, University of Copenhagen


[[alternative HTML version deleted]]

__
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
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.