[R] variance explained in a cox ph model

Federico Calboli f.calboli at imperial.ac.uk
Wed Feb 22 15:05:17 CET 2012

On 22 Feb 2012, at 14:01, Terry Therneau wrote:

> --- begin included message ---
> I have a left truncated, right censored cox model:
> coxph(Surv(start, stop, censor) ~ x + y, mydata)
> I would like to know how much of the observed variance (as a number
> between 0 and 1) is explained by each variable. How could I do that?
> Adding terms sequentially and then using anova(mod1, mod2) tells me
> whether I get a significant improvement of the fit, but does not tell me
> how much variance I actually explain? 
> --------- end inclusion --------
> There is not a "variance explained" in a Cox model, since it does not
> try to predict the actual survival time.  There are several proposals in
> the literature for an "approximate" R^2.  If you use summary(fit) it
> will print one of the older ideas, which unfortuantely was shown by
> later literature to not be a very good approach (replacing this is
> another item on my long list of "someday fix").  At present your best
> choice is to use the c-statistic, which is also printed out by summary.

I am now actually looking at the Schoenfeld residuals (which are calculated as a matter of fact by residuals()/resid()), and I am looking at 

Muller et al. Quantifying the contribution of genetic variants for survival phenotypes

Genetic Epidemiology 2008



Federico C. F. Calboli
Neuroepidemiology and Ageing Research
Imperial College, St. Mary's Campus
Norfolk Place, London W2 1PG

Tel +44 (0)20 75941602   Fax +44 (0)20 75943193

f.calboli [.a.t] imperial.ac.uk
f.calboli [.a.t] gmail.com

More information about the R-help mailing list