[R] Cox model using mfp library

Inman, Brant A. M.D. Inman.Brant at mayo.edu
Wed Nov 8 22:47:33 CET 2006

I am running R 2.3.1 on a Windows XP machine.  I have a large dataset of
over 13 000 cases of a disease for which I am attempting to build a
prognostic model using Cox proportional hazards regression.  Some of the
continuous covariates are skewed and therefore require transformation
for use in the Cox model.  I have empirically selected some simple
normalizing transformations (based on previous knowledge and the ladder
of powers) and wanted to compare these selections to those produced by
the fractional polynomial technique of Royston and Altman.

When I attempt to fit a Cox model with the mfp library for overall
survival using a single left-skewed covariate called "age", I get the
following output:

> library(mfp)
> fit<- mfp(Surv(time.dead,dead.any)~ fp(age), family=cox, verbose=T,
+ na.action=na.exclude)

        Variable        Deviance        Power(s)
Cycle 1
                        26704.42        1
                        26701.83        2
                        26695.2 -2 -2

    shift scale
age     0   100

Fractional polynomials
    df.initial select alpha df.final power1 power2
age          4      1  0.05        4     -2     -2

Null model: 27071.72
Linear model: 26704.42
Final model: 26695.2
Error in "[.data.frame"(list("Surv(time.dead, dead.any)" = c(NA, NA, NA,
        invalid subscript type


Can anyone tell me what this error message means and why I am getting
it?  I did not get this error when I tried similar code using the GBSG
dataset provided with the mfp library.

Brant Inman
Mayo Clinic

More information about the R-help mailing list