[R] survival::predict.coxph

Bernhard Reinhardt bernhard.reinhardt at dlr.de
Thu Feb 26 13:18:55 CET 2009


I just came across another question concerning predict.coxph

Terry Therneau states in "A Package for Survival Analysis in S" that

term <- predict(fit, type="terms")

yields predicted values for the individual components of the linear 
predictor X*beta

My coxmodel looks like:

        coef exp(coef) se(coef)      z      p
S0   -3.106  4.48e-02     2.88 -1.080 0.2800
S1    6.365  5.81e+02     5.20  1.224 0.2200
S2  -14.009  8.24e-07     5.32 -2.636 0.0084

The first line of my input-data looks like:	
	S0     	 S1      S2      S3
1   -1.030 -0.9500 -1.0950 -1.0700

So I thought the first line of term should be calculated by

-1.030*-3.106,  -0.9500*6.365,  -1.0950*-14.009 [..] which is
3.20,		-6.04,		15.34

Actually the first line of term contains:
           S0          S1           S2
1   3.36737346 -6.36032595  15.73846097

which is quite similar but not the same.

Can anyone shed some light on this? I guess there must be tons of 
literature on this topic but I find it quite hard to find the good one.

I´d also appreciate literature on how to choose the appropriate number 
of covariates for a coxmodel and overfitting.



More information about the R-help mailing list