[R] survival probabilities from survival tree (rpart)

[Luka Kos]

Hi!

>How to compute probabilities of survival from rpart object? Consider the 
>following example:
>
>library(rpart)
>survtree <- rpart(Surv(time,status)~., data=aml)
>
>This code will produce the following survival tree:
>
>  1. root 23 26.44184 1.0000000
>  2. x=Maintained 11 12.61182 0.6848303 *
>  3. x=Nonmaintained 12 10.81338 1.4426070 *
>
>If I am not wrong the rightmost numbers are relative risks compared to 
>whole population (e.g. root >node). How to get survival probability 
>function S(t|x)?

After (almost) one year I am finally able to answer myself "from the future" 
:).

Create rpart tree and for every population from tree's leaf generate 
Kaplan-Meier curve. For new sample use curve which corresponds to 
combination of example's attributes.
For details refer to:
M. Radespiel-Troeger et al. (2003), Comparison of tree-based methods for 
prognostic stratication of survival data, Articial Intelligence in 
Medicine 28 (3): 323--341.

Regards,
Ludvik