VarPro for Survival Problems

Min Lu ()

Aster Shear ()

Udaya B. Kogalur ()

Hemant Ishwaran ()

2025-04-10

Default call 

## ------------------------------------------------------------
## survival example: pbc 
## ------------------------------------------------------------
data(pbc, package = "randomForestSRC")
o <- varpro(Surv(days, status)~., pbc)
print(importance(o))

Survival with RMST (restricted mean survival time) 

## ------------------------------------------------------------
## pbc survival with rmst (restricted mean survival time)
## functional of interest is RMST at 500 days
## ------------------------------------------------------------
data(pbc, package = "randomForestSRC")
o <- varpro(Surv(days, status)~., pbc, rmst = 500)
print(importance(o))

Survival with a RMST vector 

## ------------------------------------------------------------
## pbc survival with rmst vector
## variable importance is a list for each rmst value
## ------------------------------------------------------------
data(pbc, package = "randomForestSRC")
o <- varpro(Surv(days, status)~., pbc, rmst = c(500, 1000))
print(importance(o))

High dimensional survival data

Our algorithm automatically implement sparse option for high dimensional problems.

data(vdv, package = "randomForestSRC")
o <- varpro(Surv(Time, Censoring)~., vdv)
print(importance(o))

To turn off the sparse option, you can use the argument sparse = FALSE:

## ------------------------------------------------------------
## high dimensional survival example without sparse option
## ------------------------------------------------------------
data(vdv, package = "randomForestSRC")
o <- varpro(Surv(Time, Censoring)~., vdv, sparse = FALSE)
print(importance(o))

We can use different split-weight methods:

## ----------------------------------------------------------------------
## high dimensional survival example using different split-weight methods
## ----------------------------------------------------------------------
data(vdv, package = "randomForestSRC")
f <- as.formula(Surv(Time, Censoring)~.)

## lasso only
print(importance(varpro(f, vdv, split.weight.method = "lasso")))

## lasso and vimp
print(importance(varpro(f, vdv, split.weight.method = "lasso vimp")))

## lasso, vimp and shallow trees
print(importance(varpro(f, vdv, split.weight.method = "lasso vimp tree")))

External outcome for interpreting external models 

data(pbc, package = "randomForestSRC")
o <- varpro(Surv(days, status)~., pbc)
print(importance(o))

## using  external rsf predictor as an example
print(importance(o,y.external=randomForestSRC::rfsrc(Surv(days, status)~., pbc)$predicted.oob))


Cite this vignette as
M. Lu, A. Shear, U. B. Kogalur, and H. Ishwaran. 2025. “varPro: variable selection for survival problems vignette.” http://www.varprotools.org/articles/survival.html. 

@misc{HemantInstall,
    author = "Min Lu and Aster Shear and Udaya B. Kogalur and Hemant Ishwaran",
    title = {{varPro}: variable selection for survival problems vignette},
    year = {2024},
    url = {http://www.varprotools.org/articles/survival.html},
    howpublished = "\url{http://www.varprotools.org/articles/survival.html}",
    note = "[accessed date]"
}