The Proportional Subdistribution Hazard (PSH) model has been popular for estimating the effects of the covariates on the cause of interest in Competing Risks analysis. The fast accumulation of large scale datasets has posed a challenge to classical statistical methods. Current penalized variable selection methods show unsatisfactory performance in ultra-high dimensional data. We propose a novel method, the Random Approximate Elastic Net (RAEN), with a robust and generalized solution to the variable selection problem for the PSH model. Our method shows improved sensitivity for variable selection compared with current methods.
RAEN can be installed from R-CRAN
install.packages('RAEN')Users can install the developmental version from Github.
library(devtools)
install_github('saintland/RAEN')The simulated data toydata contains 200 rows, time to event, censoring
status, and 1000 predictors, 60 of which are true predictors ((X1-X20)
and (X40-X80)). The variable correlation blocks are identified as the
following example.
require(RAEN,quietly = T)## Loaded lars 1.2
data(toydata)
x=toydata[,-c(1:2)]
y=toydata[,1:2]
fgrp<-deCorr(x)##
## No: 1 cluster contains : 19 , remaining 981
##
## No: 2 cluster contains : 17 , remaining 964
##
## No: 3 cluster contains : 19 , remaining 945
##
## No: 4 cluster contains : 18 , remaining 927
##
## No: 5 cluster contains : 2 , remaining 925
The variable selection is executed via the functions grpselect and
r2select. Users can call the main function RAEN to run the whole
process,
library(RAEN)
myres<-RAEN(x,y, B = 50, ncore=3)where x is the nxp predictor matrix, y is the time and
censoring status data frame, and ncore is the number of threads to use
for parallel processing. The selected variables and the regression
coefficients are returned.
id coef
x1 -0.21695698
x2 -0.41486949
x3 -0.21307438
x4 -0.05911834
...
...
x75 -0.3823193
x76 -0.6026619
x77 -0.2440929
x78 -0.3976471
x80 -0.3249638