uniCox: Function to fit a high dimensional Cox survival model using...

In uniCox: Univarate shrinkage prediction in the Cox model

Description

Function to fit a high dimensional Cox survival model using Univariate Shrinkage

Usage

uniCox(x,y,status,lamlist=NULL,nlam=20,del.thres=.01, max.iter=5)

Arguments

x	Feature matrix, n obs by p variables
y	Vector of n survival times
status	Vector of n censoring indicators (1= died or event occurred,0=survived, or event was censored)
lamlist	Optional vector of lambda values for solution path
nlam	Number of lambda values to consider
del.thres	Convergence threshold
max.iter	Maximum number of iterations for each lambda

Details

This function builds a prediction model for survival data with high-dimensional covariates, using the Unvariate Shringae method.

Value

A list with components

lamlist	Values of lambda used
beta	Coef estimates, number of features by number of lambda values
mx	Mean of feature columns
vx	Square root of Fisher information for each feature
s0	Exchangeability factor for denominator of score statistic
call	Call to this function

Source

Tibshirani, R. Univariate shrinkage in the Cox model for high dimensional data (2009). http://www-stat.stanford.edu/~tibs/ftp/cus.pdf To appear SAGMB.

Examples

library(survival)
# generate some data
x=matrix(rnorm(200*1000),ncol=1000)
y=abs(rnorm(200))
x[y>median(y),1:50]=x[y>median(y),1:50]+3
status=sample(c(0,1),size=200,replace=TRUE)

xtest=matrix(rnorm(50*1000),ncol=1000)
ytest=abs(rnorm(50))
xtest[ytest>median(ytest),1:50]=xtest[ytest>median(ytest),1:50]+3

statustest=sample(c(0,1),size=50,replace=TRUE)


# fit uniCox  model
a=uniCox(x,y,status)

# look at results
 print(a)

# do cross-validation to examine choice of lambda
aa=uniCoxCV(a,x,y,status)

# look at results
 print(aa)

# get predictions on a test set
yhat=predict.uniCox(a,xtest)

# fit survival model to predicted values
coxph(Surv(ytest,statustest)~yhat[,7])

uniCox documentation built on May 29, 2017, 2:19 p.m.