gcrrBAR: Group Broken Adaptive Ridge Regression for Competing Risks...

In erickawaguchi/crrBAR: Broken Adaptive Ridge for Competing Risks Regression

Description

Fits groupbroken adaptive ridge regression for competing risks regression. Based on the crrp package which performs penalized variable selection using LASSO, SCAD, and MCP. This package allows for ridge and broken adaptive ridge penalties.

Usage

gcrrBAR(ftime, fstatus, X, failcode = 1, cencode = 0,
  group = 1:ncol(X), lambda = 0, xi = 0, delta = 0, eps = 1e-06,
  tol = 1e-06, lam.min = ifelse(dim(X)[1] > dim(X)[2], 0.001, 0.05),
  nlambda = 25, log = TRUE, max.iter = 1000)

Arguments

ftime	A vector of event/censoring times.
fstatus	A vector with unique code for each event type and a separate code for censored observations.
X	A matrix of fixed covariates (nobs x ncovs)
failcode	Integer: code of fstatus that event type of interest (default is 1)
cencode	Integer: code of fstatus that denotes censored observations (default is 0)
group	Vector of group indicators. Must be a vector of consecutive integers.
lambda	Numeric: BAR tuning parameter value
xi	Numeric: tuning parameter for initial ridge regression
delta	Numeric: change from 2 in ridge norm dimension
eps	Numeric: algorithm stops when the relative change in any coefficient is less than eps (default is 1E-6)
tol	Numeric: absolute threshold at which to force coefficients to 0 (default is 1E-6)
lam.min	Numeric: smallest value of lambda if performing grid search
nlambda	Numeric: number of lambda values if performing grid search (default is 25)
log	Logical: Whether or not the grid search is log10 spaced (default is TRUE)
max.iter	Numeric: maximum iterations to achieve convergence (default is 1000)

Details

The crrBAR function penalizes the log-partial likelihood of the proportional subdistribution hazards model from Fine and Gray (1999) with the Broken Adaptive Ridge (BAR) penalty. A cyclic coordinate descent algorithm is used for implementation. For stability, the covariate matrix X is standardized prior to implementation.

Special cases: Fixing xi and lambda to 0 results in the standard competing risk regression using crr. Fixing lambda to 0 and specifying xi will result in a ridge regression solution.

Value

Returns a list of class crrBAR.

References

Breheny, P. and Huang, J. (2011) Coordinate descent algorithms for nonconvex penalized regression, with applications to biological feature selection. Ann. Appl. Statist., 5: 232-253.

Fine J. and Gray R. (1999) A proportional hazards model for the subdistribution of a competing risk. JASA 94:496-509.

Fu Z., Parikh C. and Zhou B. (2017). Penalized variable selection in competing risks regression. Lifetime Data Anal 23:353-376.

Breheny, P. and Huang, J. (2015). Group descent algorithms for nonconvex penalized linear and logistic regression models with grouped predictors. Statistics and Computing 25: 173-187

Examples

set.seed(10)
ftime <- rexp(200)
fstatus <- sample(0:2, 200, replace = TRUE)
cov <- matrix(runif(1000), nrow = 200)
dimnames(cov)[[2]] <- c('x1','x2','x3','x4','x5')
gp <- c(1, 1, 2, 2, 3)
fit <- gcrrBAR(ftime, fstatus, cov, group = gp, lambda = log(sum(fstatus == 1)) / 2, xi = 1 / 2)
fit$coef

erickawaguchi/crrBAR documentation built on June 6, 2019, 7:56 a.m.