predict.CoxBoost: Predict method for CoxBoost fits

In CoxBoost: Cox models by likelihood based boosting for a single survival endpoint or competing risks

Description

Obtains predictions at specified boosting steps from a CoxBoost object fitted by CoxBoost.

Usage

## S3 method for class 'CoxBoost'
predict(object,newdata=NULL,newtime=NULL,newstatus=NULL,
        subset=NULL,at.step=NULL,times=NULL,
        type=c("lp","logplik","risk","CIF"),...)

Arguments

object	fitted CoxBoost object from a CoxBoost call.
newdata	n.new * p matrix with new covariate values. If just prediction for the training data is wanted, it can be omitted.
newtime, newstatus	vectors with observed time and censoring indicator (0 for censoring, 1 for no censoring, and any other values for competing events in a competing risks setting) for new observations, where prediction is wanted. Only required if predicted partial log-likelihood is wanted, i.e., if type="logplik". This can also be omitted when prediction is only wanted for the training data, i.e., newdata=NULL.
subset	an optional vector specifying a subset of observations to be used for evaluation.
at.step	scalar or vector of boosting step(s) at which prediction is wanted. If type="risk" is used, only one step is admissible. If no step is given, the final boosting step is used.
times	vector with T time points where prediction is wanted. Only needed for type="risk"
type	type of prediction to be returned: "lp" gives the linear predictor, "logplik" the partial log-likelihood, "risk" the predicted probability of not yet having had the event at the time points given in times, and "CIF" the predicted cumulative incidence function, i.e., the predicted probability of having had the event of interest.
...	miscellaneous arguments, none of which is used at the moment.

Value

For type="lp" and type="logplik" a vector of length n.new (at.step being a scalar) or a n.new * length(at.step) matrix (at.step being a vector) with predictions is returned. For type="risk" or type="CIF" a n.new * T matrix with predicted probabilities at the specific time points is returned.

Author(s)

Harald Binder [e-mail]

Examples

#   Generate some survival data with 10 informative covariates 
n <- 200; p <- 100
beta <- c(rep(1,10),rep(0,p-10))
x <- matrix(rnorm(n*p),n,p)
real.time <- -(log(runif(n)))/(10*exp(drop(x %*% beta)))
cens.time <- rexp(n,rate=1/10)
status <- ifelse(real.time <= cens.time,1,0)
obs.time <- ifelse(real.time <= cens.time,real.time,cens.time)

#   define training and test set

train.index <- 1:100
test.index <- 101:200

#   Fit CoxBoost to the training data

cbfit <- CoxBoost(time=obs.time[train.index],status=status[train.index],
                  x=x[train.index,],stepno=300,penalty=100) 

#   mean partial log-likelihood for test set in every boosting step

step.logplik <- predict(cbfit,newdata=x[test.index,],
                        newtime=obs.time[test.index],
                        newstatus=status[test.index],
                        at.step=0:300,type="logplik")

plot(step.logplik)

#   names of covariates with non-zero coefficients at boosting step
#   with maximal test set partial log-likelihood

print(cbfit$xnames[cbfit$coefficients[which.max(step.logplik),] != 0])

CoxBoost documentation built on May 1, 2019, 9:32 p.m.