icrsf {icRSF}

Permutation-based variable importance metric for high dimensional datasets appropriate for time to event outcomes, in the presence of imperfect self-reports or laboratory-based diagnostic tests.

Description

Let N and P denote the number of subjects and number of variables in the dataset, respectively. Let N** denote the total number of visits, summed over all subjects in the study [i.e. N** denotes the number of diagnostic test results available for all subjects in the study]. This algorithm builds a user-defined number of survival trees, using bootstrapped datasets. Using the out of bag (OOB) data in each tree, a permutation-based measure of variable importance for each of the P variables is obtained.

Usage

icrsf(data, subject, testtimes, result, sensitivity, specificity, Xmat,
  root.size, ntree, ns, node, pval = 1)

Arguments

data

name of the data frame that includes the variables subject, testtimes, result

subject

vector of subject IDs of length N**x1.

testtimes

vector of visit or test times of length N**x1.

result

vector of binary diagnostic test results (0 = negative for event of interest; 1 = positive for event of interest) of length N**x1.

sensitivity

the sensitivity of the diagnostic test.

specificity

the specificity of the diagnostic test.

Xmat

a N x P matrix of covariates.

root.size

minimum number of subjects in a terminal node.

ntree

number of survival trees.

ns

number of covariate selected at each node to split the tree.

node

For parallel computation, specify the number of nodes.

pval

P-value threshold of the Likelihood Ratio Test.

Value

a vector of the ensembled variable importance for modified random survival forest (icRSF).

Examples

library(parallel)
data(Xmat)
data(pheno)
vimp <- icrsf(data=pheno, subject=ID, testtimes=time, result=result, sensitivity=1,
             specificity=1, Xmat=Xmat, root.size=30, ntree=1, ns=sqrt(ncol(Xmat)), node=1, pval=1)