khazardcond {kernhaz}

Kernel estimate of conditional hazard function for right-censored data

Description

Kernel estimate of conditional hazard function for right-censored data with one covariate. Options include two methods for bandwidth selection.

Usage

khazardcond(times, delta, covariate, h = NULL, t = NULL, x = NULL,
  tx = NULL, t.length = 100, x.length = 100, tmin = NULL,
  tmax = NULL, xmin = NULL, xmax = NULL, kernel = "epanechnikov",
  type = "interior", type.w = "nw", parallel = FALSE,
  h.method = "crossval", optim.method = "ga", tol = ifelse(h.method
  == "crossval", 10^(-6), 1), run = 2, ...)

Arguments

times

vector of observed times

delta

vector of censoring indicator. 0 - censored, 1 - uncensored (dead)

covariate

vector of covariate

h

bandwidth vector of length 2, first element is bandwidth for time and second for covariate. If missing, h is found using some bandwidth selection method.

t

vector of time points at which estimate is evaluated

x

vector of covariate points at which estimate is evaluated

tx

data frame of t and x at which estimate is evaluated

t.length

number of grid points of time

x.length

number of grid points of covariate

tmin, tmax

minimum/maximum values for grid of time

xmin, xmax

minimum/maximum values for grid of covariate

kernel

kernel function, possible values are: "epanechnikov" (default), "gaussian", "rectangular", "quartic".

type

Type of kernel estimate. Possible types are: "exterior", "interior" (default).

type.w

Type of weights. Default are Nadaraya-Watson weights.

parallel

allows parallel computation. Default is FALSE.

h.method

method for bandwidth selection. Possible methods are: "crossval" (default), "maxlike".

optim.method

method for numerical optimization of the crossvalidation or log-likelihood function. Possible methods are: "ga"(default).

tol

the desired accuracy of optimization algorithm

run

the number of consecutive generations without any improvement in the best fitness value before the GA is stopped.

...

additional arguments of GA algorithm

Details

External type of kernel estimator is defined as the ratio of kernel estimator of the conditional subdensity of the uncensored observations to the conditional survival function of the observable time. Internal type of kernel estimator is based on a convolution of the kernel function with a nonparametric estimator of the cumulative conditional hazard function.

Value

Returns an object of class 'khazardcond' which is a list with fields

time.points

vector of time points at which estimate is evaluated

covariate.points

vector of covariate points at which estimate is evaluated

hazard

matrix of hazard function values on grid or data.frame of time and covariate points and appropriate hazard values if hx is defined

h

bandwidth vector

CVML

value of crossvalidation or log-likelihood at h

details

description of used methods

GA.result

output of ga, object of class ga-class

References

Selingerova, I., Dolezelova, H., Horova, I., Katina, S., and Zelinka, J. (2016). Survival of Patients with Primary Brain Tumors: Comparison of Two Statistical Approaches. PloS one, 11(2), e0148733.

See Also

plot.khazardcond, ga

Examples

library(survival)
fit<-khazardcond(times = lung$time,delta = lung$status-1,covariate = lung$age,h=c(200,20))