| ah.2ph {addhazard} | R Documentation |
Fit Additive Hazards Regression Models to Two-phase Sampling
Description
The function fits a semiparametric additive hazards model
\lambda(t|Z=z) = \lambda_0(t) + \beta'z.
to two-phase sampling data. The estimating procedures follow Hu (2014).
Usage
ah.2ph(formula, data, R, Pi, ties, robust = FALSE,
calibration.variables = NULL, ...)
Arguments
formula |
a formula object for the regression model of the form
response ~ predictors. The outcome is a survival object created by |
data |
a data frame. Input dataset. |
R |
a phase II membership indicator. A vector of values of 0 and 1. The subject is selected to phase II if R = 1. |
Pi |
the probability of a subject to be selected to the phase II subsample. |
ties |
a logical variable. FALSE if there are no ties in the censored failure times. |
robust |
a logical variable. Robust standard errors are provided if robust = TRUE. |
calibration.variables |
a vector of some column names of the data. These are the variables available for every observation. They are used to calibrate the weight assigned to each subject in order to improve estimation efficiency. |
... |
additional arguments to be passed to the low level regression fitting functions. |
Value
An object of class 'ah.2h' representing the fit.
Note
This function estimates both model-based and robust standard errors. It can be used to analyze case-cohort studies. It allows subsampling among cases. It can incorporate the calibration procedure and analyze the combined dataset of phase I and phase II samples.
References
Jie Hu (2014) A Z-estimation System for Two-phase Sampling with Applications to Additive Hazards Models and Epidemiologic Studies. Dissertation, University of Washington.
See Also
predict.ah.2ph for prediction based on fitted additive
hazards model with two-phase sampling and nwtsco for the description
of nwtsco dataset.
Examples
library(survival)
### fit an additive hazards model to two-phase sampling data without calibration
nwts2ph$trel <- nwts2ph$trel + runif(dim(nwts2ph)[1],0,1)*1e-8
fit1 <- ah.2ph(Surv(trel,relaps) ~ age + histol, ties = FALSE, data = nwts2ph, R = in.ph2, Pi = Pi,
robust = FALSE, calibration.variables = NULL)
summary(fit1)
### fit an additve hazards model with calibration on age
fit2 <- ah.2ph(Surv(trel,relaps) ~ age + histol, ties = FALSE, data = nwts2ph, R = in.ph2, Pi = Pi,
robust = FALSE, calibration.variables = "age")
summary(fit2)
### calibrate on age square
### note if users create a calibration variable, then
### the new variable should be added to the original data frame
nwts2ph$age2 <- nwts2ph$age^2
fit3 <- ah.2ph(Surv(trel,relaps) ~ age + histol, ties = FALSE, data = nwts2ph, R = in.ph2, Pi = Pi,
robust = FALSE, calibration.variables = "age2")
summary(fit3)
#############################################################################
### When phase II samples are obtained by finite Sampling #############
#############################################################################
### calculating the sample size for each straum
### calculate the strata size
strt.size <- table(nwts2ph$strt)
ph2.strt.size <- table(subset(nwts2ph, in.ph2 == 1)$strt)
### fit an additve hazards model with finite stratified sampling
### calculate the sampling fractions
frac <- ph2.strt.size/strt.size
### treating the problem as bernoulli sampling coupled with calibration on strata sizes
### using frac as the sampling probilities
nwts2ph_by_FPSS <- nwts2ph
nwts2ph_by_FPSS$Pi <- NULL
for (i in 1:length(strt.size)){
nwts2ph_by_FPSS$Pi[nwts2ph_by_FPSS$strt ==i] <- frac[i]
}
### create strt indicators, which become our calibration variables
for (i in 1:length(strt.size)){
nwts2ph_by_FPSS$strt_ind <- as.numeric(nwts2ph_by_FPSS$strt ==i)
names(nwts2ph_by_FPSS)[ncol(nwts2ph_by_FPSS)]= paste0("strt", i)
}
### fit an additve hazards model with finate sampling
fit4 <- ah.2ph(Surv(trel,relaps) ~ age + histol,
data = nwts2ph_by_FPSS, ties = FALSE,
R = in.ph2, Pi = Pi,
robust = FALSE,
calibration.variables = c("strt1","strt2","strt3"))
summary(fit4)