aftfit: Instrumental variable estimation via semiparametric...
In jaredhuling/aftiv: Instrumental Variable Estimation for Time-to-Event Data

Description Usage Arguments Examples

View source: R/aftfit.R

Fitting function for instrumental variable estimation under the semiparametric AFT model

aftfit(formula, data, instrument, confounded.x.names, method = c("AFT",
  "AFT-IV", "AFT-2SLS", "AFT-IPCW"), smoothed = FALSE, weights = NULL,
  bootstrap = FALSE, boot.method = c("ls", "sv", "full.bootstrap"),
  B = 1000L, dependent.censoring = FALSE, na.action, init = NULL,
  return.data = FALSE, tol = 1e-05, maxit = 10L, verbose = 0,
  BB.control = NULL, ...)

`formula`	a formula object, with the response on the left of a tilde operator, and the covariates. The response must be a survival object as returned by the `Surv()` function.
`data`	a `data.frame` which contains the variables named in the `formula`.
`instrument`	a vector of length equal to the number of rows in `data` containing the instrumental variable
`confounded.x.names`	name of the exposure variable (endogenous variable) of interest
`method`	one of: `"AFT-IPCW"` The proposed IV method that uses inverse probability of censoring weighting `"AFT"` Plain rank-based AFT method that does not account for unmeasured confounding `"AFT-IV"` biased AFT method that acccounts for unmeasured confounding, but not the non-ignorable censoring (and is hence biased) `"AFT-2SLS"` Two-stage AFT method which accounts for unmeasured confounding but requires correct specification of relationship between IV and exposure
`smoothed`	should a smoothed version of the estimating equation be used? Experimental.
`weights`	observation weights. Not used currently!
`bootstrap`	logical variable indicating whether or not to run bootstrap
`boot.method`	bootstrap method to use. `"ls"` uses the LS method of Zeng and Lin (2008), `"sv"` uses the SV method of Zeng and Lin (2008). `"full.bootstrap"` uses a full boostrap where resampled estimating equations are solved for each iteration (this method is very slow!)
`B`	number of bootstrap iterations
`dependent.censoring`	for the `"AFT-IPCW"` method, should the censoring model allow for dependence on covariates? If `TRUE` a Cox model will be fit for the censoring distribution
`na.action`	how should missing data be treated? See help file for `lm` for more details
`init`	vector equal to the length of the number of parameters in the model used to initialize estimation
`return.data`	should the data be returned?
`tol`	positive tolerance threshold. Smaller values indicate more precise solutions are required
`maxit`	maximum number of restarts for `BB` algorithm
`verbose`	should messages be printed? logical variable
`BB.control`	control list for `BBsolve()` function. See `BBsolve` for more information
`...`	not used

library(aftiv)

set.seed(1)
true.beta <- c(1,-0.5,-0.5,0.25,0,0,-0.75,0.75)
dat <- simIVMultivarSurvivalData(500,1,1,-1,1,true.beta,num.confounded = 1,
                                 confounding.function = "exp")

df <- data.frame(dat$survival[c("delta", "log.t")], dat$X)
Z <- dat$Z

system.time(aftf <- aftfit(Surv(log.t, delta) ~ ., data = df, 
                           instrument = Z, 
                           confounded.x.names = "X1",
                           method = c("AFT", "AFT-IV", "AFT-2SLS", "AFT-IPCW"), 
                           boot.method = "ls", verbose = 1,
                           B = 200L, smoothed = FALSE,
                           bootstrap = TRUE))

summary(aftf)