surv_onestep: One-step TMLE estimator for survival curve
In wilsoncai1992/onestep.survival: One-step TMLE for survival analysis

Description Usage Arguments Details Value Examples

one-step TMLE estimate of the treatment specific survival curve. Under right-censored data

surv_onestep(dat, dW = rep(1, nrow(dat)), g.SL.Lib = c("SL.glm", "SL.step",
  "SL.glm.interaction"), Delta.SL.Lib = c("SL.mean", "SL.glm", "SL.gam",
  "SL.earth"), ht.SL.Lib = c("SL.mean", "SL.glm", "SL.gam", "SL.earth"),
  epsilon.step = 1e-05, max.iter = 1000, tol = 1/nrow(dat),
  T.cutoff = NULL, verbose = FALSE, ...)

`dat`	A data.frame with columns T.tilde, delta, A, W. T.tilde = min(T, C) is either the failure time of censor time, whichever happens first. 'delta'= I(T <= C) is the indicator of whether we observe failure time. A is binary treatment. W is baseline covariates. All columns with character "W" will be treated as baseline covariates.
`dW`	A binary vector specifying dynamic treatment (as a function output of W)
`g.SL.Lib`	A vector of string. SuperLearner library for fitting treatment regression
`Delta.SL.Lib`	A vector of string. SuperLearner library for fitting censoring regression
`ht.SL.Lib`	A vector of string. SuperLearner library for fitting conditional hazard regression
`epsilon.step`	numeric. step size for one-step recursion
`max.iter`	integer. maximal number of recursion for one-step
`tol`	numeric. tolerance for optimization
`T.cutoff`	int. Enforce randomized right-censoring to the observed data, so that don't estimate survival curve beyond a time point. Useful when time horizon is long.
`verbose`	boolean. When TRUE, plot the initial fit curve, and output the objective function value during optimzation
`...`	additional options for plotting initial fit curve

options to ADD: SL.formula: the covariates to include in SL

Psi.hat A numeric vector of estimated treatment-specific survival curve

T.uniq A vector of descrete time points where Psi.hat take values (have same length as Psi.hat)

params A list of estimation parameters set by user

variables A list of data summary statistics

initial_fit A list of initial fit values (hazard, g_1, Delta)

library(simcausal)
D <- DAG.empty()

D <- D +
    node("W", distr = "rbinom", size = 1, prob = .5) +
    node("A", distr = "rbinom", size = 1, prob = .15 + .5*W) +
    node("Trexp", distr = "rexp", rate = 1 + .5*W - .5*A) +
    node("Cweib", distr = "rweibull", shape = .7 - .2*W, scale = 1) +
    node("T", distr = "rconst", const = round(Trexp*100,0)) +
    node("C", distr = "rconst", const = round(Cweib*100, 0)) +
    node("T.tilde", distr = "rconst", const = ifelse(T <= C , T, C)) +
    node("delta", distr = "rconst", const = ifelse(T <= C , 1, 0))
setD <- set.DAG(D)

dat <- sim(setD, n=3e2)

library(dplyr)
# only grab ID, W's, A, T.tilde, Delta
Wname <- grep('W', colnames(dat), value = TRUE)
dat <- dat[,c('ID', Wname, 'A', "T.tilde", "delta")]

dW <- rep(1, nrow(dat))
onestepfit <- surv_onestep(dat = dat,
                            dW = dW,
                            verbose = FALSE,
                            epsilon.step = 1e-3,
                            max.iter = 1e3)