ateRobust: Average Treatment Effects (ATE) for survival outcome (with...
In bozenne/riskRegressionLight: Light version of the package riskRegression

Description Usage Arguments Details See Also Examples

Compute the average treatment effect using different methods: G-formula based on (cause-specific) Cox regression, inverse probability of treatment weighting (IPTW) combined with inverse probability of censoring weighting (IPCW), augmented inverse probability weighting (AIPTW, AIPCW).

1
2
3

ateRobust(data, times, cause, type, formula.event, formula.censor,
  formula.treatment, fitter = "coxph", product.limit = NULL,
  se = TRUE, augment.cens = TRUE, na.rm = FALSE)

`data`	[data.frame or data.table] Data set in which to evaluate the ATE.
`times`	[numeric] Time point at which to evaluate average treatment effects.
`cause`	[numeric/character] The cause of interest. Defaults to the first cause.
`type`	[character] When set to `"survival"` uses a cox model for modeling the survival, otherwise when set to `"competing.risks"` uses a Cause Specific Cox model for modeling the absolute risk of the event.
`formula.event`	[formula] Cox model for the event of interest (outcome model). Typically `Surv(time,event)~treatment`.
`formula.censor`	[formula] Cox model for the censoring (censoring model). Typically `Surv(time,event==0)~treatment`.
`formula.treatment`	[formula] Logistic regression for the treatment (propensity score model). Typically `treatment~1`.
`fitter`	[character] Routine to fit the Cox regression models. If `coxph` use `survival::coxph` else use `rms::cph`.
`product.limit`	[logical] If `TRUE` the survival is computed using the product limit method. Otherwise the exponential approximation is used (i.e. exp(-cumulative hazard)).
`se`	[logical] If `TRUE` compute and add the standard errors relative to the G-formula and IPTW method to the output.
`augment.cens`	[logical] If `TRUE` add an censoring model augmentation term to the estimating equation
`na.rm`	[logical] If `TRUE` ignore observations whose influence function is NA.

The standard errors/confindence intervals/p-values output by ateRobust do not account for the uncertainty related to the estimation of the parameters of the censoring model (only relevant for IPCW/AIPCW estimators). Note that for the AIPTW, this uncertainty is neglectable (i.e. o_p(n^-1/2)) in correctly specified models.

ate for the g-formula result in case of more than 2 treatments

library(survival)
library(lava)
library(data.table)
library(prodlim)
set.seed(10)
# survival outcome, binary treatment X1 

ds <- sampleData(101,outcome="survival")
out <- ateRobust(data = ds, type = "survival",
         formula.event = Surv(time, event) ~ X1+X6,
         formula.censor = Surv(time, event==0) ~ X6,
         formula.treatment = X1 ~ X6+X2+X7, times = 1)
out
dt.out=as.data.table(out)
dt.out

# competing risk outcome, binary treatment X1 
dc=sampleData(101,outcome="competing.risks")
x=ateRobust(data = dc, type = "competing.risks",
          formula.event = list(Hist(time, event) ~ X1+X6,Hist(time, event) ~ X6),
         formula.censor = Surv(time, event==0) ~ X6,
         formula.treatment = X1 ~ X6+X2+X7, times = 1,cause=1,
                     product.limit = FALSE)
## compare with g-formula 
fit= CSC(list(Hist(time, event) ~ X1+X6,Hist(time, event) ~ X6),data=dc)
ate(fit,data = dc,treatment="X1",times=1,cause=1)
x
as.data.table(x)