msm_vimshift {txshift}

Working marginal structural model for causal effects of an intervention grid

Description

Working marginal structural model for causal effects of an intervention grid

Usage

msm_vimshift(
  W,
  A,
  C_cens = rep(1, length(Y)),
  Y,
  C_samp = rep(1, length(Y)),
  V = NULL,
  delta_grid = seq(-0.5, 0.5, 0.5),
  msm_form = list(type = "linear", knot = NA),
  estimator = c("tmle", "onestep"),
  weighting = c("identity", "variance"),
  ci_level = 0.95,
  ci_type = c("marginal", "simultaneous"),
  ...
)

Arguments

W

A matrix, data.frame, or similar containing a set of baseline covariates.

A

A numeric vector corresponding to a treatment variable. The parameter of interest is defined as a location shift of this quantity.

C_cens

A numeric indicator for whether a given observation was subject to censoring by way of loss to follow-up. The default assumes no censoring due to loss to follow-up.

Y

A numeric vector of the observed outcomes.

C_samp

A numeric indicator for whether a given observation was subject to censoring by being omitted from the second-stage sample, used to compute an inverse probability of censoring weighted estimator in such cases. The default assumes no censoring due to two-phase sampling.

V

The covariates that are used in determining the sampling procedure that gives rise to censoring. The default is NULL and corresponds to scenarios in which there is no censoring (in which case all values in the preceding argument C must be uniquely 1. To specify this, pass in a NAMED list identifying variables amongst W, A, Y that are thought to have played a role in defining the sampling/censoring mechanism (C).

delta_grid

A numeric vector giving the individual values of the shift parameter used in computing each of the estimates.

msm_form

A list specifying the type of working MSM to fit to summarize the counterfactual means. The list has two components: (1) "type", which may be either "linear" or "piecewise", and (2) "knot", which, if specified, must be a value in delta_grid. See examples for its use.

estimator

The type of estimator to be fit, either "tmle" for targeted maximum likelihood estimation or "onestep" for a one-step augmented inverse probability weighted (AIPW) estimator.

weighting

Whether to weight each parameter estimate by the inverse of its variance (in order to improve stability of the resultant MSM fit) or to simply weight all parameter estimates equally. The default is the option "identity", weighting all estimates identically.

ci_level

A numeric indicating the desired coverage level of the confidence interval to be computed.

ci_type

Whether to construct a simultaneous confidence band covering all parameter estimates at once or marginal confidence intervals covering each parameter estimate separately. The default is to construct marginal confidence intervals for each parameter estimate rather than a simultaneous confidence band.

...

Additional arguments to be passed to txshift.

Details

Computes estimates of the counterfactual mean over a grid of shift stochastic interventions and fits a working marginal structural model to summarize the trend through the counterfactual means as a function of the specified shift intervention. The working marginal structural model may be linear in the shift parameter or piecewise linear with a single knot point. Provides support for two weighting schemes, may be used with either of the one-step or TML estimators, and also allows the construction of marginal or simultaneous confidence intervals.

Value

A list containing estimates of the individual counterfactual means over a grid in the shift parameters (delta_grid), alongside the estimate of a marginal structural model that summarizes a trend through these counterfactual means.

Examples

if (require("sl3")) {
  n_obs <- 100
  W <- as.numeric(replicate(1, rbinom(n_obs, 1, 0.5)))
  A <- as.numeric(rnorm(n_obs, mean = 2 * W, sd = 1))
  Y <- rbinom(n_obs, 1, plogis(2 * A - W))
  msm <- msm_vimshift(
    W = W, A = A, Y = Y, estimator = "tmle",
    g_exp_fit_args = list(
      fit_type = "sl",
      sl_learners_density = Lrnr_density_hse$new(Lrnr_glm$new())
    ),
    Q_fit_args = list(
      fit_type = "glm",
      glm_formula = "Y ~ ."
    ),
    delta_grid = seq(-1, 1, 0.25)
  )

  # fit a linear spline with knot at 0
  n_obs <- 100
  W <- as.numeric(replicate(1, rbinom(n_obs, 1, 0.5)))
  A <- as.numeric(rnorm(n_obs, mean = 2 * W, sd = 1))
  Y <- rbinom(n_obs, 1, plogis(0.1 * A * (A >= 0) - 3 * A * (A < 0) - W))
  msm <- msm_vimshift(
    W = W, A = A, Y = Y, estimator = "tmle",
    g_exp_fit_args = list(
      fit_type = "sl",
      sl_learners_density = Lrnr_density_hse$new(Lrnr_glm$new())
    ),
    Q_fit_args = list(
      fit_type = "glm",
      glm_formula = "Y ~ ."
    ),
    delta_grid = seq(-1, 1, 0.25),
    msm_form = list(type = "piecewise", knot = 0)
  )
}