Parametric G-Formula
Implements the non-iterative conditional expectation (NICE)
algorithm of the g-formula algorithm (Robins (1986)
, Hernán and Robins (2024, ISBN:9781420076165)).
The g-formula can estimate an outcome's counterfactual mean or risk under
hypothetical treatment strategies (interventions) when there is sufficient
information on time-varying treatments and confounders.
This package can be used for discrete or continuous time-varying treatments
and for failure time outcomes or continuous/binary end of follow-up
outcomes. The package can handle a random measurement/visit process and a
priori knowledge of the data structure, as well as censoring (e.g., by loss
to follow-up) and two options for handling competing events for failure time
outcomes. Interventions can be flexibly specified, both as interventions on
a single treatment or as joint interventions on multiple treatments.
See McGrath et al. (2020) for a guide on
how to use the package.
gfoRmula: Parametric G-Formula
Installation
You can install the released version of gfoRmula from CRAN with:
install.packages("gfoRmula")
After installing the devtools package (i.e., calling
install.packages(devtools)), the development version of gfoRmula can
be installed from GitHub with:
devtools::install_github("CausalInference/gfoRmula")
Usage
Please refer to McGrath et
al. (2020) for a detailed
guide to the gfoRmula package. Also, see the following vignettes
regarding updates since McGrath et al. (2020):
- “A Simplified Approach for Specifying Interventions in gfoRmula”
- “Using Custom Outcome Models in gfoRmula”