Skip to contents

bayesmsm

Overview

bayesmsm is an R package that implements the Bayesian marginal structural models to estimate average treatment effect for drawing causal inference with time-varying treatment assignment and confounding with extension to handle informative right-censoring. The Bayesian marginal structural models is a semi-parametric approach and features a two-step estimation process. The first step involves Bayesian parametric estimation of the time-varying treatment assignment models and the second step involves non-parametric Bayesian bootstrap to estimate the average treatment effect between two distinct treatment sequences of interest.

Reference paper on Bayesian marginal structural models:

  • Saarela, O., Stephens, D. A., Moodie, E. E., & Klein, M. B. (2015). On Bayesian estimation of marginal structural models. Biometrics, 71(2), 279-288.

  • Liu, K., Saarela, O., Feldman, B. M., & Pullenayegum, E. (2020). Estimation of causal effects with repeatedly measured outcomes in a Bayesian framework. Statistical methods in medical research, 29(9), 2507-2519.

Installation

Install using devtools package:

## install.packages(devtools) ## make sure to have devtools installed 
devtools::install_github("Kuan-Liu-Lab/bayesmsm")
library(bayesmsm)

Dependency

This package depends on the following R packages:

  • MCMCpack
  • doParallel
  • foreach
  • parallel
  • R2jags
  • coda

Quick Start

Here are some examples demonstrating how to use the bayesmsm package:

# Load example data without right-censoring
testdata <- read.csv(system.file("extdata", "continuous_outcome_data.csv", package = "bayesmsm"))

# Calculate Bayesian weights
weights <- bayesweight(
  trtmodel.list = list(
    a_1 ~ w1 + w2 + L1_1 + L2_1,
    a_2 ~ w1 + w2 + L1_1 + L2_1 + L1_2 + L2_2 + a_1
  ),
  data = testdata,
  n.iter = 250,
  n.burnin = 150,
  n.thin = 5,
  n.chains = 2,
  seed = 890123,
  parallel = TRUE
)

# Perform Bayesian non-parametric bootstrap
model <- bayesmsm(
  ymodel = y ~ a_1 + a_2,
  nvisit = 2,
  reference = c(rep(0, 2)),
  comparator = c(rep(1, 2)),
  family = "gaussian",
  data = testdata,
  wmean = weights$weights,
  nboot = 1000,
  optim_method = "BFGS",
  seed = 890123,
  parallel = TRUE,
  ncore = 2
)

# View model summary
summary.bayesmsm(model)

License

This package is licensed under the MIT License. See the LICENSE file for details.

Citation

Please cite our software using:

@Manual{,
  title = {bayesmsm: An R package for longitudinal causal analysis using Bayesian Marginal Structural Models},
  author = {Xiao Yan and Martin Urner and Kuan Liu},
  year = {2024},
  note = {https://github.com/Kuan-Liu-Lab/bayesmsm},
  url = {https://kuan-liu-lab.github.io/bayesmsm/},
}

Contact