Overview
Welcome to the {beastt} package! This R package is designed to assist users in performing Bayesian dynamic borrowing with covariate adjustment via inverse probability weighted robust mixture priors for simulations and data analyses in clinical trials. For the sake of this package, we use the term IPW BMB to refer to this inverse probability weighted robust mixture methodology.
What is IPW BDB?
Inverse Probability Weighted Bayesian Dynamic Borrowing (IPW BDB) is a statistical approach designed to enhance the estimation of marginal (i.e., population-averaged or unconditional) treatment effects in clinical trials. This method employs inverse probability weighted robust mixture priors to adjust for covariate differences between a new internal study and external (i.e., historical) control data.
Why use IPW BDB?
By using propensity score-based inverse probability weighting, IPW BDB effectively balances prognostic variables between trial participants and historical controls, improving inference accuracy and reducing biases due to differences in covariate distributions. This technique increases the statistical power and reduces potential biases in estimating average treatment effects, which are critical for health policy decisions and drug approval processes.
IPW BDB has two mechanisms by which it can account for drift from different sources: 1. The use of a robust mixture prior alleviates prior-data conflict by dynamically down weighting external data when there is a significant level of drift between studies. 2. Inverse probability weighting can account for explainable causes of drift by balancing covariate distributions between external and internal control participants.
Augmenting the standard robust mixture prior (RMP) approach to incorporate IPWs does not add substantial computational burden associated with other Bayesian approaches; e.g., in cases where conjugate priors exist for the standard RMP approach, they will still exist for the IPW BDB approach.
When can IPW BDB be used?
IPW BDB should be considered in clinical trial settings where individual level external control data is available and you want to integrate this data with your current trial. This method is particularly useful when there are differences in distributions of key prognostic factors between the current study population and the external controls, which could otherwise introduce bias. It is especially relevant in oncology and rare disease trials, where using external data can help overcome challenges such as slow patient enrollment due to reluctance to join control groups. IPW BDB is well-suited for contexts where Bayesian dynamic borrowing is already applicable but could benefit from additional adjustments for confounding.
Installation
You can install the development version of {beastt} from GitHub with:
# install.packages("devtools")
devtools::install_github("GSK-Biostatistics/beastt")Usage
At the moment {beastt} covers borrowing from external control data for normal, binary, and time to event endpoints. For more information, see the vignettes.
