Skip to contents

decider R package: decision making in multiple-arm oncology dose escalation trials with logistic regression

The decider package was created to allow the use of Bayesian logistic regression models (BLRMs) for decision making in oncology dose escalation trials with multiple arms. It focuses on the use of the meta-analytic (“joint”) BLRM for mono- and combination therapy, which uses a hierarchical prior structure. The package comes with various functions allowing to evaluate the performance and operating characteristics of the BLRM and a chosen prior in a given trial setting using trial simulations or hypothetical data scenarios.

Background and getting started

The main model implemented in the package is the hierarchical (exchangeable) combination therapy BLRM following the model described e.g. by Neuenschwander et al. (2016) for oncology dose finding in combination therapy. The posterior computations for the joint BLRM are implemented using the rstan package. For a detailed model description, refer to the documentation of scenario_jointBLRM(). The deciderpackage allows to use the basic hierarchical combination therapy model with exchangeable prior structure together with two variants of the general model:

  • Inclusion of a binary covariate

  • Use of two different interaction terms following the methodology described
    by OncoBayes2, namely, the linear interaction term and the saturating interaction term.

The core functions are scenario_jointBLRM() and sim_jointBLRM(), for evaluation of hypothetical data scenarios, respectively trial simulations. The latter function allows simulations of up to 6 parallel ongoing trial arms, with arbitrarily many historical trials (mono and/or combination therapy), and provides various options for customization – for instance, cohort size, order of cohort enrolment in different arms, dose escalation rules, and further things can be adjusted. Additionally, the simulations allow for the inclusion of so-called back-fill cohorts, where lower doses are backfilled while escalation is ongoing on higher doses.

Please refer to the vignette Evaluating prior specifications of a joint BLRM for oncology dose finding for an introduction to how the functions of the decider can be used to conduct BLRM simulations with multiple arms or evaluate hypothetical data scenarios.

Installation

The development version can be installed as follows:

if(!requireNamespace("remotes", quietly = TRUE)){
  install.packages("remotes")
}
remotes::install_github("https://github.com/Boehringer-Ingelheim/decider")

Documentation

The package documentation is hosted here.