Skip to contents

Characterizing the robustness of policy recommendations to uncertain assumptions is a crucial concern of modelers. While the computational cost of uncertainty analyses can be high, High-Performance Computing (HPC) is increasingly ubiquitous and accessible. Nevertheless, using HPC tools involves a steep learning curve, which may hinder their adoption.

This repository houses crcrdm – an R package designed to facilitate the analysis of Cancer Screening Models using HPC resources. crcrdm provides an interface between your model and a parallel back-end - an HPC workflow system that orchestrates concurrent model runs. Currently, we use this package with EMEWS. This package aims to make it easier to define and efficiently run large experimental designs, reducing the need to create ad-hoc analytical code for each analysis.

The package includes features useful in ongoing analyses performed with the CRC-SPIN model. First, it partitions the memory usage of models to a manageable size (e.g., the same model run is parallelized across different computing processes and nodes). The package also supports multi-model experimental designs. This package may also help facilitate modeling transparency by separating the analytical workflow from the model.

This package implements the crcmodel and the crcexperiment R6 classes and can be used to perform Robust Decision Making Analyses of multiple cancer screening models using High-Performance Computing resources. The package supports large-scale computational tasks that have historically been deemed unfeasible for microsimulation models, such as defining and conducting Probabilistic Sensitivity Analyses (PSAs) or robustness analyses of large models and large combinations of parameter sets.

Installation

This package can be installed from GitHub with:

# install.packages("remotes")
remotes::install_github("c-rutter/crcrdm")

Status

This package is being actively used by the crcspin model, which itself is not contained in the package.

This repository does not contain code for the models used within those analyses, and is made public to allow for automatic installation across HPC systems and to support continuous integration workflows in repositories that depend on this package.

Documentation

A documentation page is available at this link. This documentation page describes the package main classes and their methods.

Key classes: crcmodel, and crcexperiment:

This package implements two R6 classes:

  1. crcmodel : Is the basic unit and encompasses a single model structure. A crcmodel can be either calibrated or not. if it is calibrated, the model object can include the posterior distribution of its parameters. A single model may contain multiple posterior distributions when those parameter sets were created by different calibration runs, with potentially different targets or priors.
  2. crcexperiment : Contains the definition of an experiment experiment to be applied over the crcmodels included in it.

Automated Tests and Test Coverage

This package is tested automatically after every commit across a few platforms. Results from these automated checks can be found here. A test coverage report can be found here.