This repository contains code associated with the paper Reopening California: Seeking Robust, Non-Dominated COVID-19 Exit Strategies. This paper will be first released as a working paper then submitted to a journal.
R Package Developer: Pedro Nascimento de Lima
Model Developers: Raffaele Vardavas, Pedro Nascimento de Lima
Data Pipeline Developers: Pedro Nascimento de Lima, Lawrence Baker
-
Working Paper: Reopening California: Seeking Robust, Non-Dominated COVID-19 Exit Strategies.
-
Working Paper: Modeling COVID-19 Nonpharmaceutical Interventions: Exploring periodic NPI strategies.
-
Original Decision Support Tool: The Health and Economic Impacts of Nonpharmaceutical Interventions to Address COVID-19: A Decision Support Tool for State and Local Policymakers
Please use the .Rproject to work in this project with R Studio. When necessary, all scripts assume that the working directory is the base directory.
This repository uses git lfs. Make sure to download all git lfs files before trying to run the model.
We developed the c19randepimod R package specifically to inform
policies during the COVID-19. pandemic The package includes a series of
functions to gather data, define a c19model model class, calibrate the
model, define experimental designs, run experiments and generate
results. By defining our model as a class, we allow ourselves and future
users to extend our original c19model over time. Currently, models
based on the c19model class necessarily need to be compatible with the
deSolve package, but future versions can relax this requirement and
allow stochastic models or ABMs. This allows us to create new model
classes that inherit the functions implemented for the c19model class
regardless of the model structure. Since our first
publication, we used
different model classes for each of our publications as we learned more
about COVID-19 and as our response to COVID-19 changed. For example, our
original state policy tool published in May 2020 did not include
vaccination, behavioral responses to vaccination and hesitancy,
seasonality, and increases in transmissibility from variants, but the
model used in the Reopening California
paper does.
This flexibility and model design choices proved to be helpful and
instrumental for our work. They allowed us to publish a widely used tool
within six weeks from conception to publication, and to follow up with
analyses that reflected progress in the pandemic.
The figure below illustrates the main steps necessary to run the model.
The R package starts by gathering data from the covidtracking
api and a spreadsheet containing
model inputs using the get_augmented_inputs function. This function
creates a model object that contains everything we need to calibrate the
model. Then, we use the calibrate or the recent calibrate_imabc
function to find parameters that can produce outcomes consistent with
observed time-series.
Finally, we can run the model after the calibration period by using the
functions set_parameter to set different types of paramters and use
the function set_experimental_design to define the future experimental
design. We then use the evaluate_experiment function to evaluate the
experimental design. The section below indicates how we perform these
steps for the Reopening California paper.
This folder is organized as follows:
-
./c19randepimod: contains a snapshot of the c19randepimod R package code. We created this package to generalize functions that will be useful beyond this analysis. The package has been used for all the publications we list here, but this reposistory only contains code for the “Reopening California” paper.
-
./00_dependencies: contains the dependencies necessary to run all analyses. Run the install_dependencies.R file for the first time if needed. This will install the dependencies you need to run the model. Make sure you are using R > 3.6.1.
-
./01_calibration: contains scripts three R scripts. The first called 01_calibration_imabc.R sets up and runs the calibration by calling the function calibrate_imabc. This function is contained in the file in the file calibrate_imabc.R. The inputs folder contain the spreadsheets that are read in as inputs for the model.
-
./02_future_runs: contains three R scripts. The first called 01_setup_experiments.R is a simple script that sets the experimental design, as specified in the setup_rdm_experiments.R file. This is where the future experimental design is defined. Then, we use the hpc_run_experiments to run the experiments defined previously in parallel. One can run this script manually as well, but we use the hpc_run_experiments.sbatch file contained in the root directory to run the model in HPC clusters using slurm’s array jobs. This allows us to run the model across several nodes.
-
./03_regret_analysis: contains the script 03_regret-analysis.R that calls functions in regret-functions.R. This is where the results are read, and this is where we analyse the regret from different policies. This script produces the plots we use in our papers and presentations.
-
./04_tableau_plots: contains a tableau workbook we use to generate plots and summary tables. I use the tableau workbook to generate the summary tables in the strategy_summaries.xlsx file.
For questions, reach out to Pedro at plima [at] rand dot org
Copyright (C) 2021 by The RAND Corporation. This repository is released as open source software under a GPL-2.0 license. See the LICENSE file.