Skip to main content
Dryad logo

An interactive tool to forecast us hospital needs in the Coronavirus 2019 pandemic

Citation

Locey, Kenneth et al. (2021), An interactive tool to forecast us hospital needs in the Coronavirus 2019 pandemic, Dryad, Dataset, https://doi.org/10.5061/dryad.1ns1rn8rx

Abstract

We developed an application (https://rush-covid19.herokuapp.com/) to aid US hospitals in planning their response to the ongoing COVID-19 pandemic. Our application forecasts hospital visits, admits, discharges, and needs for hospital beds, ventilators, and personal protective equipment by coupling COVID-19 predictions to models of time lags, patient carry-over, and length-of-stay. Users can choose from seven COVID-19 models, customize a large set of parameters, examine trends in testing and hospitalization, and download forecast data.

The data and scripts contained herein are used to generate Figure 1 of the associated manuscript, which presents general forms of the models used by our application and presents results for each model across time.

Methods

Date provided here are based on data our application uses. Specifically, aggregated reports of cumulative cases across US states and territories from the Johns Hopkins University Center for Systems Science and Engineering (JHU CSSE) public GitHub repository, and state and territory population sizes based on publically avialable US Census Bureau data (2010 – 2019). See the associated mansuscript and/or our application's public GitHub repository (https://github.com/Rush-Quality-Analytics/SupplyDemand) for greater details. 

To recreate Figure 1 of the main manuscript, run the following scripts in a python environment or suitable terminal window with python 3.6+ installed:

1. ModelFxns_Figs.py -- A script to generate the general predicted forms of each model; associated with Figure 1 of the associated manuscript. Generates the Model_Forms.png file located in the figures folder.

2. ModelPerformance_Figs.py -- A script to generate results figures for each model; associated with Figure 1 of the associated manuscript. Generates the Model_Performance.png file located in the figures folder.
 

Included data files

1.  model_results_dataframe.pkl -- A python-specific pickle file located in the data folder and which contains results for each model's predictive accuracy across time and US states and terroritories. Used by ModelPerformance_Figs.py.

2. COVID-CASES-DF.txt -- A file located in the data folder and which contains data downloaded and curated from JHU CSSE.

3. StatePops.csv -- A file located in the data folder and which contains data on US State and territory population sizes as well as reported dates of COVID-19 arrival (gathered from state/territory public websites).

Additional python scripts

1. results_dataframe.py -- A script to regenerate the model_results_dataframe.pkl file. Warning: This script may take several days to run because of the many iterations needed for the SEIR-SD model. Running this script will overwrite the existing model_results_dataframe.pkl file, so take necessary precautions.

2. model_fxns.py -- A script containing functions for running models used by our application. This script is used by results_dataframe.py to generatth the model_results_dataframe.pkl file.

Usage Notes

1. The model_results_dataframe.pkl file is a python specific file format.

2. Running the ModelFxns_Figs.py and ModelPerformance_Figs.py is all that is needed to recreate the subplots of figure 1. The user should have the following libraries/softwares installed:

Python 3.6 or greater
numpy 1.16 or greater
pandas 0.24 or greater