Challenges of COVID-19 Case Forecasting in the United States, 2020-2021

This repository contains R scripts for all analyses included in the manuscript titled “Challenges of COVID-19 Case Forecasting in the United States (US), 2020-2021”. COVID-19 case forecasts submitted to the US COVID-19 Forecast Hub, a large consortium of researchers that leverage information from individual models to predict the near-term burden of COVID-19 in the US, were evaluated for precision and accuracy. Seventy-four weeks (between July 28, 2020, and December 21, 2021) of forecasts for US states and counties (approximately 9.7 million forecasts) were included in the evaluation.

Repository structure

All included code can be run in R, with the required packages indicated at the top of each script. Analyses were performed in RStudio (RStudio 2022.07.0+548) using R (v. 4.2.1).

All R scripts are in the Code folder and figures included in the manuscript are in the Figures folder.

Ground truth data, estimates of the state-level time-varying reproduction number, and county population size are in the Data folder, while submitted forecasts are publicly available for download. Code is provided for pulling submitted forecast data.

Data sources

The following data sources were used in the analysis and are publicly available for download:

• Forecast data are pulled from a cloned repository of the US COVID-19 Forecast Hub at https://github.com/reichlab/covid19-forecast-hub.

• Time-varying reproduction number estimates data were downloaded from https://github.com/epiforecasts/covid-rt-estimates. The data pulled for this analysis are located in the Data/rt estimates folder

• County population size estimates from 2018 (SVI2018_US_COUNTY.csv): https://www.atsdr.cdc.gov/placeandhealth/svi/data_documentation_download.html

Forecasts were evaluated against reported COVID-19 case reports collated by the Johns Hopkins Center for Systems Science and Engineering (CSSE). Surveillance data are processed by the COVID-19 Forecast Hub and available for forecast teams. We used ground truth, surveillance data as of April 2, 2022, from the COVID-19 Forecast Hub repository (see truth-Incident Cases.csv)