Background

This report provides a brief summary of the weekly ensemble forecast from the COVID-19 Forecast Hub. In collaboration with the US CDC, our team aggregates COVID-19 forecasts from dozens of teams around the globe. Typically on Wednesday or Thursday of each week, a summary of the week’s forecasts from the COVID-19 Forecast Hub appear on the official CDC COVID-19 forecasting page.

Every week, teams submit their forecasts to the COVID-19 Forecast Hub. This past week, 44 models were submitted, including one model from a new team.

Each Tuesday, we combine the most recent forecasts from each team into a single “ensemble” forecast of reported COVID-19 cases at the county, state, and national level and deaths at the state and national level. At the moment, we only generate ensemble forecasts for four weeks into the future, as we don’t have reliable evidence that the models are accurate past that horizon.

Weekly reports from the COVID-19 Forecast Hub can be found at this page. We expect that these reports will become more detailed as the weeks go on.

COVID-19 Mortality Forecasts

National level

This week, our ensemble combined forecasts from 32 different models. We note that this week several states, most notably Texas, modified the way in which they were reporting “probable” COVID-19 deaths. In Texas, this resulted in a one-time additional count of over 600 deaths in the data that we consider to be “ground truth”.

At the national level, the ensemble model’s best guess is that we will see between 6,500 and 7,400 deaths each week for the next four weeks with around 181,000 deaths by August 29th (Figure 1). However for the week ending August 29th, the ensemble shows substantial uncertainty, with observed deaths between 4,500 and 10,500 deemed possible (95% prediction interval: 4,560 - 10,587).

Throughought most of July, models have in general shown broad agreement about the trajectory of the outbreak over the coming weeks. However, the recent surge in cases has left models with quite different interpretations about what the next few weeks hold in terms of how many reported deaths from COVID-19 we will see. You can explore the full set of models, including their forecasts for past weeks online at our interactive forecast visualization.

Figure 1: Weekly observed and forecasted COVID-19 deaths. Observed data from JHU CSSE and forecasts from the COVID-19 Forecast Hub.

Figure 1: Weekly observed and forecasted COVID-19 deaths. Observed data from [JHU CSSE](https://github.com/CSSEGISandData/COVID-19/tree/master/csse_covid_19_data/) and forecasts from the [COVID-19 Forecast Hub](https://covid19forecasthub.org/).

State-by-state Breakdown

The ensemble model estimates that 41 states and territories of the US have a greater than 50% chance of having more deaths in the next two weeks compared to the past two weeks (Table 1). But the model shows substantial uncertainty about the increases, with only 11 states and territories showing a greater than 75% chance of an increase over the next two weeks (Alaska, Arkansas, Hawaii, Minnesota, New Jersey, Oklahoma, Wyoming, American Samoa, Guam, Northern Mariana Islands, Puerto Rico). Note that for the purposes of the tables below, we did not count any of the 631 “probable” COVID-19 deaths newly reported last week in Texas as having occured in the last two weeks.

The sortable and searchable table below shows the total number of reported COVID-19 deaths at the US level and by state over the last two weeks (ending Saturday, August 1, 2020) and the forecasted counts for the subsequent two weeks (ending August 15, 2020).

The sortable and searchable table below shows the total number of reported COVID-19 deaths at the US level and by state as of Saturday, August 1, 2020 (“Total COVID-19 Deaths”) as well as the rate of reported COVID-19 deaths in the population (standardized per 100,000 population) over the last two weeks and over the next two weeks. Looking at the rates allows for easier comparison across states, where you can see which states have had or are predicted to have propoportionally higher rates in comparison to other states. These tables calculate an average daily number of deaths per 100,000 population across the last two weeks (ending Saturday, August 1, 2020) and forecasted for the following two weeks (ending August 15, 2020).

This report was reproducibly and dynamically generated using RMarkdown. The code for the report can be found here.