COVID-19 Forecasts: Deaths

COVID-19 Forecasts: Deaths
Updated Aug. 6, 2020

Observed and forecasted new and total reported COVID-19 deaths as of August 3, 2020.

Interpretation of Forecasts of New and Total Deaths

  • This week CDC received forecasts of national COVID-19 deaths over the next 4 weeks from 34 modeling groups.  Those forecasts predict:
    • The number of new COVID-19 deaths reported each week (32 forecasts), which indicates how reported deaths are likely to increase or decrease in the coming weeks
    • The total number of COVID-19 deaths reported by the end of each week (33 forecasts), which helps us understand the likely overall impact of the pandemic in the coming weeks

Of the 34 modeling groups, 31 provided forecasts of both new and total deaths, one provided forecasts of new deaths only, and two provided forecasts of total deaths only.

  • This week’s national ensemble forecast predicts that weekly reports of new COVID-19 deaths may decrease over the next 4 weeks, with 4,500 to 10,600 new deaths reported during the week ending August 29. The ensemble forecast predicts that 175,000 to 190,000 total COVID-19 deaths will be reported by August 29.
  • State-level ensemble forecasts predict that the number of reported new deaths per week may increase over the next four weeks in Hawaii and Puerto Rico and may decrease in Florida, Mississippi, New Mexico, the Northern Mariana Islands, Ohio, Texas, Vermont, and the Virgin Islands.

National Forecast

National forecast as of 8/3/2020
  • The top row of the figure shows the number of new COVID-19 deaths reported in the United States each week from May 30 through August 1 and forecasted new deaths over the next four weeks, through August 29.
  • The bottom row of the figure shows the number of total COVID-19 deaths in the United States each week from May 30 through August 1 and the forecasted number of total COVID-19 deaths over the next four weeks, through August 29.
  • Models make various assumptions about the levels of social distancing and other interventions, which may not reflect recent changes in behavior.  See model descriptions below for details.

State Forecasts

State-level forecasts figures show observed and forecasted state-level new and cumulative COVID-19 deaths in the US. Each state forecast uses a different scale, due to differences in the numbers of COVID-19 deaths occurring in each state.

Forecasts fall into one of two categories:

  • The Auquan, CMU, DDS, Columbia-UNC, ERDC, ESG, Geneva, GT-DeepCOVID, ISU, Karlen, LANL, LNQ,  LSHTM, MIT-CovAlliance, MIT-ORC, MOBS, Oliver Wyman, NotreDame-Mobility, QJHong, RPI-UW, STH, UA, UCM, UM, UMass-MB, USC, and UT forecasts assume that existing control measures will remain in place during the prediction period.
  • The Columbia, COVID19Sim, GT-CHHS, IHME, JCB, JHU, NotreDame-FRED, PSI, UCLA, and YYG forecasts make different assumptions about how levels of social distancing will change in the future.

Download state forecasts pdf icon[29 pages]1

Download forecast data excel icon[1 sheet]

Additional forecast data and information on forecast submission are available at the COVID-19 Forecasting Hubexternal icon.

Modeling Groups

Forecasts were provided by the following modeling groups. Each modeling group submitted forecasts for both new and total deaths, unless otherwise indicated.

Model names: Auquan

Note:  This group submitted forecasts of total deaths only.

Intervention assumptions: These projections do not make specific assumptions about which interventions have been implemented or will remain in place.

Methods: SEIR model

Note: This group submitted state-level forecasts of new deaths only.

Model name: CMU

Intervention assumptions: These projections do not make specific assumptions about which interventions have been implemented or will remain in place.

Methods: Autoregressive time-series model.

Model name: Columbia

Intervention assumptions: This model assumes that contact rates will increase 5% per week over the next two weeks. The reproductive number is then set to 1 for the remainder of the projection period.

Methods: Metapopulation SEIR model.

Model name: Columbia-UNC

Intervention assumptions: This model assumes that transmission intensity will peak in early July and then gradually decline.

Methods: Statistical survival-convolutional model.

Model name: Covid19Sim

Intervention assumptions: This model is based on assumptions about how levels of social distancing will change in the future.

Methods: SEIR model.

Model name: DDS

Intervention assumptions: This model assumes that the effects of interventions are reflected in the observed data and will continue going forward.

Methods: Bayesian hierarchical model.

Model name: GT_CHHS   (state-level forecasts only)

Intervention assumptions: This model assumes that once stay-at-home orders are lifted, contact rates will gradually increase.  It also assumes that some households containing symptomatic cases will self-quarantine.

Methods: Agent-based model.

Model name: GT-DeepCOVID (formerly GA_Tech)

Intervention Assumptions: This model assumes that the effects of interventions are reflected in the observed data and will continue going forward.

Methods: Deep learning.

Model name: IHME

Intervention assumptions: Projections are adjusted to reflect differences in aggregate population mobility and community mitigation policies.

Methods: Combination of a mechanistic disease transmission model and a curve-fitting approach.

Model name: ISU

Intervention Assumptions: These projections do not make specific assumptions about which interventions have been implemented or will remain in place.

Methods: Nonparametric spatiotemporal model.

Model name: JCB

Intervention assumptions: The incidence of COVID-19 in the population determines the strength and resulting impact of control measures in the future.

Methods: Phenomenological statistical model.

Model name: JHU

Intervention Assumptions: This model assumes that the effectiveness of interventions is reduced after shelter-in-place orders are lifted.

Methods: Stochastic metapopulation SEIR model.

Model name: Karlen

Intervention assumptions: This model assumes that the effects of interventions are reflected in the observed data and will continue going forward.

Methods: Discrete-time difference equations.

Note: This group submitted forecasts for new deaths only.

Model name: LNQ

Intervention assumptions: This model assumes that the effects of interventions are reflected in the observed data and will continue going forward.

Methods: Machine learning.

Model name: LSHTM

Intervention assumptions: These projections assume that current interventions will not change during the forecasted period.

Methods: This forecast is an ensemble of three different models: A time-varying reproductive number-based model, a time series model based on numbers of deaths, and a time series model based on numbers of cases and deaths.

Model name: LANL

Intervention assumptions: This model assumes that currently implemented interventions and corresponding reductions in transmission will continue, resulting in an overall decrease in the growth rate of COVID-19 deaths.

Methods: Statistical dynamical growth model accounting for population susceptibility.

Model name: MIT-CovAlliance

Intervention Assumptions: The projections assume that current interventions will remain in place indefinitely.

Methods: SIR model.

Note: This group submitted forecasts of total deaths only.

Model name: MIT-ORC

Intervention Assumptions: The projections assume that if cases reach a certain threshold, interventions will be reinstated.

Methods: SEIR model.

Model name: MOBS

Intervention assumptions: The projections assume that social distancing policies in place at the date of calibration are extended for the future weeks.

Methods: Metapopulation, age-structured SLIR model.

Model names:

  • NotreDame-Mobility
  • NotreDame-FRED (state-level forecasts only)

Intervention assumptions: These forecasts assume that population-level mobility is a reliable proxy for adherence to social distancing, and that recent trends in mobility will continue over the coming weeks.

Methods:

  • NotreDame-Mobility: SEIR model fit to data on deaths, test positivity, and population mobility.
  • NotreDame-FRED: Agent-based model.

Model name: Oliver Wyman

Intervention assumptions: These projections assume that current interventions, will remain unchanged during the forecasted period.

Methods: Time-dependent SIR model for detected and undetected cases.

Model name: PSI

Intervention assumptions: These projections assume that current interventions will not change during the forecasted period.

Methods: Stochastic SEIRX model.

Model name: QJHong

Intervention assumptions: These projections assume that current interventions will not change during the forecasted period.

Methods: Machine learning using case data and mobility data.

Model name: RPI-UW

Intervention assumptions: These projections calibrate the rate of transmission to the average rate of population mobility since the start of the epidemic and assume that this relationship will not change in the next four weeks.

Methods: SIR model fit to mobility data.

Model name: ESG

Intervention assumptions: These projections assume that current interventions will not change during the forecasted period.

Methods: Fitting reported data to multiple skewed gaussian distributions.

Model name: STH

Intervention assumptions: These projections assume that current interventions will not change during the forecasted period.

Methods: Statistical growth model.

Model name: ERDC

Intervention assumptions: These projections assume that current interventions will not change during the forecasted period.

Methods: SEIR model.

Model name: UA

Intervention assumptions: This model assumes that current interventions will remain in effect for at least four weeks after the forecasts are made.

Methods: SIR mechanistic model with data assimilation.

Model name: UCLA

Intervention assumptions: This model assumes that contact rates will increase as states reopen. The increase in contact rates is calculated for each state.

Methods: Modified SEIR model.

Model name: UCM

Intervention assumptions: These projections assume that current interventions will not change during the forecasted period.

Methods:  SEIR model.

Model name: Geneva

Intervention assumptions: The projections assume that social distancing policies in place at the date of calibration are extended for the future weeks.

Methods: Exponential and linear statistical models fit to the recent growth rate of cumulative deaths.

Model names: UMass-MB, Ensemble

Intervention assumptions:

  • UMass-MB: These projections do not make specific assumptions about which interventions have been implemented or will remain in place.
  • Ensemble: The national and state-level ensemble forecasts include models that assume certain social distancing measures will continue and models that assume those measures will not continue.

Methods:

  • UMass-MB: Bayesian SEIRD model.
  • Ensemble: Equal-weighted combination of 2 to 11 models, depending on the availability of national and state-level forecasts. To ensure consistency, the ensemble includes only models with 4 week-ahead forecasts and models that do not assign a significant probability to there being fewer cumulative deaths than have already been reported by the day of submission. Only one model was available for the forecasts for Guam and the Northern Mariana Islands.

Model name: UM

Intervention assumptions: These projections assume that current interventions will remain unchanged during the forecasted period.

Methods: Ridge regression.

Model name: USC

Intervention assumptions: These projections assume that current interventions will remain unchanged during the forecasted period.

Methods: SIR Model.

Model name: UT

Intervention assumptions: This model estimates the extent of social distancing using geolocation data from mobile phones and assumes that the extent of social distancing will not change during the period of forecasting. The model is designed to predict confirmed COVID-19 deaths resulting from only a single wave of transmission.

Methods: Nonlinear Bayesian hierarchical regression with a negative-binomial model for daily variation in death rates.

Model name: YYG

Intervention assumptions: The model accounts for individual state-by-state re-openings and their impact on infections and deaths.

Methods: SEIS mechanistic model.

Footnote

1 The full range of the prediction intervals is not visible for all state plots. Please see the forecast data for the full range of state-specific prediction intervals.