COVID-19 model FAQs

Table of contents

Overall
Scenarios
Deaths
Infections and testing
Hospital capacity
Social distancing
Methods/modeling
About the project

Overall

What has changed with the IHME COVID-19 projections?

The forecast IHME released on March 26, 2020, was geared to helping hospitals plan for a surge in demand for their resources (e.g., beds, ICUs, ventilators) to fight COVID-19. We have since updated our modeling approach, and with many locations around the world lifting social distancing measures, attention is now on how best to prevent and manage a resurgence of the disease while enabling people to get back to work and school safely. In June, we developed a model to extend our forecasts to October 1, 2020, and developed three scenarios to show how different policy decisions on social distancing mandates and mask use could affect the trajectory of the pandemic.

Why are you showing different scenarios in the tool?

In making these forecasts, we aim to help policymakers plan for the days and months ahead, and take action to change the course of the pandemic for the better. We have developed three different scenarios that can help policymakers understand how different policy decisions could affect the trajectory of the pandemic in their location.

Why do the numbers keep changing?

The world’s understanding of and response to COVID-19 is rapidly evolving. In June, we extended our forecasts to October 1, 2020. These forecasts indicate that COVID-19 infections could rise in the Northern Hemisphere between August and September as the fall flu season begins, human contact increases, and social distancing mandates continue to be relaxed. Cases are also projected to rise in many Latin American and Caribbean as well as Arab League countries where mandates on social distancing and mask wearing have been eased.

We strive to incorporate new evidence as soon as it becomes available, and our estimates will change in response to these data. We do this to make the tool as useful as possible to governments and hospital administrators, who need to know how the situation is changing in real time. Our aim is to produce the best possible predictions given what we know today – and to continually improve these estimates tomorrow. Also, we improve our methods based on feedback that we receive. If you have questions or suggestions, please contact us here. Learn more about these changes by visiting our estimation updates page.

Scenarios

How should I interpret the different scenarios?

The “Current projection” scenario assumes that social distancing mandates will continue to be lifted, but will be re-imposed for six weeks if daily death rates reach 8 per million. The “Mandates easing” scenario assumes that mandates will continue to be lifted and will not be re-imposed. The “Universal masks” scenario assumes that mask wearing will reach 95% in 7 days, and social distancing mandates will continue to ease, but will be re-imposed for six weeks if daily death rates reach 8 per million.

What assumptions are included in each scenario?

The assumptions for each scenario include:

Scenario	Mask use in the population?	When are mandates removed?	Threshold at which mandates are re-imposed?	What mandates?
Current projection	Assumes mask use continues at currently observed rates	Assumes that the gradual easing of social distancing mandates continues	Assumes that mandates will be re-imposed for six weeks if daily deaths reach 8 per million	Educational facilities closed Non-essential businesses closed People ordered to stay at home Large gatherings banned + Mandatory mask use put in place
Mandates easing	Assumes mask use continues at currently observed rates	Assumes that the gradual easing of social distancing mandates continues	Assumes that mandates are never re-imposed	Not applicable
Universal masks	Assumes mask use rises to 95% within 7 days	Assumes that the gradual easing of social distancing mandates continues	Assumes that mandates will be re-imposed for six weeks if daily deaths reach 8 per million	Educational facilities closed Non-essential businesses closed People ordered to stay at home Large gatherings banned + Mandatory mask use put in place

Deaths

How many overall deaths will there be?

Our model is updated to account for new data and information, and the estimates may change as a result. For the latest estimate, visit our COVID-19 projections tool. At present, the forecast extends to October 1.

Why do your results show a wide range in the forecast for deaths?

Larger uncertainty intervals – or the range within which estimated deaths are likely to fall – can occur because of limited data availability, small studies, and conflicting data. A smaller range generally reflects extensive data availability, large studies, and data that are consistent across sources.

Why did the estimates for my location change?

In June, we extended our forecasts to October 1, 2020. These forecasts indicate that COVID-19 infections could rise in the Northern Hemisphere between August and September as the fall flu season begins, human contact increases, and social distancing mandates continue to be relaxed. Cases are also projected to increase in Latin American and Caribbean as well as Arab League countries where mandates on social distancing and mask use have been eased.

Also, as the pandemic unfolds and we receive more data on COVID-19 deaths, our projections adjust to follow new trends that emerge. The largest changes tend to be seen in locations with little data, which are in the early phases of the pandemic.

Along with data on reported deaths, we are now using trends in reported cases and hospitalizations to project deaths, which we believe improves the accuracy of our forecasts. In addition to using anonymized mobile phone data along with data on testing and population density, we recently added data on self-reported mask use, annual pneumonia death rate, and self-reported contacts to understand transmission of the virus. We use seasonal patterns of pneumonia transmission in the model since they seem to closely resemble patterns of COVID-19 transmission.

To learn more, please visit our update page.

Why is the peak for daily deaths still forecast in the future when it looks like it has already occurred in my location?

The date of peak daily deaths depends on the model’s projections. If the model projects that the number of daily deaths will continue to rise, then the peak will be projected for a future date. It is important to note that the data on daily deaths may fluctuate dramatically due to inconsistencies in reporting. Health care workers are extremely busy caring for COVID-19 patients, so they may fall behind on reporting deaths. Once health care workers catch up on their reporting, however, it may appear as though there has been a spike in daily deaths.

Why are the “actual deaths” shown in your results for my location different from what is shown on the government’s official page?

For deaths, we primarily use the COVID-19 death data aggregated by the Johns Hopkins University (JHU) data repository. The JHU repository uses Coordinated Universal Time (UTC), which means new days start at 8 p.m. Eastern time. The JHU counts may differ slightly from local government data as a result of these timing differences.

Also, the JHU repository is not necessarily synchronized to the update schedule of every location, so there may be a short lag that is reflected in a difference between our recorded daily deaths in a given location and those ultimately reported on government websites. Although this will be corrected when we update our analysis, in some cases, these differences may persist for several days.

We have added additional data from government websites for Brazil, Canada, France, Germany, Italy, Japan, Mexico, Spain, the UK, and the US states of Hawaii, Indiana, Illinois, Kentucky, Maryland, and Washington. For New York, we use data from the New York City Department of Health and Mental Hygiene and the New York Times GitHub repository.

Do you have the number of deaths by age group?

We do not currently have the number of deaths estimated to occur within each age group, but we are working on making these numbers available as part of our regular updates to the model and results.

News reports have described probable under-reporting of COVID-19 deaths, particularly in the early data from Wuhan – does your model account for this?

The model no longer relies on any data from Wuhan, China.

There are reports of deaths being under-reported in places. How does this impact your forecast?

We are learning that not all deaths due to COVID-19 that occur at home or in nursing homes have been attributed to COVID-19. As awareness increases, the number of reported deaths is growing, with some locations now reporting presumptive COVID-19 deaths. Another challenge is that COVID-19 death data fluctuate substantially each day, with some locations reporting more deaths on Tuesdays than on Sundays and Mondays. We believe this variation is due to data reporting practices instead of actual death patterns. To mitigate the impact of inconsistent reporting on our forecasts, our published predictions are based on averaging multiple iterations of projections. As new data emerges, we incorporate it into our model, and our projections will shift up or down in response to the data. To learn more, see our estimation updates.

For Ecuador and Peru in particular, the number of reported deaths due to COVID-19 appears to be improbably low. Instead of using reported COVID-19 deaths for these countries, we are approximating deaths from COVID-19. To approximate COVID-19 deaths, we used the number of excess deaths occurring in Ecuador and Peru during the COVID-19 pandemic and observations from other countries where we had weekly reports of total deaths and high-quality data on COVID-19 deaths.

Infections and testing

Why does the dotted line for infections grow more rapidly in certain places?

Increases in human mobility, loosening of social distancing measures, and seasonal disease transmission patterns (COVID-19 transmission appears to be highest during the fall and winter) can drive up infections. However, taking precautions such as maintaining at least six feet between individuals at any gathering, wearing cloth masks or face coverings in public areas, and regular handwashing and sanitation could reduce the risk of disease transmission. Insufficient testing (testing that doesn’t keep pace with infections) as well as insufficient contact tracing capabilities, or lack of isolation for known infections, could also contribute to this rise.

Why are you modeling tests and estimated infections?

We are modeling testing because it is significantly related to disease transmission. As testing increases, the contact tracing and isolation associated with it help decrease disease transmission.

Estimated infections are a key input to the disease transmission component of our hybrid model. Estimated infections are also useful for understanding when the pandemic is likely to be contained in different locations.

Comparing testing to estimated infections allows government officials to understand how well testing is keeping pace with infections. If testing in a particular location falls below estimated infections on a given day, the risk of undetected community transmission and resurgence increases.

How are you defining estimated infections versus confirmed infections?

We define estimated infections as prevalent infections – that is, all cases that exist in a location on a given day, not just new ones. Confirmed infections are those infections that have been identified through testing.

How are you modeling estimated and confirmed infections?

For estimated infections, we start with death estimates, then work backward, using infection fatality ratios to estimate infections based on deaths. Confirmed infections are based on case data from the Johns Hopkins University (JHU) data repository on Github, averaged over the last 3 days to account for delays in reporting.

How did you determine the infection fatality ratio?

We reviewed data from all locations where extensive COVID-19 testing has occurred, then used the infection fatality ratio from the locations where the ratios were lowest: the Diamond Princess cruise ship and New Zealand.

What data are you using to model tests?

Our primary sources for US testing data are compiled by the COVID Tracking Project. For other locations, we rely primarily on data reported by Our World in Data. However, for Brazil and the Dominican Republic, we use government data.

How are you estimating tests and projected tests?

For tests, we average the data points from the last 3 days to account for delays in reporting. In cases where there are gaps in reported testing data, we extrapolate testing levels back to the date of the first confirmed case. We model projected tests based on past trends observed in the testing data.

Hospital capacity

Why are you estimating hospital capacity?

These forecasts were developed to help hospitals and health systems prepare for the surge of COVID-19 patients.

Why have your estimates for hospital resource needs changed?

As data continue to come in, our estimates may change. Specifically, new death data and new information about the number of COVID-19 patients who need hospital beds have changed our projections.

What if hospitals end up over-preparing for the surge in demand?

We believe that it is better for health systems to prepare for the upper range of values in the forecast, instead of the middle or lower range, to ensure that sufficient resources are in place. Many planners also prefer to utilize the upper range in their preparations.

Are your estimates for the number of available beds and ICU beds just for COVID-19 patients or for all patients?

The hospital resources shown are those estimated to be available for COVID-19 patients. We have excluded non-COVID patient needs, that is, the typical percentage of hospital beds occupied by other patients and emergencies.

Why do your projections for number of hospital beds needed for COVID-19 patients not match what I see in my location?

These discrepancies typically stem from the limitations of the datasets that we are using to estimate hospital and ICU beds needed for COVID-19 patients. We obtained our data from sources such as government websites, hospital associations, the Organisation for Economic Co-operation and Development, WHO, and published studies. As hospitals rapidly increase their capacity to deal with the influx of COVID-19 patients, we do not have access to data that reflect how bed counts are changing in real time. Given these limitations, government staff and health system administrators should continue to compare any available local hospital capacity data against our projections.

Are “invasive ventilators needed” additional ventilators needed for COVID-19 patients?

These are total ventilators needed for COVID-19 patients, excluding those needed by non-COVID-19 patients. These numbers do not reflect shortfalls in ventilators needed for COVID-19 patients because we lack data on the number of ventilators that states and countries currently have available.

Why doesn’t the number of beds needed in your visualization tool match the hospitalization admissions I’m seeing reported elsewhere?

Public records of the number of hospitalizations on a particular day do not account for the number of people who are already occupying beds.

Do you have models predicting health workforce needs at the peak of the outbreak?

We do not currently have this information. Having a patient in every bed, or running out of beds entirely, may mean that a hospital is facing substantial human resource shortages.

How will hospitals get more supplies? Is there a time when they will just run out?

Whether hospitals will be able to obtain sufficient supplies depends on the number of patients as well as any extra support that they may receive from governments or others.

What are the strategies that hospitals have to deal with the shortfall in capacity?

There are a variety of options for hospitals facing shortfalls in resources, including canceling elective procedures, setting up additional beds, constructing temporary facilities, and using mobile military resources. Scaling up production of ventilators, masks, and other personal protective equipment is likely to be needed to ensure these resources are available to hospitals as demand grows.

Social distancing

What are you trying to show on the social distancing chart?

We are showing how human mobility has changed relative to background levels for each location. These mobility patterns have changed as social distancing measures have been implemented and/or eased. Individual decision-making also factors into mobility patterns, as individuals in certain locations choose to increase or decrease their movement regardless of government mandates.

What do you mean by "mobility"?

Mobility refers to personal movement by a population and is based on anonymous cellphone data several technology companies have made available for the purposes of fighting COVID-19. Mobility is an indicator of greater potential for personal contact, which can contribute to the spread of the disease. When mobility is high, the risk of COVID-19 spreading may also be high. However, as mobility increases, taking precautions such as maintaining at least six feet between individuals in any gathering, wearing cloth masks or face coverings in public areas, and regular handwashing and sanitation could reduce the risk of disease transmission.

What do you mean by “mandates?”

Mandates include the following government orders, which are based on the New Zealand Government 4-level alert system:

Educational facilities closed
All levels of educational instruction (primary, secondary, and higher education) are required to implement distance learning and are closed for in-person teaching activities.
Non-essential businesses ordered to close
Only locally defined “essential services” are in operation. Typically, this results in closure of public spaces such as stadiums, cinemas, shopping malls, museums, and playgrounds. It also includes restrictions on bars and restaurants (they may provide take-away and delivery services only), closure of general retail stores, and services (like nail salons, hair salons, and barber shops) where appropriate social distancing measures are not practical. There is an enforceable consequence for non-compliance such as fines or prosecution.
People ordered to stay at home
All individuals are ordered to stay at home unless traveling to essential services. Physical contact is only allowed between residents of the same household. Exercise may be permitted, as a solitary, distanced exercise, or with members of the same household. There is an enforceable consequence for non-compliance such as fines or prosecution.
Severe travel restrictions
Location borders are closed to all incoming traffic except for those in provision of essential services and returning residents isolated in foreign territories. Automobile travel is restricted to accessing and working at essential services. Public transit is closed.
Any gathering restrictions
Mandatory restrictions on gatherings of individuals of any number are in place. These can apply to public and private gatherings.
Any business closures
The mandatory closure of any businesses is in effect. These restrictions need not apply to all businesses but can apply to just a specific subset (like bars and restaurants).

How do social distancing measures factor into your model?

We use social distancing measures in our model to project COVID-19 deaths. Data from Germany, Italy, the US, and Spain have demonstrated how social distancing measures can save lives. Using data from these countries helps us understand how quickly we can expect deaths to decline after a location implements social distancing measures.

Starting May 4, we began modeling the correlation between increases in mobility and disease transmission, which has allowed us to project how deaths from COVID-19 may increase in places where the pandemic has not yet been curbed. Also, if people are being more cautious as they move around, maintaining six feet of distance and wearing masks, this could reduce the risk of disease transmission as mobility increases.

Methods/modeling

Why are your estimates different than those produced by other organizations?

Our model is designed to be a planning tool for government officials who need to know how different policy decisions can radically alter the trajectory of COVID-19 for better or worse.

Our model aimed at helping hospital administrators and government officials understand when demand on health system resources will be greatest. To learn more, see our estimation updates page.

What kind of model does IHME use to produce these estimates?

IHME uses a hybrid modeling approach to generate our forecasts, which incorporates elements of statistical and disease transmission models. Our model is grounded primarily in real-time data instead of assumptions about how the disease will spread. We update our model frequently to respond to new data and new information, describing the changes on our estimation updates page. More detail on the prediction models can be found in articles published as preprints here and here.

Why are you modeling three different scenarios?

In making these forecasts, we aim to help policymakers plan for the days and months ahead, and take action to change the course of the pandemic for the better. We do not know which course of action policymakers will choose, so we decided to model the impact of three different courses of action that policymakers may be evaluating. For more on the scenarios see the question, “What assumptions are included in each scenario?” above.

What are the advantages of your modeling approach?

Our model:

Shows how different policy decisions can impact the trajectory of COVID-19
Incorporates data on deaths, hospitalizations, and cases adjusted for scale-ups in testing and populations tested (i.e., symptomatic individuals and active case detection efforts among high-risk populations in factories, prisons, nursing homes, and homeless shelters)
Factors in mobility, population density, self-reported mask use, seasonal patterns of pneumonia (these patterns closely mirror transmission of COVID-19), and self-reported contacts to understand transmission of the virus
Relies primarily on real-world data instead of assumptions about how the disease will spread
Takes into account variation in transmission across locations and over time
Makes sense of data that fluctuate frequently
Performs well for smaller epidemics

You have expressed concerns about disease transmission models overestimating the number of likely COVID-19 deaths. Why are you now using one?

Our current model is not a disease transmission model. It is a hybrid model that combines both a statistical modeling approach and a disease transmission approach, leveraging the strengths of both types of models, and scaling the results of the disease transmission model to the results of the statistical model.

Why aren’t you using the hybrid model to make projections for Ecuador and Peru?

The number of reported deaths due to COVID-19 in Ecuador and Peru appear to be improbably low. Instead of using reported COVID-19 deaths, we are approximating deaths from COVID-19 in these countries. To do this, we used the number of excess deaths occurring in Ecuador and Peru during the COVID-19 pandemic along with observations from other countries (13 European countries and the US) about the number of excess deaths that were caused by COVID-19.

Why does uncertainty interval (range of values) shrink to zero for future estimates? As the mean estimate (the point where it is equally likely for the estimate to be above or below) approaches zero, the uncertainty interval automatically shrinks because a range of values below zero is not possible.

About the project

Why did you decide to produce these forecasts?

We were initially asked by colleagues at the University of Washington School of Medicine to develop models to help in planning their response to COVID-19. As other hospital systems and governments around the world reached out for help in determining the impacts of COVID-19 on their health systems, we developed forecasts for all 50 US states, and are continuing to add other regions of the world facing similar questions about COVID-19.

Where does IHME obtain its data?

These forecasts include data from local and national governments, hospital networks and associations, the World Health Organization, third-party aggregators, and a range of other sources. We use the Johns Hopkins University (JHU) data repository on Github to collate daily death numbers due to COVID-19. We supplement this dataset as needed to improve the accuracy of our projections. For example, we use data from government websites for Brazil, Canada, France, Germany, Italy, Japan, Mexico, Spain, United Kingdom, and the US states of Hawaii, Indiana, Illinois, Kentucky, Maryland, and Washington. For New York, we use data from the New York City Department of Health and Mental Hygiene and the New York Times GitHub repository. We obtain subnational data from government websites. We obtain testing data from Our World in Data, COVID Tracking Project, and government websites. Our models are updated regularly, as new data are available, to provide the most up-to-date planning tool possible.

For a complete list of our supporting organizations, please see our Acknowledgements page.

Why are you only making projections through October 1?

Going forward, we will forecast four months into the future, updating the timeframe for the forecast at the beginning of each month. We believe that projecting four months into the future will allow us to strike a balance between providing a planning tool for policymakers, while producing the highest-quality results.

What case data do you use in your models?

We use the Johns Hopkins University (JHU) data repository on Github to collate daily COVID-19 case data. We supplement this dataset as needed to improve the accuracy of our projections. For example, for Washington state and New York, we use case data from the New York Times. We use case data from government websites for Japan, Brazil, Canada, France, Germany, Italy, Mexico, Spain, and the UK, and for the US states of Hawaii, Indiana, Illinois, Kentucky, and Maryland.

Since you use case data in your models, are you accounting for greater case detection as locations scale up testing?

Yes – we adjust trends in cases to account for scale ups in testing.

Why are you no longer using temperature in your models?

We have found that seasonal patterns in the transmission of pneumonia better predict transmission of COVID-19 than temperature.

What population density data do you use in your models?

We use gridded population count estimates for 2020 at the 1 x 1 kilometer (km) level from WorldPop.

Can you make estimates for my country/county?

We have heard from many hospitals, government officials, and other organizations asking us to generate projections for other locations. We plan to add different locations to the tool, including additional countries, as well as major metropolitan areas in the US. Based on our current resources, we cannot meet all of the requests we are receiving. However, you are welcome to submit requests by completing our Inquiry Form, and we will evaluate whether we have the resources to address them on a case-by-case basis.

Where can I find explanations for changes in specific forecasts?

Regular updates on changes to the underlying data and model structure accompany each iteration of the model. These can be found on our estimation updates page.

Where can I download previous versions of the projections?

You can download previous versions on our estimate downloads page.

The tool isn’t loading correctly in my web browser. What should I do?

Troubleshooting tips include:

Use the latest version of Chrome, Internet Explorer (currently IE11), Edge, or Firefox.
Make sure to disable compatibility mode in Internet Explorer. In some cases, it renders a site unusable
Use Ctrl+0 to reset the browser zoom. Sometimes having the browser zoomed in or out interferes with the layout or a few mouse-dependent properties.
If the visualization isn’t loading or looks distorted, try clearing your cache. See http://www.wikihow.com/Clear-Your-Browser's-Cache for pointers.

Where may I find the technical appendix for your projections?

For technical appendices, see the supplementary appendices of our COVID-19 papers, which are located here. On April 21, we released detailed technical appendices with the article “Forecasting the impact of the first wave of the COVID-19 pandemic on hospital demand and deaths for the USA and European Economic Area countries.” Changes to the model are also regularly communicated through our estimation updates page.

How frequently do you update your projections?

We aim to release as frequently as possible and will be working to make sure the model reflects what we’re learning each day about the pandemic, and that our forecast reflects the most up-to-date information available from all locations we track.

Can I use your projections in my report/publication?

Please refer to our terms and conditions of use page.

Why did you publish estimates for the US, Europe, and Canada before creating estimates for the rest of the world?

Our goal is to produce COVID-19 estimates for all nations as quickly as possible to assist decision-makers. We were initially asked by colleagues at the University of Washington School of Medicine to develop models to help in planning their response to COVID-19. As other US hospital systems asked for our help, we recognized the need to develop forecasts for all 50 US states. After producing estimates for the US, we then expanded the analysis to Europe since we found readily available data and were already using data from Italy and Spain in our model to understand the progression of the epidemic. We have now published projections for most countries around the world reporting 50 or more COVID-19 deaths. We will publish findings for other countries as soon as we adapt our models to best reflect and respond to the data available and the needs of these contexts. Also, in order to be useful and relevant to policymakers and those planning responses to COVID-19, models for countries with more complicated data landscapes require different approaches than those with robust and publicly available records of deaths.

Coming up next in our research pipeline, we plan to release projections for other countries as soon as possible. We are focusing on expanding our forecasts to locations that have reported a minimum of 50 deaths to date since estimation in locations with few reported deaths can be less reliable.

Why is Puerto Rico listed separately from the US in the drop-down menu in the data visualization?

In our work at IHME, we measure outcomes for territories all over the world. In our drop-down menus, we list these territories at the same level as nations. To be consistent, we have used the same approach for Puerto Rico.

COVID-19 resources