What Makes The COVID-19 Mortality Forecasts Upon Which The White House Relies Seem So Low

What Makes The COVID-19 Mortality Forecasts Upon Which The White House Relies Seem So Low

I read an interesting article regarding the modeling being used by the US Federal government (and some state governments) and how realistic it might be at predicting the magnitude of the current epidemic. An important question to answer when debating how and when to make policy decisions such as opening the economy back again.

According to the model being used by the federal government; this epidemic will claim the lives of roughly 68,000 people and be in its conclusion around August.

The forecasts of U.S. COVID-19 deaths, upon which the White House relies, imply that by August 4 the outbreak will be fully contained. By then, daily death rates will have dropped to zero, and total deaths will have reached 68,000. These same forecasts tell us that one month from now, the outbreak will be 96% contained, meaning that of the total eventual U.S. deaths from COVID-19, only 4% will occur after five weeks from now.

The source of the mortality forecasts just described is the University of Washington's Institute for Health Metrics and Evaluation (IHME)

How does the IHME model, which is being used by the Feds, work?

IHME Forecasts

According to the IHME, the peak for new or daily deaths in the U.S. occurred on April 12; and on April 9, the forecast for April 12 was 2,212 (revised down from an earlier estimate of 3,130). Actual new deaths appear to have peaked at 2,087, and then declined for the next three days. 

During the c-Span interview, Professor Mokdar explained that the IHME methodology for projecting deaths is based on models that are different from most other research groups, because of IHME’s emphasis on fitting the patterns of daily mortality observed in the experiences of other geographic areas such as Wuhan, Italy and Spain.

the research methodology employed by IHME implicitly builds in the containment measures being followed by the population, be those measures voluntary or government mandated. As a result, the number of deaths forecasted by the IHME model factors in the intensity of those measures.

Per the above the IHME attempts to use real world information to make its predictions. It pulls in information in how the virus out break has worked in other countries and tries to account for mitigation similarities and differences to predict what might happen in the US given the measures being taken.

This is in contrast to the more traditional way to tract a disease; which is a more top down approach of mathmaticaly calculating the expected death rate based on the size of the population and the given transmission and death rate. Mitigation efforts are factored in later to bring down the over all death count.

Herd Immunity

Standard epidemic models, such as the Kermack-McKendrick model from 1927, emphasize that in the absence of vaccines and treatments, epidemics only conclude when herd immunity is achieved in the population.

methodologies such as those used by researchers at Imperial College London, herd immunity is not a salient part of IHME communications. The Imperia l College approach begins with herd immunity and then analyzes changes that result from containment measures. The IHME approach instead appears to focus on asking how behavioral patterns observed in Wuhan, Italy, and Spain carry over to the U.S.

Using the mathematical approach the blog author tries to calculate what the transitions rate would have to be for the two methodologies to coincidence with each other.

And what he concludes is that the transmission rate would have to be artificially brought down to around 1.04, through mitigation such as isolation, in order for the IHME prediction to match the heard immunity modeling.

Computations For Herd Immunity

For COVID-19, there are varying estimates of R0, ranging from 2.2 at the low end to 5.7, with the upper end of a 95% confidence interval being 8.9. There are also varying estimates of the death rate per infection. Evidence from the Diamond Princess, suggests a death rate of 0.99%. Economists Eichenbaum, Rebelo, and Trabandt report a similar statistic from South Korea, adjust the rate for age, and arrive at a value of 0.5%. The size of the U.S. population is approximately 330 million.

Taking the “best” case, with R0 equal to 2.2 and the death rate per infection equal to 0.5%, the estimated total number of deaths in the U.S. from COVID-19 turns out to be 900,000. This is 13 times as great as the IMHE estimate.

Is a forecast of 68,841 eventual deaths unrealistically optimistic? To answer this question, consider what transmission rate is associated with that number of deaths. It will be below 2.2, but how much below?

The answer is 1.04, barely above the threshold of 1.0 associated with a zero infection growth rate: that is, 1.04 is a lot below 2.2! 

A transmission rate of 1.04, through mitigation, does sound optimistic.

But than again the herd immunity calculations assume some very rigged factors; when the transmission started, the natural transmission and death rates. Factors that are just guesstimates at present and won't be fully know until far after we are over this pandemic.

In that regard the IMHE model may have some superiority. It does not rely on rigged use of guesstimated factors; but attempts to input real world information and than fit its modeling to it...

...assuming the information being fed to it is correct; which is also in question with the way China is reporting and now the growing evidence that other states and countries are manipulating their data up and down in order to frame it in the best possible light for the polititions involved.


The blog author also spends some time setting up a test strategy to track how well the IMHE model preforms over the coming weeks and months.

We can't trust the numbers to use any formula effectively. We scoffed at China's number's but I'm sure very few countries have the correct totals regardless if it's intentional or not. About the only semi-reliable number would be deaths, but even those likely have additions that are highly questionable cases. Modestly let's say 5 percent.

I learned a while back that if you want to get someone to believe something is real that is not real, a good aid for accomplishing that is graphs and other visual aids. There was research on that. Graphs and charts are used way more to fool people than to teach people something true.

The evidence they are using to create their graphs is not correct, so it is not going to show a real result with this social isolation crap.

If you got a bug like this going around, everyone should have put a covering over their face, even a bandana would have worked. also social distancing is a good idea, big crouds should not be allowed for a period of time. They did not have to do the stay at home crap. They did not have to jeopardize the economy. If many people are having just mild symptoms, their immune system knows how to attack it and kill it. Exposure to some other similar microbe trained their immune system to respond and they only showed mild symptoms. It would have been better to protect the people at risk better, giving those ten percent unemployment for a couple of months would be better than what they did. I have masks, I would wear them if others did, I have been wearing minee when I go to the store now since lots of people are doing it, the government should supply pharmacies and stores with masks to give out if people want them . They do not have to be expensive masks, those ten cent masks can stop most of the virus transmission. People with immune deficiencies could wear better ones, but those should not cost more than a quarter.....they really overcharge for some masks. They could sell reusable cloth masks cheaply, they do not need to be new every time. That should be happening more already and in the future, it could slow progression of lots of different diseases.

What Makes The COVID-19 Mortality Forecasts Upon Which The White House Relies Seem So Low

What , do you wish it were higher?

I like the way HUD Director Dr. Ben Carson put it during an interview this week.

"It's simple why the models got it so wrong Martha. It's the Garbage-In = Garbage-Out principle on full display here".

a reply to: carewemust

POPULATION DENSITY
Models don't really account for population density or a variety of mitigation factors and completely ignore data collection errors (it's reported that CDC has enacted policies that promote data collection errors for COVID-19).

Population density in NYC/NJ is over 50,000 / sq. mi. in some places, yet the State of New York weighs in at 417/sq. mi. Texas has 107 people per sq.mi. - but neighboring Dallas weighs in at 18,000 / sq.mi.

More people available = more testing, so high density urban populations are over-sampled while less-densely populated areas - states like AR, AZ, NM, TX, UT... - have little effect on predictive models but contribute to post-analysis adjustments.

It's not JUST GarbageIn-GarbageOut the actual models are tinkertoys the research guys pull off the shelves and play with to make pretty (deceiving) pictures. Note the economists commenting on models, they usually have superior math skills to most medical guys.

ganjoa