More On the IHME Study

April 1, 2020 11:33 a.m.

I wanted to add a bit more about that IHME study – some more perspective on what it is and isn’t telling us. Before I do, a quick point. Last night an exasperated TPM Reader FT wrote in saying in so many words all the debate about models is morbid. What are we accomplishing by debating technicalities about why a horrid number of Americans are going to die versus twice a horrid number? And if we’re saying it’s so bad in advance what incentive does anyone have to social distance?

First, this is all overwhelming stuff. I’m sitting here this morning with the news that the number of people who die in this calamity is likely to be counted in the hundreds of thousands. I’m simultaneously numbed and overwhelmed by it. We all have to pace ourselves. And we all have to be gentle with ourselves and others around us. One of those things is taking some breaks from the news when we need to.

On the specific question, these models aren’t the equivalent of watching polls, where they’re basically just scrambling efforts to predict the future. These models are the only maps we have for hospitals, governments, front line first responders to plan their response. They are imperfect but critically important. In covering the news it is critical to figure out whether the planning is being made on the basis of the best science. That is the main reason for our focus. But even that isn’t the whole story. We have a human need to understand what is happening to us, to our society, even for those of us who are mainly – we hope – observers rather than participants.

With that said, here’s a follow up by TPM Reader JL, a academic from a relevant background from the West Coast on the IMHE model we’ve been discussing and which now appears to be central to the White House’s efforts …

I’ve had time to look read over the IHME study and think about it and have some thoughts that might address concerns in the TPM site and twitter.

What the study says: I think that if you read the study, the best guess is that it is talking about what the initial epidemic curve will look like, and what cases, deaths, demands on health care system can be expected, IF the maximal stringent social distancing measures are put in place, AND MAINTAINED THROUGHOUT the duration of the initial epidemic curve. I think the problem is that this is not stated explicitly in the paper, but you have to figure it out by reading the paper and math appendix carefully, maybe more than once, and reading between the lines. And I think my conclusion is consistent with the Abraham Flaxman tweet I just saw in the @joshtpm twitter feed.

Is the ‘curve fitting’ approach OK? I think good bet it is, given the place most US states are in the course of the epidemic. The only tractable way to generate the true epidemic curve that is generated by the mathematical modelling of the true biology is to let the differential equation model run and write down the level of infected, susceptible, etc. that it produces. The math of the ‘true’ curve cannot be reduced to a dependent variable you want to see on the left hand side of the equal sign, and variables that produce it on the right hand side. If you let an outbreak turn into a full blown epidemic, you are going to ride on the same general shape of the true epidemic curve, no matter what you do. All you can do is squish down the same general shape of the curve through some drastic control action. Mathematical epidemiologists and applied mathematicians have developed a library of different curves to use in different situations. This modelling is over 100 years old now, goes all the way back to Ross and malaria control around 1900, so what is reasonable is well understood.

What are the weaknesses of the IHME approach?: (IMHO!) It is only really good for trying to understand the course of an epidemic where there is no two way feedback between different types of control efforts, and the behavior of the epidemic. So, it doesn’t capture the OODA loop that the humans and virus population will be playing with each other after the epidemic gets past the peak. Bottom line, doesn’t really tell us what to do after we get through the crisis of handling the epidemic peak.

Is not ‘modelling the biology’ a valid criticism?: (IMHO!) No, not for understanding what will influence the epidemic in current crisis mode when we have to understand behavior of epidemic until shortly after peak occurs. What the ‘true biology’ is cannot be answered. All models are false, but some are useful. For understanding drug treatment of AIDS in different types of cases, you have a multiple population epidemic of different strains of the AIDS virus praying on cells inside a type of AIDS patient, for example. For other purposes, you might need to model covid-19 as a predator prey model, with multiple strains of the virus competing for opportunities to prey on the human population.

Bottom line: (IMHO!) the IHME is only useful for understanding what maximum social distancing policies will probably get us in current maximum crisis stage. One of the assumptions behind the analysis in terms of understanding the whole epidemic curve is unrealistic: that the maximum social distancing measures can be maintained THROUGHOUT the epidemic. I don’t think they can. We have to transition to other modes of control after the peak, and the IHME model is not built to answer that question.

Support The TPM Journalism Fund
  • Contributions allow us to hire more journalists
  • Contributions allow us to provide free memberships to those who cannot afford them
  • Contributions support independent, non-corporate journalism
Masthead Masthead
Founder & Editor-in-Chief:
Executive Editor:
Managing Editor:
Senior Editor:
Special Projects Editor:
Editor at Large:
General Counsel:
Head of Product:
Director of Technology:
Associate Publisher:
Front-End Developer:
Senior Designer: