Models and Death Rates

The widely used model from the Institute for Health Evaluation and Metrics has taken a lot of heat and some more is piled on in a Stat article.   (Stat Article)   The article points out a variety of criticisms, including that it isn’t a traditional epidemiological model.  It was fundamentally a “curve-fitting” exercise which basically assumed the course of the epidemic, including with mitigation measures, will be similar to the course in other countries, and then modified those projections as US experience developed.    Their projections have large uncertainty ranges, but so do all models.  In fairness, the IMHE modelers had a different goal from the start and were working with the same lack of data to inform assumptions that everyone else was.  They were more taking a snapshot in time and trying to help people see if there were adequate health resources to deal with likely cases.  Where I would most fault them is poor communication about what they were doing, how inaccurate the results were likely to be, and therefore that they shouldn’t be relied on for decision-making.  They have said they are working on a more traditional SEIR model (I will explain that in more detail in a future post).  So we will see what that looks like.  What I am most concerned about is that a number of people are still relying on models that seem to assume that the epidemic peters out in the next few weeks.  I don’t see how that happens.  We have to loosen up current restrictions, but we need to be realistic that doing so will mean more infections, serious illness and deaths.  Models should run out for at least a year to capture all those effects.

And given the fixation on deaths, one of the most difficult aspects of the modeling is capturing death rates.  And in real life, there have been concerns about the attribution of deaths to coronavirus, both that some are being missed and some are being mis-attributed.  A New York Times article explores some of these issues.   (NYT Article)   Problems with the raw number of deaths include identifying people who died at home, and all deaths with coronavirus present being treated as due to coronavirus, even though other conditions, especially serious chronic ones, were likely to be as much or more a contributor.  Aside from deciding if a person died because of coronavirus, the raw death numbers feed into people’s estimates of metrics like the infection fatality rate (“IFR”) or the case fatality rate (“CFR”)  or the one I think is most meaningful, but really has to wait to reveal itself, the per capita rate.  While rates vary widely across the world, what might be thought of as the CFR appears to currently be around 6.4%.  But this is only identified cases, which means ones that were usually serious enough for the patients to seek medical attention.  It is a highly misleading rate.  Everyone believes there are multiples of people infected compared to actual “cases”.  If you think there are ten unidentified infections for every identified one, than the CFR or IFR falls to .64%.  If it is fifty unidentified for every identified one, that rate drops very dramatically.  Emerging evidence suggests that there really are an extremely large number of unidentified infections.  I anticipate an actual IFR of around one-tenth or two-tenths of a percent, at most.  And that rate is extremely skewed toward the elderly.  We have a lot of people very anxious because they are convinced they are going to get the virus and die.  Most people are not at risk.  And policymakers are over-reacting in part because they have been spooked by far two high estimates of death, so better data may help them dial back on shutdowns.