Written by John Furlan
Taming the Disease
Data Limitations of Epidemiological Models Make Projections a Moving Target
According to the IHME model I mentioned above, it now shows peak deaths per day of 3,130 on April 16, ten days from now, the same day as peak hospital resource use. Like national total deaths, these also are probably down from earlier estimates, though I don’t have the data to confirm that. This estimate is nationally, there are wide variations in projections by state available on the site, which I discuss below.
Please share this article – Go to very top of page, right hand side, for social media buttons.
Again, I can not over-emphasize strongly enough that these “COVID-19 projections assuming full social distancing through May 2020.” Presumably these numbers would be higher, perhaps significantly so, without that assumption, though I don’t see a model showing that case on the site, I may have overlooked it.
These new IHME projected U.S. total deaths of 82,000 are a very far cry from the millions of deaths from some modelers in earlier headlines, e.g. more than 2.2 million by the Imperial College model in mid-March, IF no action taken, which presumably that model then helped catalyze, see this chart from an April 2 Nature article.
As the estimates from the various credible models by experts like IHME come down, it’s important to emphasize that none of these models have been deliberately misleading, as perhaps some on the right seem to imply in order to further expand “big government.” Rather, the modelers are just working with very limited data, as I mentioned in my first article on the subject on Mar 23, and as the Fauci quote above indicated.
Yet you wouldn’t know this from the media coverage, which often report the projections of various models as the best guide for policy, rather than something that must be taken with the proverbial huge grain of salt for obvious data reasons, as Fauci said.
The modelers, extremely smart, highly educated professionals, obviously know and acknowledge this issue. From the April 2 Nature article mentioned just above, titled “Special report: The simulations driving the world’s response to COVID-19,” which I very strongly urge you to read.
John Edmunds, who is a modeller at the LSHTM says:
“You can project forwards and then compare against what you get. But the problem is that our surveillance systems are pretty rubbish. The total numbers of cases reported, is that accurate? No. Accurate anywhere? No.”
Only Serology Antibody Testing Can Overcome Models Data Limitations
A major issue with the lockdown policy approach has always been the lack of useful data, especially on the reported case fatality rate, which is reported deaths divided by reported cases. Estimates of that have often seemed too high, because the denominator does not include people who have had Covid-19 but were not reported.
That huge data shortfall can only be changed with a serology test for Covid-19 antibodies given on a random basis to various populations, both national and locally. Such a test is analogous to public opinion polls, which have been done for decades. The first results of such tests may be available in 2 – 4 weeks. Again, from the April 2 Nature article:
“But some crucial information remains hidden from the modellers. A reliable test to see who has been infected without showing symptoms – and so could be moved to the recovered group – would be a game changer for modellers, and might significantly alter the predicted path of the pandemic, says Edmunds.”
The testing that has been available although on a far too limited, criminally negligent basis is called PCR, polymerase chain reaction, which is used to detect bits of viral genetic material, it can show if someone with symptoms has COVID-19 or the flu, but it cannot show if someone has had COVID-19, indicated by the presence of antibodies, which is what serology antibody tests can do.
The media has focused on the PCR test to deal with the intolerable situation of people with mild symptoms being turned away from testing due to its lack of availability, and to the long wait for test results when they are given. The months-long failure to address and resolve this issue has already destroyed Trump’s credibility on COVID – 19, perhaps irreparably.
For a medical view of the serology antibody test issue, read the March 24 WSJ article by two Stanford professors titled “Is the Coronavirus as Deadly as They Say? Current estimates about the COVID-19 fatality rate may be too high by orders of magnitude,” in which they write:
“Given the enormous consequences of decisions around Covid-19 response, getting clear data to guide decisions now is critical. We don’t know the true infection rate in the U.S. Antibody testing of representative samples to measure disease prevalence (including the recovered) is crucial. Nearly every day a new lab gets approval for antibody testing, so population testing using this technology is now feasible.”
They also write:
“Epidemiological modelers haven’t adequately adapted their estimates to account for these factors,” and before that cite “high estimated case fatality rate – 2% to 4% of people with confirmed Covid-19 have died, according to the World Health Organization and others.”
I think this may be misleading, though not deliberately so, in that those high estimates were thrown around shortly before the time of the WSJ article, but I highly doubt, without knowing for sure, that reputable modelers like IHME are using rates above 1% now. Still, there is a huge difference between 1% and a flu-like 0.1% or even lower, which the authors of the WSJ article suggest is possible.
If you think the WSJ is too conservative, free-market biased, then check out the March 31 YouTube interview of one of the article’s authors, Jay Bhattacharya, he comes across as a humble, humane, caring, compassionate person, much more so than his right-wing Hoover Institution interviewer, whom he happens to be the physician for.
Frankly, without invoking a conspiracy theory, it just may even have been both socially necessary and politically expedient to have such high initial case fatality rate estimates used in models to get the social distancing policies implemented.
But we are now well past the point where that may have been the case. It’s long past due for much better data to guide policy. Whether or not Trump will use it is a whole other question, which I think you may know the answer to.
Part 3 to be published.
Photo credit: Vox.com
Adapted from an article on Medium 06 April 2020.
.
