On navigating coronavirus blindfolded

In all of my previous posts on COVID-19, I’ve attempted to help readers understand the virus quantitatively to separate fact from fiction. In recent days we’ve seen societal lockdowns all around the world in an effort to combat the virus, which I believe will have catastrophic— and poorly understood — impacts on the world from an economical, sociological, and psychological perspective. I think these lockdowns are a big mistake.

What concerns me most is these drastic policy decisions were based upon a very poor understanding of the actual severity of the virus and where we sit on its epidemiological growth curves. The root of this issue is a dearth of random population-wide testing studies, which are only now starting to emerge. Without an accurate way to measure the prevalence of infections, the world has essentially been navigating coronavirus blindfolded.

Early reports of coronavirus from Wuhan, China, the epicenter, suggested a CFR (Case Fatality Rate) as high as 3.4%. This number got the attention of policymakers around the world, even though it’s essentially meaningless without a firm understanding of the denominator — the correct number of coronavirus cases.

I previously spent over a decade running online market research firms. In our industry, we randomly sample a small proportion of a population to estimate how prevalent certain characteristics are across the entire population. Thanks to the power of statistics, we can do so quite accurately with sample sizes in the hundreds or low thousands, even projecting to population sizes in the hundreds of millions.

Epidemiologists can do this as well, by randomly testing a small proportion of a population to estimate how widespread disease is across an entire population. It’s important to use a random sampling methodology because a biased sample (e.g., only people who visit a hospital) will lead to biased conclusions.

Random population studies are only now starting to occur with coronavirus, and early results indicate that the virus may be far more prevalent, and thus far milder, than was previously thought.

Iceland has recently shared the results of their population-wide testing, which I’ve summarized and analyzed here. Their data suggest a current estimated coronavirus fatality rate of just 0.032%. Granted many of their infections will have health consequences that are yet to be realized since it typically takes 2–3 weeks for an infection to become fatal, but even if you estimate they’ll eventually end up with 4x as many fatalities as today from current infections, the implied CFR is just 0.128%.

Another place where population-wide testing occurred is the Diamond Princess cruise ship, where, as of today, 7 out of 700 infected passengers and crew have died of the virus, implying a CFR of 1%. However, there are some problems with both the numerator and denominator of this equation. The numerator is likely inflated compared to other populations since cruises attract a much more elderly clientele, and the denominator is likely deflated because the type of test used only accounts for current infections, not people who were infected but already recovered. John Ioannidis, a professor of medicine, epidemiology, population health, biomedical data science, and statistics at Stanford University, adjusted mathematically for these factors and concluded that if one projects the Diamond Princess data across a large general population “the death rate among people infected with Covid-19 would be 0.125%”

Notice this is about the same number that I estimated above based upon the recently published Iceland study. For context, the seasonal flu has a CFR of about 0.1%, which would make coronavirus only about 25-30% more fatal if the results of these early population-wide studies hold up elsewhere, and they very well might.

“But what about Wuhan? What about Italy??” you ask.

We’ve seen the terrible human toll in these areas, and my heart goes out to all of the families who lost loved ones, and all of the heroic medical professionals who had to make impossible triage decisions as their hospitals were bombarded with COVID-19 cases seemingly overnight.

However, I suspect looking these numbers, that they were much further along in the epidemic than they thought, and thus much closer to reaching herd immunity.

To help explain, I’ve borrowed the base case scenario epidemiological curve from isee systems’ COVID-19 Simulator and marked it up for illustration:

I didn’t calculate the placement of these arrows on the curve for Wuhan or Italy, they’re merely meant to illustrate the importance of knowing where you are on the curve before taking drastic public policy actions like lockdowns. If the virus is actually 20x more prevalent in a population than once thought, which it would be if you were projecting based on deaths and assumed a ~3% CFR when it’s actually a 0.128% CFR, then you grossly misplaced your position on the curve.

Lockdowns make sense if the alternative is thousands or millions of deaths in a matter of weeks, but not if you’re dealing with a relatively mild virus that is far too prevalent to be contained anyway. Then the appropriate reaction would be to slow it’s spread just enough to avoid overwhelming healthcare infrastructure while it runs its course. Even though coronavirus may not be very fatal, it is very speedy, so aggressive action should be taken to slow and mitigate its impact on high-risk groups (e.g., targeted advisories for at-risk populations, sanitation education campaigns, contact tracing, widespread testing, etc) but not shutting down an entire economy.

If the population studies continue to show many uncounted infections and thus a much lower CFR than originally thought, then coronavirus will turn out to be less dangerous than the draconian measures we’re taking to address it, especially within the context of excess mortality. History would then judge this as a story of media sensationalism, mass hysteria, and catastrophic public policy decisions based upon poor epidemiological assumptions more so than a story of a killer new virus. I sadly suspect that will be the case.