A little more about “On testing.”

Here’s a response to some additional questions/comments I’ve gotten on my original essay and the follow-up.

11. “Wait, so what are you saying are the takeaways from that Harvard team’s modeling data?”

I’m sorry I wasn’t sufficiently clear when I wrote about the Harvard team’s data.  The two big take-aways from the graphs that are Figure 4 and Figure 5 in the paper are the height of the cumulative incidence curve (the total number of critical cases) and the slope (the rate at which people seek care in the hospital).

• 

• 

When the researchers model the timecourse of critical cases for a one-time epidemic addressed with a one-time shutdown, the shutdown produces either little or no reduction in the eventual number of critical cases, and the rate at which critical cases appear is only slowed (preventing hospitals from being overwhelmed) if we do a mediocre job of it.  Otherwise we preserve a large pool of unexposed people, and when the shutdown ends it would be as though that were simply the new start date of the epidemic (look at the steep slope of the green curve in each graph of Figure 4).

But the shutdown can extend the epidemic for a long time, which can cause Covid-19 to become a seasonal burden.  These infections would become seasonal if there were:

1.) seasonal changes in the rate of transmission

2.) genetic drift sufficient to change the viral surface, at which point your old antibodies won’t bind

3.) waning immunity

4.) a non-zero pool of maintained infections

#

My gut reaction is that Covid-19 would have seasonal changes in the rate of transmission – the rate of transmission of influenza when people spend more time clustered indoors.  We’ll know more about whether or not this is true for Covid-19 if we can get data on the rate of transmission in places around the world (especially on opposite hemispheres), but for now it seems reasonable to assume that human behavior would increase the rate of transmission if there were no intervention like intentional social distancing.  And the Harvard team included this in the model used for Figure 5 (although it’s not clear whether you’d have the same behavioral changes in the midst of a shutdown).

In terms of genetic drift, we’re lucky that Covid-19 contains a proofreading enzyme.  It should change more slowly than other viruses that use an RNA-dependent RNA polymerase.  Making an HIV vaccine has been a nightmare in part because HIV mutates so rapidly; making a Covid-19 vaccine will be easier.

Waning immunity, though, might be a big problem.  In their introduction, the Harvard team cites data showing that some coronaviruses confer immunity that wanes within a year; other coronaviruses confer longer-lasting immunity.

Your immunity to almost any antigen will eventually wane.  This can happen at very different rates – you only need a tetanus vaccine every ten years (and might maintain immunity even longer than that).  By way of contrast, immunity to influenza can wane within months — best practice may be to receive two vaccinations against influenza each year. Please read this excellent review for details.

You might need a Covid-19 vaccine every year (if there is still a reservoir of infectious virus circulating through the population).  We just don’t know.

The Harvard team discusses waning immunity as a motivation for including seasonality in their model, but for Figure 6, where they model the impact of repeated shutdowns, they don’t include waning immunity.  They instead model our population steadily approaching a herd immunity threshold.  This may be unrealistic.  If immunity wanes, there will also be constant downward pressure on our total population-level immunity.  It’ll take longer – and there will be more total critical cases – to reach herd immunity.

There’s another scary result that you can get from constant downward pressure on our population-level immunity.  By staggering infections, you are staggering the times when people will lose their immunity, possibly ensuring that the virus will always be able to find a new host.  By slowing its spread, we could cause Covid-19 to become a constant burden on our species until there is a vaccine.

I feel optimistic that we can develop a vaccine for this within four or five years, but the cumulative infections would continue to rise for this entire time period if immunity wanes.  And that would be bad.

The other big take-away from their model is that, no matter what we do, if the shutdown isn’t permanent, the same number of people are eventually getting sick.  This, too, is bad.  The shutdown, or repeated shutdowns, are saving only the small number of lives that would be lost to triage if our hospitals were overwhelmed.

We can do better.  From the data we have now, it looks like we can save lives by stratifying our recommendations and focusing our efforts on reducing transmission to the most vulnerable people among us.

In my opinion, this is the big reason why we need to see data from large-scale antibody studies in places like New York City.

I expect we’ll see that at least a quarter of the population of New York City has already recovered from Covid-19.  My father (whose guess is probably more trustworthy than mine, since he’s both a physician specializing in infectious diseases and a professor of virology & immunology), expects to see a percentage even higher than that.

And, yes: if we see that a percentage of the population smaller than that has recovered, we should go back to being terrified.  We’ll need to look for other risk factors, like whether air quality has a huge impact on people’s ability to survive Covid-19 infection.

But a big dataset showing which people had successfully recovered from Covid-19 would be really helpful.  As we learn more about who successfully recovers from this, we can try to skew the demographics of exposure so that these people are enriched among the population who gets infected.

The dataset would help us save lives by limiting exposure of precisely those people whose biomarkers don’t resemble those we find with antibodies in their blood.  It’s the same statistical analysis that Abraham Wald used to protect fighter planes during World War II (which is a nifty story if you haven’t read about it).

Because we bungled the beginning of this epidemic, this is probably our best option.  The only ways we have to actually save lives are either to skew the demographics of exposure or else continue the shutdown until there’s a vaccine.  But that’s unthinkable – a three-month shutdown will already cause a lot of deaths.  A five year shutdown would be catastrophic.