Thursday Throughput: Missing COVID Deaths Edition

Derived from NIAID image.

[ThTh1] Over the last few weeks, we have a seen sudden and alarming rise in confirmed COVID-19 infections, to the point where many states are instituting mask requirements, pulling back on reopening plans and considering new stay-at home orders. I’m not particularly interested, at this point, in figuring out who is to blame for this. I suspect, like most things surrounding the virus, there are many factors at play. But there seems little doubt that we are now in the grips of a resurgence, with 38 states or territories seeing an increase in cases over the last two weeks, 16 seeing a stable case load, and none seeing a decrease.

Or are we? Proponents of keeping the economy open are countering that the recent rise in cases is due to increased testing, not an actual resurgence of the virus. President Trump has been particularly vocal on this point. This belief has faded as the number of cases has skyrocketed and the percentage of positive tests has grown. But one point the diehards keep hitting is that there has been, so far, no corresponding rise in COVID-related deaths. The number of COVID deaths should be relatively insensitive to testing, so why aren’t they rising if the virus is really surging?

Well, first of all, people don’t die the second they are diagnosed with Coronavirus. The typical time between when a case is diagnosed and when a death is reported is about three weeks. And we are just about at three weeks since the new surge began. Moreover, that delay between confirmed test and confirmed death may be increasing right now because of increased testing. We are diagnosing cases faster and earlier than ever. That leads to a problem called lead-time bias, in which you mistake earlier detection for better outcomes.

The second reason is that the new surge — so far — is hitting a younger demographic. The first hit older Americans hard — nursing homes in New Jersey and New York were literally decimated. Older people are far more likely to die from Coronavirus than younger people. So while we may be seeing a rise in deaths from the new surge, it is being countered by a decline in deaths from the first surge, at least for the moment.

The third reason we are not seeing a rise in deaths (yet) is that COVID-19 is probably less lethal than it was three months ago. In the early days of the pandemic, we had no idea what we were dealing with. Since then we have discovered that the antiviral remdesivir has a positive effect if given early in the disease. The cheap steroid dexamethasone may cut the fatality rate by as much as a third for the most desperately ill patients. We’ve also learned that we were overusing ventilators during the early months. The use of a ventilator, as Andrew can tell you in detail, comes with risks and potential complications. The air pressure can be high enough to damage the lungs. Doctors are now being a lot more cautious in using them (which is why we don’t have a ventilator shortage right now). I’ve heard from several critical care doctors and nurses that they are now only being used if a patient is in critical danger, with supplemental oxygen and Non-Invasive Positive Pressure being favored if possible.

This is part of the bigger issue which is that we know have thousands of doctors and nurses that have seen dozens or hundreds of COVID cases. My father is a surgeon and he likes to say that medicine is as much art as science. You can know all the textbooks just fine. But a lot of the time, you have to rely on instinct and experience to know what’s wrong with a patient and how to treat them. Four months ago, that experience did not exist for COVID-19. Now it does. And thus our medical personnel are much better equipped to handle COVID cases and keep patients alive. They have the instincts to know when a patient is in serious trouble and when they’re not; when they need a ventilator and when they don’t; when to use steroids and how much to use. All of this adds up to better overall outcomes.

But there’s one other factor: we are still seriously estimating just how bad things were in March and April.

Everyone with a spreadsheet program and a little knowledge is producing COVID graphs these days, so I thought I’d give it a whirl, using data from the COVID tracking project.¹ The problem with tracking the progress of COVID-19 is that, even now, we don’t detect all the infections or even the majority of them. And this was especially true in the early months of the disease when testing was woefully inadequate. So on any particular day, the number of new COVID infections is some convolution of the number of new infections and the number of tests we’re doing.

There is, however, a much more solid number: deaths. While there are reasons to believe we are undercounting the number killed by COVID, the number is much more stable since a death is a recorded event while a test is a maybe thing. So what I did was take the number of deaths on any particular day and calculate how many infections led to those deaths, assuming that it takes three weeks from infection to a recorded death and that the fatality rate is 0.7%.² Here is what the infection curve looks like, back-projecting from deaths.

The project number of COVID-19 infection, given recorded deaths. The X-axis measures date since February 1.

The data go through mid-June and they show just how bad the initial wave of infections was. At its worst, we may have been getting as many as 200-300,000 infections a day. But because testing was so inadequate, we didn’t see it very clearly. We knew it was bad; but we didn’t quite know how bad.

Here is another plot comparing the number of projected infections against the number of confirmed ones. This is basically the percentage of cases we were detecting.

A crude estimate of the fraction of COVID infections confirmed by testing.

It’s certainly no secret that we weren’t detecting most COVID-19 infections early on. But the sheer scale of the problem is hard to grasp. For several weeks, the epidemic was raging invisibly, infecting millions while only thousands of cases were being confirmed. We are now detecting about a third of cases, possibly higher. That might be as good as it gets since many people have no symptoms or mild symptoms and never get tested.

Let’s use the deaths as a measure of the number of infections up to three weeks ago. Then we’ll switch to the number of actual confirmed cases, assuming a 30% detection rate. This hybrid plot looks like so:

Hybrid estimate of number of COVID-19 infections.

Now I would not take this plot to the bank. It is rife with assumptions, many of which are known and are throwing the analysis off. The age demographic has changed, which makes the use of deaths to back-project infections increasingly inaccurate. But then again, our treatments have improved so we’re likely getting more infections for every recorded death.

In the end, however, with a large helping of salt, I think that curve gives an accurate if blurry picture of what has happened: the initial wave of infections was massive, much worse than it looks if you only use confirmed detections. But the resurgence is bad and getting worse. This is not just testing; this is an actual resurgence of the virus. Ignore the numbers on the Y-axis; the shape is what we need to be thinking about. ³

So, to bring it all home: should we expect deaths to start rising? Maybe. They are already leveling off, indicating that the pandemic is still going. If I had to guess, I think we will see a rise to about 1000-2000 deaths a day. Not as bad as the initial outbreak, but still bad. However, this will be short-lived if we enact the right policies — banning mass gatherings, wearing masks and social distancing with stay-at-home orders where the outbreak is at its worst.

By then, however, we’re getting into fall and winter. And that’s the time everyone’s afraid of: when we may get a real second wave that will make the first one look mild.

Of course, by the time this post goes live, the point may be moot. The first two days of this week saw a sudden and sharp uptick in reported COVID deaths. It’s only two days worth of data so far, so I would be reluctant to interpret it too aggressively. But it may indicate that the honeymoon is over and the carnage about to get worse.

And it always bears mentioning that even if we have brought deaths down, COVID is still a serious and possibly debilitating disease. For every person who dies, another dozen have a potentially life-altering course of the illness and another few dozen get seriously ill for weeks. When the 1918 flu pandemic hit, the survivors, including those in utero, had shorter lifespans because of the long-term health impact. So while the lag in deaths is interesting, let’s not lose sight of what’s really going on: hundreds of thousands of people getting very very sick. And in a way that might have been preventable.

It didn’t have to be this way. Other countries have managed to beat this down to a hundred confirmed infections and single-digit deaths every day. The United States produces unique challenges, of course, since we’re a far-flung country of fifty states. But I can’t help but think that with better leadership, better testing and less politicizing…I might not have had to write this post.

[ThTh2] Speaking of COVID-19 … this video uses an engineering technique to show why masks are so effective in slowing the spread of the virus. Wear them. Please.

[ThTh3] And continuing on our theme of disease…we now know that malaria is spread by mosquitos. But in the 19th century, that belief was regarded as mistaking correlation for causation.

While researching my book, I was struck by this quote by British geographer Richard Burton. Written in 1856, it dismissed suggestions that mosquitoes were linked with malarial fevers.

It would be several decades before European researchers finally made the same connection. pic.twitter.com/VBDuRttMDl

— Adam Kucharski (@AdamJKucharski) July 1, 2020

[ThTh4] Some months ago, we talked about Betelgeuse fading and whether it meant an explosion was imminent. We now have a possible explanation for the fading: giant sunspots. It is well-known that stars like Betelgeuse can get truly massive starspots — we’ve actually seen them in images of Betelgeuse. So this is one possible explanation for what happened.

[ThTh5] Numbers so big they can create a mental black hole? Well, only if you had a brain big enough to know all the numbers (H/T: Crazyfillyjonk

[ThTh6] Wait. Massive stars can just disappear? Well…maybe. A very luminous star in a nearby galaxy has become too faint to detect. There are a number of explanations: the star may have exploded behind a thick shroud of dust; we may have simply missed the explosion; or the earlier detections were of a very bright outburst that has since faded. But one possibility is that it by passed the supernova phase. Normally, when a massive star runs out of fuel in its core, that core implodes, crushing down to either a neutron star or a black hole, depending on how massive it is. The upper layers of the star crash down on the newly-formed remnant and then rebound, creating a supernova explosion that can outshine an entire galaxy. But if the star is too massive (not not quite massive enough to blow itself apart), there may not be enough energy for the shock wave to break out and create the intense light we see. Instead, that energy is pulled into the black hole.⁴ (H/T: Em)

[ThTh7] This has been a good week for the sky. Comet NEOWISE is visible. And we had a nice alignment of the planets. If you ever spot two planets and the Sun, you can draw a line through them that define the plane of our Solar System. Whenever I see it, something turns around in my head and I feel like I’m looking at the Solar System from the outside. And the scale of is is humbling.

Jupiter and Saturn are so bright tonight I can get a picture of them with my freakin phone! And they're lined up perfectly with the Moon.

(Moon not to scale) pic.twitter.com/uh1jaTM3qg

— Kojac! (@kojachusky) July 5, 2020

1. I apologize in advance for the poor quality of my plots. I usually use professional software for this sort of thing and Excel and I … get along OK, but not great.
2. The assumed IFR and time to death don’t actually matter very much; the absolute numbers changes but the overall result is unchanged.
3. Several other outlets have done this kind of analysis using different techniques. They all get roughly the same result that I do.
4. This really stretches my understanding of stellar physics, so I may be mis-stating some things here.