Updated: Oct 26, 2020
Dr. Osterholm of University of Minnesota recently commented that the next 6-12 weeks are going to be the darkest of the pandemic. He notes that we recently had 70,000 cases, the same level as at the height of the July surge. Now, there are several issues with this, which we can see from Figure 1 below. First, 70K cases in July are not the same as 70K cases in October. Why? We have increased our testing capacity since mid-July by 25%--from 800K daily to more than 1 million. The point at which this testing surge started, in mid-September, (which is rather oddly exactly when IHME predicted a second wave. Coincidence? Probably not), both hospitalizations and deaths start to to diverge from confirmed case counts. In Late July, there was no such divergence. More important, is that hospitalizations now are at right around 35,000 hospitalizations. In mid-July, when Dr. Osterholm says we were at the same point where we are now, there were more than 60,000 people hospitalized--which still only accounted for 8% of hospital capacity, according to the CDC. And if you're going to say "but Texas," no, Texas's hospitals were never actually even close to overwhelmed, which I have gone into in excruciating detail in this post, but the short hand is that at Texas's peak, COVID patients were using ~12% of hospital capacity (at the height of the 2017-18 flu in Texas, 25% were in use) and approximately 40% of ICU capacity. It's also notable, that Governor Cuomo recently claimed that NY's hospitals were never overwhelmed, despite NYC hospitals reporting at least 40% of beds in use for COVID patients... OK, if you say so governor Cuomo.
Figure 1: Using Rolling Case Fatality Rate (RH axis) to Estimate Historical Reported Cases, if Testing Had Been Uniform
Source: COVID tracking project. Estimated CFR computed by averaging (1 week of deaths)/(prior week's cases), shown on RH axis. "Estimated cases observed if consistent testing" derived by dividing 7-day rolling average of deaths by 10/19 estimated CFR. Note differences in units, done to allow comparison, cases, hospitalizations are in 10's, tests in 100's, deaths are as recorded).
All this is to point out that we are clearly NOT where we were at the height of the summer, nor does it look like we will be heading in that direction. Might we see a slight hump? Sure. But are we going to see the "Darkest 6-12 weeks of the pandemic"? I highly doubt it. First of all, you have to define when the darkest 6-12 weeks of the pandemic actually were. Those weeks were not in July, they were from the end of March to the beginning of June. This is my primary reason for creating figure 1. During those truly dark days, if we had been testing at the rate that we are testing now, how many cases would we have seen? The double-red in Figure 1 above, is an estimate.
There has been much bad science, and even more bad media coverage, about the mortality rate going down, suggesting that the virus itself is getting less deadly. This is mostly balderdash. President Trump likes it because it allows him to talk about how much better we're treating the disease. Public health officials like it, because they don't like to talk about the actual spread of the disease. What is actually happening is that we are testing roughly 10x more than we were at the beginning of April. Thus, we are catching a ton more case (still a minority of the actual cases), increasing the denominator of the mortality rate, and, thereby decreasing the observed mortality--or the case fatality rate, CFR--but not the actual mortality rate (the Infection fatality rate, IFR). As of today, nationally, we have an estimated CFR of 1.4% (This is still much higher than the IFR, which includes all cases, known and unknown. We are still missing many cases, which I will delve into a little later in this post). Today's CFR is down from 40% at the very beginning of the epidemic. Some of this is due to improvements in care. In NYC at the peak, the ratio of new hospitalizations/deaths was 3/1. In some places (like Utah), the ratio is, and has been 6/1 (flu is 7/1). But most states still have a rate of around 4/1, hospitalizations/death. This has remained pretty constant throughout. Because of this consistency, we can use today's observed CFR--1.4%--to calculate the number of observed cases we would have expected to see earlier in the pandemic, had we been testing at the same rate throughout. As noted, the double-red line in Figure 1, above, represents this.
What we can see is that at the peak of the pandemic--truly the darkest 6-12 weeks--if testing were what it was today, we would likely have seen more than 150,000 cases/day--more than half of which were concentrated in just 7 states, with 16% of the population. There is no way we are getting to that same case level, even though the virus is currently much more widely spread, meaning that the burden is spread out across all states, and more manageable in each individual state. To put this into perspective, if all 50 states were to suddenly experience the spread seen in the northeast in April and May, we wouldn't see just 150,000 new PCR confirmed cases/day, we would expect to see 450,000 confirmed cases per day (and that would still be only about 1/6 or 1/7 of the real number of cases).
I think the other reason that people like Dr. Osterholm think we are headed to a massive third wave, is because of the large second waves being seen throughout Europe. This is mis-understanding the situation. Most of Europe had very hard lockdowns, much harder than ours. What we are seeing in Europe, is what we saw in the summer here--spread in the parts of the population that were not extremely hard hit in spring. Nowhere did our deaths (which are the only really good proxy) go down as low as in Europe, meaning the virus was still circulating, and pretty broadly at that. Our second wave, wasn't really even much of a second wave, when you tease apart the early-peaking states of the northeast from the rest of the country. For most states, the summer peak represented the culmination of the first wave, just spread out over a much larger period of time, i.e. a flattened curve--which was, after all, what we thought we signed up to do. When we see spikes (none of which have been anything like what we saw in the Northeast), they are in states that had not previously seen significant, widespread transmission. The same is true in Europe.
Most states in the U.S. are broadly open, with people broadly socializing with each other, despite what our epidemiological overlords might want of us. There are a few exceptions, like Massachusetts, where I live, and New York, but it is unlikely when these states do finally open that they will have large spikes given the size of their original spikes (and hence infections and population immunity). Certainly, there are places that are still largely closed, and which, like Europe, never saw large initial spikes, like the west coast states, and Hawaii. These will certainly have some significant increases in deaths whenever they do re-open, but for the most part, even these do not lag that far behind their demographic peers. For better or worse, the rest of the country is now behaving far more like Sweden. And just as Sweden is not seeing a significant resurgence, nor in all likelihood will we. Instead, we will keep plodding along, slowly working our way along our flattened curve, until we reach its end, my guess being some time in the first half of January--with or without a vaccine.
Figure 2: Comparing Deaths/Day/Million for 7 early-peaking states, and the other 43 states.
Source: The COVID Tracking Project. Rolling 7 day average daily deaths/million as of 10/10/20, "Early Hot Spot" states (NY, NJ, CT, MA, RI, MI and LA), vs. the rest of the country. Analysis and visualization provided by Emily Burns.
What were the darkest weeks of the pandemic really like?
I'd like to return for a moment to Dr. Osterholm's rather preposterous statement about the next 6-12 weeks being the darkest of the pandemic. To say that we are in for the darkest 6-12 weeks is to grossly misunderstand the magnitude of the disaster that occurred in the northeast in spring--as I noted above. When we look at the death curve of the epidemic in the U.S., the summer peak is smaller, 1200 deaths/day vs. 2100 in April. However, what that obscures is that in April, half of the deaths were driven by only 7 states, 7 states with a population of 55 million. The other 43 states, with a population of 270 million made up the other half. For the magnitude of disaster that we saw to be revisited, we would need to see the same level of deaths and hospitalizations in those other 43 states, that we saw in those first 7. Those 7 states peaked at 27 deaths/day/million in population. That would mean, nationally having 9400 deaths/day (27 * 270+2100). In terms of hospitalizations, which are shown below in figure 3, those 7 states saw at their peaks, 38,000 people hospitalized. At 5 times their population, the rest of the country would need to see close to 200,000 simultaneous hospitalizations--240,000 if all 50 states were peaking simultaneously. As a note--this would still only be 28% of our national hospital capacity (we are currently using 4%, and have never exceeded 8%). I suppose it is worthwhile to remember that even in NY, the hospitals were purportedly not overwhelmed, and the overflow hospitals that were built remained unused.
Figure 3: Comparing hospitalizations and Deaths/Day/Million for 7 early-peaking states, and the other 43 states
Source: The COVID Tracking Project. Rolling 7 day average daily deaths/million as of 10/10/20, "Early Hot Spot" states (NY, NJ, CT, MA, RI, MI and LA), vs. the rest of the country. Analysis and visualization provided by Emily Burns.
Given this actual apples to apples comparison, what Dr. Osterholm is proposing seems utterly preposterous: not just verging on, but well into fear-mongering.
I noted that if we were testing at the same level that we test now, we would have found 150,000 tests per day in April (incidentally, on a population/adjusted basis, France and the UK are seeing these kind of numbers, now), and that if the whole country had been on fire as those 7 states were, we would have seen 450,000 confirmed daily cases. But what if we had been testing enough to catch ALL of the cases, how many daily new cases would we have seen in April? How many would we be seeing now?
Well, we can do that, too, by backing into the real number of cases based on the Infection Fatality Rate (the IFR), which measures the ratio of all deaths to every person who actually catches the disease (this is how we get the 0.1% IFR for flu). Thanks to NY State's seroprevalence work, we know the actual IFR for that area. Based on the 15,000 person seroprevalence survey done in NY, we know that the actual IFR for NY State was 0.5% roughly. This is probably a good IFR to use for these 7 early-peaking states, all of which have seen significantly higher excess death rates than the rest of the country. For the rest of the country we can use the lower, 0.23% IFR, calculated across many different locales (including many U.S. states) by John Ioannidis, and now in use by the WHO. Ioannidis notes that with proper management, i.e. shielding of the vulnerable, that number could be substantially lower. In fact, Utah, which is extremely open, has fewer excess deaths from COVID than in a bad flu year (a good indication of the veracity of Dr. Ioannidis's claim). This 0.2% IFR is further supported by the summer in Europe. During the beginning of Europe's second wave, calculating CFR the same way that I used above yielded 0.2% in Spain, i.e. for 10,000 cases, there would be 20 deaths about a week later--inline with Dr. Ioannidis's estimated IFR--meaning they were literally catching every single case (that is no longer the case anywhere in Europe, though they are still catching far more than in spring, as we are, but explains why case levels that are the same or higher than the spring do not see the same levels of deaths as in the spring).
All right, so using these IFRs to back into cases, how many daily new cases were there really at the peak? Based on the analysis below, the real peak in April would have been just shy of 600,000 daily cases. If the whole country had been as bad as those 7 states, we'd have had 1.8 million cases/day. That would indeed have been a dark dark time. By the same analysis, we are now roughly around 280,000 cases/day (versus the 60,000 PCR-confirmed).
Figure 4: Daily U.S. COVID-19 Cases Estimated Based on Differential Seroprevalence
Sources: Case & test numbers from COVID tracking project. Estimated mortality (IFR) calculated from seroprevalence in NY state, and extrapolated for other 7 states. Rest of U.S. IFR based on aggregated seroprevlance work done by Dr. John Ioannis
Now, Figure 4 is an estimate, but it's directionally accurate. Let me first provide a few data points that give credence to these numbers. First off, note that using this methodology, there would have been 6.2 million people who had COVID by April first. Researchers who looked at the existing surveillance network for influenza-like illness (ILI) estimated that there were 8.7 million COVID cases between March 8 and March 28th. That would mean that this model is conservative (which was my goal). One thing to note about the linked article, it says that "cases were being undercounted by as much as 80%." This is a fairly embarrassing mathematical error. As of March 28th, there were 127,000 cases, meaning that they were actually under-counting by 98%. I presume that they used the total number of cases as of their submission date, which would have been in June (for July publication), when cases were around 2 million. Sadly, this is the kind of lazy math that is ruling the day. Regardless, the study gives credence to this model. A few other data points of note. On June 25th, CDC director Redfield estimated there were 20-25 million people who had gotten COVID. The model above estimates 33 million by that date. I have always thought director Redfield's number seemed quite low, based on seroprevalence, and the ILI surveillance work done by the researchers mentioned above. A study published in the Lancet, showed that through July 15th, roughly 10% of study participants had COVID antibodies, and thus estimated that approximately 33 million people had had the disease by mid-July. The model above would suggest 46 million. However, the Lancet study was done in dialysis patients. Patients on dialysis have end stage renal disease, the complication most associated with death from COVID. One would hope that such patients were less likely to get COVID than others in their communities, not more. Further, we now know that antibody studies can undercount significantly due to a) antibodies waning below detectable levels (less likely in this population, as severe disease is linked to stronger immune responses), and b) that 33% of patients with neutralizing antibodies do not generate positive results with such tests--despite generating antibodies, and mounting effective immune responses.
All of these things lead me to feel confident that this model is not far from reality. Back in mid-July, we had recorded 3.5 million cases to the Lancet article's estimate of 33 million. Since that time we have registered an additional 5 million cases, but increased testing only 30%. Further--and far more important--in mid-July, the "rest of the country" the 43 states "normal" states had logged 63,000 deaths (the early-peaking 7 states had logged 65,000 all by themselves at that time. Since that time, the 43 states that make up the "rest of the country" have logged an additional 76,000 deaths (the early-peaking states, 6400). This would mean that if mortality had remained identical, we would be no less than 50 million cases. IFR's calculated in multiple states suggest about 1/2 the mortality of that observed in the northeast. Which would mean that each death represents twice the number of cases (lower mortality, means more cases for every death), again, making this 70 million-ish number not a bad one.
50-70 million cases, and we're still talking about "Stopping the Spread". Seriously?
The authors of the Lancet journal use these numbers to argue explicitly that we cannot pursue a herd immunity strategy. Doing so ignores the reality their data implies. They continue to flog non-pharmaceutical interventions such as lockdowns and masks in pursuit of "reducing the R," stopping the spread. They refuse to recognize, despite their own research making it abundantly clear, that the spread is far, far greater than what we find through testing. If, for example, we were to make an effort to find every case, we would see our "R" rate skyrocket, despite real cases decreasing. This is why using R, and case-tracking to define success when testing is not consistent, and catches nowhere near the actual number of cases, verges on the absurd. This is particularly clear when considering a country that has in all likelihood never had fewer than 200,000 new actual COVID cases in a single day since the beginning of March.
Deaths are going down, social interaction is going up (which means COVID is circulating more). It's hard to think that this decrease in deaths is anything other than herd immunity coming into play. Our public health officials dismiss this, because they did some simple math back in March that told them that 67% of people would have to get the disease before the spread was stopped. They have willfully ignored myriad research papers showing significant pre-existing T-cell immunity in up to 50% of the population, especially amongst the healthy, which indicates a far lower threshold for herd immunity, particularly if the vulnerable are protected, as opposed to being the only people left exposed, and forced to shelter-in-place with lower-risk family members who might have an asymptomatic but still infectious case. This is supported by the case of Sweden whose seroprevalence studies show just 20% of the population with antibodies--just like New York--and which like New York has ceased to see any meaningful growth in cases or deaths attributable to COVID-19.
Ignoring the possibility of herd immunity as an approach to get through this epidemic when you are clearly failing to stop the spread by other means, consigns you to a herd immunity "strategy" by default. With a disease like COVID where there is a such a widely stratified risk profile, this "accidental herd immunity" will undoubtedly result in higher death tolls. This is because the disease is only able to circulate amongst the most at-risk within those communities. The early-peaking states prove this in spades. Locked-down societies are locking down (and keeping out of circulation) exactly those people who would be least likely to contract the infection--due to T-cell immunity from prior coronavirus infections-- and least likely to develop serious complications or die. These are precisely the people who would provide a desperately need immunological shield to those who are at-risk (and this is how diseases have progressed throughout all of evolutionary history). Back in March, the Imperial College published this report which shows how highly stratified the mortality is--someone over 85 is 4500 times more likely to die of COVID than someone under 10.
Table 1: Hospitalizations, ICU and Fatality Rates by Age in Symptomatic COVID-19 Patients
Source: Imperial College of Medicine
4500 to 1? This is why the Focused Protection approach of community immunity proposed by the scientists behind the Great Barrington Declaration makes so much sense. 4500:1 actually obscures the real discrepancy in mortality amongst groups. 4500:1 is simply the age-based discrepancy. It hides the additional compounding factors of co-morbidities. Throughout the world, 90+% of deaths are associated with patients with co-morbidities, nor are these trivial co-morbidities like slightly elevated hypertension. The truth is age, is a proxy for all of these co-morbidities. If you look at a chart of age-based deaths by COVID, vs. deaths due to the co-morbidities that increase risk of death by COVID, they look almost identical. Not surprisingly, the average age of death for COVID-19 is 78--the same as the average age of death in the U.S., period.. Thus, while age is indicative of these co-morbidities, it is not the determining factor of likelihood of survival. More importantly, within all of these age groups, there are people who are at significantly elevated levels of risk, well beyond this 4500:1.
We Need to Re-Evaluate to Start Protecting the Vulnerable
Because our Public Health officials have so little confidence in our ability to handle nuanced messages,, they feel it is better to scare 96% of us too much, rather than direct a very pointed message to 4%. This does a great disservice to the part of the population who is at serious risk, not simply to the part of the population who is not at-risk. For every two hundred 20-45 year olds that view their risk as a 1000 or 10,000 times greater than it actually is, there is some person 65+ person who is within a year or two of entering a nursing home, with very fragile health, who despite understanding that they are at greater risk, is underestimating their risk by a thousand, or ten-thousand fold. Likewise, there are many other healthy 65+ people who are significantly over-estimating their risk, and missing out on what for them, is a significant portion of their remaining years, despite a not greatly elevated risk of death.
I performed an analysis of New York City's seroprevalence studies, and was able to come up with the following estimate of fatality risk by age group, for groups without co-morbidities. One of the great canards of this pandemic (and there have been many) is that we have so many people with so many co-morbidities, that we cannot possibly follow a community or herd immunity approach--people add up the people who are obese, then add up the people with hypertension, and before they're done, they're at 300% of the population (which ought to raise questions for them about their analysis, but of course does not). In truth, co-morbidities, like birds, flock together--people who are obese are more likely to have hypertension, more likely to have diabetes, people who have diabetes are more likely to have renal disease, and on, and on. The following chart shows the results of my analysis.
Table 2: Estimated Mortality Rate by Age and Co-morbidities by Age
Source: All of the analysis and sources are linked here.
The table above shows that 59% of New Yorkers across age groups have no co-morbidities, including 22% above 65, and 15% above 75+. Even those with mild co-morbidities in these age groups, are likely not to die, though they may have serious disease. We can see that based on the myriad elderly public figures we have seen contract the disease, many of whom have been hospitalized, nearly all of whom have come through (the exceptions being people like Herman Cain with stage 4 lung cancer).
You can contrast that with the raw mortality rates by age in New York City here (which are very, very close to those predicted by the Imperial college above--though hospitalization rates are significantly off, especially in the young--but that's another post):
Table 3: Estimated Mortality Rate by Age in NYC, Based on Seroprevalence Data
Source: All of the analysis and sources are linked here.
It is tempting (but lazy) to say, "well, if there are 64% of people between the age of 18 and 44 who have no co-morbidties, and their mortality rate is 0.003%, then the mortality rate must be 0.2% for the 36% of people in that group with co-morbidities". While this is statistically accurate, again, it would mask where the real risk lay, which is with those people in every age group with multiple, severe, and likely, life-threatening co-morbidities.
Herd immunity has repeatedly been characterized as heartless, "survival of the fittest." Rather, the suppression strategy we are currently pursing, ends up being a strategy of "protect the rich, infect the poor," by encouraging those who can to stay home and out of harm's way, with the people able to do that being the youngest, healthiest and wealthiest in our communities. Community immunity with focused protection would be far more compassionate. 65% of the U.S population is under 50. About 70% of those are neither obese, nor have co-morbidities, close to 50% of the population. An additional 10% or so above 50 do not have any co-morbidities. Even if the 60% herd immunity threshold were required, there are nearly enough people whose risk of death from COVID is significantly lower than their risk of accidental death (car crashes, overdoses, suicides). If the CDC would be honest with us about death rates among those with mild co-morbidities, the population whose risk is the same as, or lower than their risk from accidental death, is probably even higher. The fact that suicide and overdose rates are sky-rocketing makes this approach even more compelling.
The table below shows an estimate of deaths by age group if 100% of the healthy population in the U.S. contracted COVID-19.
Table 4: Estimated Mortality Amongst Healthy in U.S. if All Healthy Persons' in U.S. Contracted COVID Compared with Mortality for Same Group due to Accidental Causes
Source: Based on analysis of April NY seroprevalence surveys and age- and co-morbid status mortality in NYC. Methodology detailed here.
Table 4 shows the mortality that we would see if a) we were able to completely limit spread solely to the healthy population and b) if the herd immunity threshold was 55%. As for a) it is unlikely that we would be able to limit spread solely to the healthy, even with the best protection for the vulnerable. As for b) it is also unlikely that the threshold is this high given the observed case study of Sweden. Nonetheless, this is an important place to start, because while is it unlikely that such a focused protection strategy would be 100% effective at shielding the vulnerable, it has proven equally impossible to stop the virus, as evidenced by the still un-checked spread of the virus in every country in the world, save New Zealand and China (maybe).
With those caveats, let's dive a little deeper into the numbers above. What the chart above shows is that in a worst-case scenario where every single healthy person in the U.S. had to contract COVID before the disease's spread petered out, we could have expected to see 17,000 total deaths in the U.S. across all age groups--a 90% reduction in deaths relative to the strategy we are currently pursuing--and fewer deaths in every age group. Many people will say "but one death is too many." This is a false choice. First of all, the estimated risk of death due to COVID for the healthy across all age groups is significantly lower than their risk of accidental death (car wrecks, suicide, drug overdose). These people all presumably choose to drive. Therefore, given accurate information about their relative risk, they would also presumably feel comfortable continuing to live their lives in the midst of COVID--if not, they could choose to opt-out, as so many are already doing now. That the risk of accidental death due to suicide and drug overdose among this healthy group is being significantly elevated due to our current mitigation measures, makes this approach yet more appealing. At the same time that we reduce the deaths amongst the vulnerable, we also reduce the likelihood of death due to suicide or overdose in the healthy population--literally a win-win, live-live solution. Compare that to the COVID deaths we have seen in each age group to date, in our current "stop the spread," lockdown-light approach to managing the virus. Even in the younger groups, we have seen 3x the number of deaths due to COVID. In the oldest group, our current approach has yielded more than 100x greater mortality. Even if this focused protection approach were not 100% effective, it would still likely result in substantially reduced deaths among all populations, including vulnerable populations.
Those who reject this approach are simply not facing the reality, nor the magnitude of the failure of this world-wide "suppression" approach, most particularly as regards the most vulnerable in our communities. At the height of the first peak in the Northeast, a random sampling of nursing home residents in Connecticut found that 28% had active COVID infections. That PCR tests miss about 30% of infections, and that many residents likely contracted it before or after means that this number is the bare minimum. Seroprevalence surveys from the same time show that the general population in Connecticut had just a 5% infection rate, meaning that our "stop the spread" strategy, has resulted in the very most vulnerable in our population being infected at a rate at least 6x greater than the rest of the population. This is what failure looks like. When the data shows that your strategy has failed this monumentally, it's time to look at other options.
Let's hope that our Public Health Officials can adjust course during the remainder of this pandemic and begin to make an effort to protect the vulnerable. If they do not, they will be continuing to pursue what amounts to a slow herd immunity strategy, which builds our immunity on the backs of the very most vulnerable without our community.