As the Omicron wave recedes, I’m on a quest to determine when the risks imposed by Omicron will be no higher than the risks imposed by typical seasonal flu. Based on what I’m reading, that “flu benchmark” has intuitive appeal to a lot of people. If you don’t obsess about flu risks every year, that’s a reasonable starting point for discussing a return to normalcy with endemic Omicron.
Best guess? If you’re fully vaccinated and boostered against COVID-19, once Omicron falls below 40 new cases / 100K / day, your risk of hospitalization or death from Omicron is no higher than your risk of those outcomes from flu, during an average week in a U.S. flu season.
That estimate embodies an extremely conservative assumption that COVID-19 vaccines only reduce your risk of infection, with no additional protection against hospitalization or death, if infected. I’ve erred on the side of caution. By contrast, if you think the vaccine plus booster provides (say) an additional 50% protection against hospitalization and death, beyond mere protection against infection, then you can more-or-less double that 40 case threshold to 80.
With six states already below 100 cases / 100K / day and falling, maybe it isn’t too soon to start thinking about a return to normalcy for those who are fully protected.
This post provides the background and outlines the calculation behind that estimate.
Background
The overwhelming consensus of scientific opinion is that we are headed toward “endemic COVID-19”, whatever that may mean (Post #1400, part 2).
The most common model for “endemic COVID-19” seems to be seasonal flu. It’s always present at some level in the population, almost everyone has some immunity to it, every year maybe 10 percent of the population has a symptomatic case, and small fraction of those — mostly elderly and frail — will be hospitalized or die from it.
Some seasons and some variants will be worse than others. If you are at high risk, get your flu COVID shot every year and stay out of crowds during peak flu COVID season.
I can’t quite get my mind around that simple picture of COVID-as-flu, for two main reasons. First, the current variant (Omicron) is ten times as contagious as typical seasonal flu. (R-nought of 15, versus maybe 1.5 as a median value for flu in a typical year). Second, immunity seems to fade quickly.
That combination makes it unlike any other endemic disease that we routinely deal with successfully. A lot of childhood diseases are as infectious as Omicron, but we have long-lived vaccines for those. If you mandate vaccination for schoolchildren you’ve more-or-less prevented large-scale epidemics. With Omicron, by contrast, the only way I see to avoid epidemics is to mandate that everyone get vaccinated every year. You can practically hear the squawking start at the very mention of that.
We have a while to think about it, in most places. Based on the typical rates of decline of the Omicron wave that have been observed internationally, Omicron will still be circulating in most of the U.S. at pandemic levels for some time yet.
In addition, in the long run, after the presumed end of the Omicron wave, there are at least three obvious unknowns that will govern how this will play out: New variants, new vaccines, and immunity following infection. We already have the BA.2 variant coming, reported to be 1.5 times as infectious as Omicron. Manufacturers are slated to release at least one Omicron-specific vaccine in March 2022, of as-yet-to-be-announced efficacy. And we still don’t know the extent to which an Omicron infection provides long-lasting immunity against a subsequent Omicron (or BA.2) infection.
Putting all those uncertainties aside, for me, it’s getting to be time to start figuring out when the world will feel “normal” again. With the assumption that we end up with a “flu-like” endemic omicron. Which seems to be the current consensus of where we are heading.
Last summer, when we were still facing the original (Wuhan) strain and new case counts were in the single digits, I went back to most of my normal routines. Went back to the gym, started seeing movies again, and just generally freely used areas considered to be places with relatively high risk of COVID-19 transmission (Post #813, Qualitative rankings of activities by risk of COVID-19 infection ). That was done with forethought, after calculating the risks (Post #1163, June 2021).
Then, when case counts went up again, I stopped doing that.
In that context, this current blog post is just another variation on that calculation. At what level of new Omicron case counts will I judge risks to be low enough to be, for all intents and purposes, I can ignore them?
Prior estimate for all persons combined
In the third section of Post #1400, I did head-to-head comparison on risk of hospitalization and death from Omicron, versus typical U.S. seasonal flu. I did that for all persons combined. My best guess, at that time, is that once Omicron gets below 30 new cases / 100K / day, the average person faces no more risk from that, than from flu at the peak week of a typical U.S. flu season. At 16 new cases / 100K / day, Omicron poses no more risk then flu does for the average week of the entire flu season (instead of the peak week), for the average American. Those are the rates at which, by my calculation, the average American faces no more risk of hospitalization or death from COVID-19 than from normal seasonal flu.
The arithmetic there isn’t rocket science. Using CDC data, I estimated that a typical week during flu season sees about 49 / 100K / day new symptomatic flu cases. Then, again calculating from CDC data, if you catch an infection, Omicron is about three times more likely to hospitalize you or kill you, compared to flu. So, to equalize your chances of hospitalization or death, your risk of catching Omicron has to be one-third that of flu. And 49/3 = 16 or thereabouts.
I now want to refine that, and talk about the fully-vaccinated-and-boostered only. That’s not only because that’s what’s relevant to me, but also because it’s pretty clear that the people won’t get vaccinated aren’t worried about the risks they impose on themselves and others.
Doing that more detailed calculation turns out to be a whole lot harder, for a wide range of reasons to be discussed below. The results are best viewed as a refinement on my prior one-size-fits-all estimate, and not as any definitive final answer.
Details and calculation, Part 1: Simplifying the issue, or, all the stuff I can’t do.
This is a tough and imprecise calculation for a variety of reasons. I’m going to list all the things you just have to turn a blind eye to, or can’t get information for, or can only get scattered information for. Skip to the next section of that’s of no interest.
1: Risks vary widely by age and frailty, my target audience for analysis of risk is older adults.
The elderly have a much higher risk of hospitalization and death from Omicron or from flu, compared to other groups. (The sole exception is a slightly elevated risk of death of flu in infants, compared to others.)
Here’s how the case rates line up for flu (here, the 2017-2018 flu season), for hospitalizations and deaths per case, by age. Just under 1 percent of children up to age 4 with symptomatic flu were hospitalized, as was about 1 percent of the 50-64 year old population. And then the elderly had an average case hospitalization rate of about 9 percent. Flu deaths were distributed in a roughly similar manner, except that deaths were not proportional to hospitalizations for small children.
Source: Calculated from CDC Disease Burden of Flu, 2017-18, Table 1.
You can see a similar distribution by lining up the case mortality rates for COVID-19 (for the entire pandemic, not for Omicron!) against flu, by the same age ranges. Except for the scale, the lines are virtually identical. The COVID-19 line is ten times the flu line because this is for the entire pandemic, not just for the later sections including Omicron.
Source: Calculated from CDC Disease Burden of Flu, 2017-18, Table 1, and data from the CDC COVID data tracker demographics section.
What’s worse, the observed rates in those tables reflect the variation in vaccination rates by age group. And they reflect the concurrent frailties of old age. (That is, the elderly aren’t just old, they have a lot chronic illness burden that goes along with being old.)
The upshot is that I’m going to average across all of that variation. If you’re 30, you’ve probably never given the idea of death from flu a second though. As you approach Medicare age, that’s more of a reality. Roughly speaking, the average 65-year-old would have just about the mean risk rate that I’ll be talking about.So, in some sense, my target audience for this analysis is older adults.
The low cases hospitalization and mortality rates for younger age groups bring up at two other aspects of this calculation that are not covered by the simple likelihood of you, yourself, suffering these outcomes.
First, this ignores other morbidities that were associated with COVID-19, at least early on in the pandemic, that may be far more prevalent in the young. That is, even if risk of hospitalization and death are low in certain age groups, there was also the additional risk of long COVID, temporary or long-term loss of sense of taste and smell, temporary or long-term organ damage, and the like. It’s possible that younger people still face significant risks from those conditions, but as far as I can tell, there’s no hard data on those other morbidities under Omicron. (Other than for myocarditis, which is its own separate topic.)
So you, personally, face health risks beyond hospitalization and death. It’s not possible to bring those into the calculation.
Second, this also touches on the “public” part of public health in this area, which, for starters, asks you not to spread disease to others. Even if you yourself are not particularly at risk of great harm, somebody further down a chain of infection that you helped to perpetuate might be. So a “total harm” calculation would include not just harm to self, but plausible harm caused to others by failing to (e.g.) get vaccinated or adopt agreed-upon rules of COVID-19 hygiene.
To an economist, this effect — the fact that you might cause harm to others without having to pay for it — is an “externality”. It is, in a sense, a “missing market”, in that you don’t have to pay for the damage you cause. Republicans are ideologically blind to externalities — in health care, in environmental policy, in areas of consumer protection, and so on — because controlling them in an economically efficient manner generally requires the government to step and and do something to try to approximate that missing market. As a result, lots of famous Republicans publicly and proudly act as if they really, truly don’t care whom they infect. From the standpoint of Republican orthodoxy, that’s not a bug, that’s a feature. To an economist, it’s just inefficient.
2: Published current illness rates commingle risk classes.
Not only are the data averaged across all ages, every population statistic you see is an average for the boostered, vaccinated, and un-vaccinated populations combined. So when I note that the U.S. is around 170 new cases per 100K per day now, that’s a combination of a much higher rate of infection among the unvaccinated, and a much lower rate among the boostered.
This means that the first step of the process is to estimate the rates separately, for each population, using some known population proportions, some estimates of vaccine effectiveness, and some algebra.
3: Observed rates broken out by risk classes (vaccinated versus not) do not provide reliable data on the effect of vaccination.
You have to take your estimates of vaccine effectiveness from controlled studies of some sort. You can’t just read them off a table of simple average rates by those who were vaccinated and not.
The longer I’m at this, the more convinced I am that most people really, truly do not understand the difference between an experiment, such as a randomized controlled trial, and “observational data”, meaning, whatever shows up in the population.
People routinely (and incorrectly!) take observational data and treat it as it were the results of a controlled experiment. And the professional liar class that infests social media makes it a point to seek out such data, when it seems to make some point that they wish were true, and deliberately misrepresent the simple comparison of of averages as if it were the results of a controlled clinical trial.
If I randomized individuals into two groups, then vaccinate one and not the other, any difference in infection rates between the groups will be attributable to the act of vaccination. Plus or minus a bit of statistical uncertainty, particularly if I’m only using small numbers of individuals. That’s by design, because a) the experimenter chooses whom to vaccinate, and b) assuming the randomization is done well, all other factors affecting infection rates will be equal across the two groups.
But if I observe the infection rates of people who chose to get vaccinated or not, I’m looking at not only the impact of vaccination, but also the effect of all other differences between those two self-selected groups.
In this case, let me call those other factors the Palin Effect. The same people who refuse to get vaccinated are likely to engage in riskier behavior across-the-board. But they are also probably younger, on average. They might work in a different set of careers from those who chose vaccination. They might hang with a different peer group. And so on.
Because of the Palin Effect — the unvaccinated really are different from the vaccinated — the averages for those groups are often vastly different from the actual vaccine impact, which is what you get from controlled clinical trials. In Virginia, for example, I routinely see that the un-vaccinated get COVID-19 infections at ten times the rate of the vaccinated. For the week ending 1/1/2022, it was sixteen times:
Does the observational data from Virginia mean that vaccines are actually vastly more effective against Omicron than the clinical trials demonstrated? No, of course not. Almost certainly, vaccination by itself provides only modest protection, and vaccination plus booster is only about 70 effective in preventing any symptomatic infection. The difference from what you would expect at best (1/.3 = 3.33 times the infection rate) and the observational data (16.6 times the infection rate) is almost certainly a large Palin Effect. It’s the unvaccinated and unmasked hanging out with like-minded people, going out to party over the holidays and spreading COVID-19 at a greatly heightened rate, compared to the vaccinated.
And it’s easy enough to convince yourself of this. Just look at some very different state, such as Washington. There, the raw difference in infection rates between vaccinated and un-vaccinated is typically about 4-to-1. Same disease, same vaccine. The fact that the observational number is all over the map should clue you into the fact that it’s not showing you the impact of vaccination in isolation.
Scientifically, there’s also a murky middle ground between proper randomized trials and just taking average rates by group. These go by a wide range of names (“case-control study”, “propensity score analysis”, “regression analysis”, “pre-post with control comparison”, “instrumental variables”, “simultaneous equations”, “natural experiment”, “cross-sectional analysis”, …), but they all boil down to using statistical techniques to try to separate out the effect of (say) vaccination from the effect all those other factors (covariates). That gray area is where I lived all my professional life, for the simple reason that there are no controlled experiments in economics. Much of what you read in the newspapers consists of results of studies like that. They can be well done and provide useful information. They can be poorly done and be completely misleading. Typically, the researchers themselves don’t really know which, and for sure, newspaper reporters have no way of knowing the difference. Hence we end up with a rich and diverse array of bad science that gets public notice. Of course, the more bizarre the results, the more attention they will gain.
4: Other stuff I’m just going to ignore.
4.`1: Risk aversion and assumption of risk-neutral persons. I’m going to treat people as if they are risk-neutral, in the way economists use that term. The crux of that is that I’m only going to pay attention to the average rate of bad outcomes, not the distribution.
For example, I’m going to say flu and Omicron generate equal risk if (say) you’re half as likely to catch COVID as flu, but each COVID cases is twice as likely to land you in the hospital. The fact that I’m valuing those two scenarios as equivalent to one another is saying that I’m risk-neutral. But people can (and will) rationally value those two scenarios differently. You might rationally fear Omicron more, because if you catch it, there’s a much higher risk of hospitalization. That’s not irrational, that’s just typical risk-averse behavior. I’m just ignoring that.
4.2 There are numerically important quirks and differences in how we estimate infection rates and case rates for COVID and flu. I went over those in Post 1400, Part 3, and I won’t repeat them here.
4.3 The data on effectiveness of booster shots at preventing hospitalization and death from Omicron, above and beyond their ability to prevent infection in the first place, is scant. Due to the small number of events, most of what I’ve seen is observational data based on relatively few observations, or just straight-up anecdote.
There are good theoretical reasons to expect vaccination or prior infection to be more effective at preventing the worst Omicron outcomes, compared to preventing infection alone. Preventing any infection relies on the rapid-response part of the immune system (antibodies), and Omicron has found ways to get around existing antibodies (and/or antibody levels fade). But other parts of the immune system remain primed to fight COVID even if antibodies fail, and these parts react more rapidly to a new COVID infection than would occur in an un-vaccinated, un-infected individual.
That said, I struggle to find a consensus on just how large that effect is.
Details and calculation, Part 2: Calculation and results.
I’m only going to give the barest outline here. Really, I’m going to show you the assumptions I made, and then you’ve got to trust that I did the arithmetic correctly.
Step 1 of the calculation estimates the number of COVID-19 infections and flu infections you would expect in the vaccinated and un-vaccinated portions of the population, all other things held equal. This is the part where I break the published total into what I would expect to see, if the only difference between the vaccinated and unvaccinated populations was vaccine. (I.e., absent any Palin Effect).
- Start with the known average number of cases / 100K / day.
- Factor in the experimental estimates of vaccine efficacy at preventing infection.
- Work in the fraction of the population that is vaccinated.
- Solve for the infection rates in each segment of the population that, when averaged together, would give you the observed rate.
I’m omitting the details of the calculation, but the upshot is that, absent any Palin Effect, even if you are vaccinated and boosted, your likelihood of getting an Omicron infection per day is higher than your likelihood of getting a flu infection, per day, for a typical flu season.
You will not be able to follow the calculation from the table above because I’ve omitted numerous intermediate columns. You’re going to trust that I’ve done it right.
I’m also not going to explain where all the assumed numbers come from. The basic numbers on flu prevalence and such are from CDC:
The rest of it is my summary of the literature. Flu vaccine is only about 40% effective, on average, at preventing symptomatic infection. COVID vaccine and booster is about 70% effective against Omicron. And so on.
Step 2: Once you have that, do the same trick again, but this time solving for the hospitalization case rates by population segment, and converting that to the risk of hospitalization per day per 100K persons.
This time, you have to start with the expected number of infections per 100 population. As before, you can’t follow the calculation from the numbers shown. But I’m showing you that I assumed vaccines provided no additional protection against hospitalization, above and beyond the reduction in getting infected in the first place.
The upshot is that, right now, for equivalent populations, those with vaccine and booster for COVID are about five times as likely to be hospitalized for Omicron, per day, than they would be hospitalized for flu, in a typical flu season. That’s the result of much higher case rates for Omicron right now (relative to flu), offset by a somewhat more effective vaccine, but also factoring in the much higher average case hospitalization rate for Omicron compared to flu.
Step three is to go back and change the current Omicron case rate to something lower, until those final rates are equalized. In this case, those final hospitalization rates would be equal if there were just 40 new cases of Omicron / 100K / day.
I can do the exact same calculation with mortality rates, using my most recent case mortality rate estimate. And I find roughly the same thing. If Omicron were to fall to 40 cases / 100K / day for the entire population, the expected mortality rate from Omicron, for the fully vaccinated and boostered, would match that for flu on a typical day of flu season, for a person who has had flu vaccine. (Which I estimate to be about 0.05 deaths / 100K / day).
Does that all hang together? At a population average of 40 cases per day, the fully vaccinated and boostered population would see a theoretical average of just 18 cases per day. I have assumed no additional protection against hospitalization or death. The estimated case mortality rate for Omicron is about 0.3%. And, sure enough 18 x .003 = .05.
What I’m trying to say is that although the many assumptions may be questionable, I think I’ve done the arithmetic right. At 40 new Omicron cases / 100K / day for the population as a whole, under these assumptions, the fully boostered population would face:
- 18 cases of Omicron / 100K / day.
- 0.54 hospitalizations for Omicron / 100K / day.
- 0.05 deaths from Omicron / 100K / day.
That 40 / 100K / day level of new Omicron cases (for the entire population) would give those fully vaccinated and boostered individuals a much lower chance of catching Omicron COVID than of catching flu on a typical U.S. flu season day. And that would give roughly the same chance of hospitalization or death, per day, as those individuals incur in a normal flu season, assuming they get their flu shot each year.
FYI, if you believe that vaccination plus booster provides even more protection against hospitalization and death than it does against mere infection, then “flu-equivalent” rate of Omicron infections would go up. If you think it cuts your case rate of hospitalization or death in half, then the “flu-equivalent” rate of daily new COVID-19 cases for the entire population would rise to about 80.
Epilog
All I’m shooting for here is a rough guideline for when we can reasonably expect a return to normalcy. Personally, I’m sketching out the new Omicron case rate at which I’ll be going back to the gym, going to the movies, and so on.
What’s your alternative? You can sit around until the CDC tells you it’s OK. If they ever do so. You can depends on some random internet source. You can try to go with the herd.
For me, I like to figure the odds.
I’ve built a few safety factors into this estimate. So it’s fairly conservative. But my best guess is that when the overall population case rate drops below 40 / 100K / day in my area, then pretty much all COVID-19 hygiene becomes optional. For the simple reason that I don’t sweat the risk of hospitalization or death from flu. And I face that every year, mask- and restriction-free.
YMMV. If you can find a better guide to where “normal” starts, use it.
At any rate, this finally ends post #1400. This is my estimate of the case rate at which the fully-vaccinated-and-boostered population can start to ignore COVID-19. This is my best guess for the psychological point at which endemic Omicron starts.
I think I’m going to reserve the right to re-write this one. It was quite a chore to crank it out, and I’m not sure I’ve been very clear.
Bottom line: When I see a daily case rate of 40 / 100K or lower, here in Fairfax County VA, I’m just going to stop worrying about my COVID risk. I’ve never worried much about flu risk. Below that level, it would be irrational of me to worry about risk from Omicron.
I might still wear as mask where convenient. Because, why not? I already own what I hope is a more-than-lifetime supply of N95s. And, honestly, sometimes wearing that mask is just an act of politeness, if the people your with are more worried about COVID than you are. Nothing wrong with that. Otherwise, on or about that time, it’ll be back to business-as-usual for me.
