As pictured here, we’re still on track for a peak of the Delta wave in early September.
The U.S. is now at 48 new COVID-19 cases per 100,000 population per day, down a bit from yesterday. The seven-day increase was 7%, not materially different from yesterday.
As we saw in past waves, a second “wavelet” of states is emerging. New case rates have peaked in the three states that led the wave (FL, LA, MS). Those states are now being overtaken by KY, SC, and TN, all of which have current new case counts in the 90s and double-digit growth rates. Georgia and Alabama are not far behind.
Data source for this and other graphs of new case counts: Calculated from The New York Times. (2021). Coronavirus (Covid-19) Data in the United States. Retrieved 8/28/2021, from https://github.com/nytimes/covid-19-data.” The NY Times U.S. tracking page may be found at https://www.nytimes.com/interactive/2020/us/coronavirus-us-cases.html.
Above is the now-familiar graph of the U.S. broken into six regions. Of note, the center of case growth appears to be moving north. (It could hardly move south, as the epicenter of the Delta wave was the U.S. Gulf coast).
Below we can see the crest of the first wavelet (FL, LA, MS) in gray, being replaced by a new set of states in color.
A note on trends in the U.S. COVID-19 death rate.
I’ll make this short: I have no clue what’s going on.
I can make up some stories, but every one I make up appears contradicted by the data.
The best story I can tell is that vaccines prevent death far better than they prevent illness. And so, with breakthrough infections (infections among the vaccinated), we’re seeing cases, but (mostly) no deaths. And so the case mortality rate (deaths per diagnosed case) is falling.
The first problem with that story is that there aren’t that many breakthrough infections. Not enough, at first glance, to depress the mortality rate by the amount shown above.
That said, to a degree, breakthrough infections are appearing somewhat disproportionately among the elderly. And so, because almost all COVID-19 deaths are in the elderly, maybe that modest count of breakthrough infections, if concentrated on the elderly, really could depress the observed mortality rate that much, with or without account for changes in the age mix of new cases.
The second problem is that if this were true, we’d be seeing a whopping decline in the case mortality rate of the elderly. All those breakthrough infections not resulting in death with bring the ratio of deaths to cases down. But, as far as I can tell, there’s no sign of that. That’s part of the data analysis below. The case mortality rate in the elderly is no different from what it was at the start of the Delta wave.
In other words, that’s a nice, plausible story, but it’s not supported by the data.
I could tell you some other plausible stories, but none of the other ones I’ve run across appear supported by the data either.
Compared to prior waves, the Delta wave appears to be generating more hospitalizations and fewer deaths per case (per individual diagnosed with COVID). This, per U.S. CDC data as shown below. See how the tail end of the yellow line flips up and down. One of those lines is hospitalizations (up), the other is deaths (down). The red line is new COVID-19 cases.
The increased hospitalizations per case is no particular surprise. The Delta strain of COVID-19 is more virulent than prior strains. For a given population, it’s about twice as likely to result in hospitalization (reported succinctly here).
The surprise is the apparent decline in the case death rate, as shown above.
I’m pretty sure the lack of deaths isn’t just some artifact of data reporting. (E.g., that it just takes a really long time for deaths to be reported, so this curve always looks like this at the end). Mainly, I’m not the only one to have noted the low death rate. You can read reporting about it. This article attributes low U.S. and U.K. death rates in the Delta wave to vaccination. And Florida has had far more cases in the Delta wave than in prior waves, but has not yet seen as many deaths (per this reporting). And Ohio. Minnesota. You get the drift. This is a widespread phenomenon appearing under a variety of administrative data reporting systems. I don’t think it’s a data reporting glitch.
Vaccines: Everybody says so, nobody has the numbers to prove it.
By far, the most common explanation for this is vaccines. The plausible story is that vaccines prevent deaths better than they prevent infection. And so, as the fraction of cases that are “breakthrough” cases rises, we’re seeing a reduction in deaths per case. They show up as infections, but rarely as deaths.
That’s a plausible story, but I don’t think there are enough breakthrough infections to account for that. And nobody seems to be able to put all the numbers on one page to see whether that story holds water or not.
It’s true that vaccinated individuals, once infected, have a substantially lower risk of death. But in Virginia, in any given week, at most, we might have 20% of all reported infections as breakthrough infections. I think that’s roughly typical. Even if the vaccine absolutely prevented death, all other things equal, you’d expect to see the case mortality rate being 20% lower than you would otherwise expect. And clearly, based on the graph above, the drop is more than that.
There is some evidence to suggest a generally higher rate of breakthrough infections among the elderly. If true, then the 20%-of-case could be a higher percentage of cases-at-high-risk-of-death.
Here, there’s less information available. What I found is that both Oregon and Washington report that about 25% of breakthrough cases are in the age 65+ population. Near as I can tell, both states show about 10% to 12% of all cases have been in that age group. Hence, on a per-capita basis, breakthrough infections appear about twice as likely in the elderly.
(Note that this is NOT the relative risk of infection for a vaccinated individual. I’m not saying that a vaccinated elderly person is twice as likely to have a breakthrough infection. I’m not saying that because a higher fraction of the elderly are vaccinated, and they’ve been vaccinated for a longer period, than the rest of the population. The “twice as likely” above combines both the sheer number of vaccinated individuals, their longer exposure time post-vaccination, and then, probably, some higher probability of breakthrough for any given vaccinated individual.)
Based on that estimate, if breakthroughs account for 20% of all cases, they might account for 40% of all cases in the elderly. Because almost all deaths are in the elderly, now we’re getting to the point where this might plausibly explain the reduction in case mortality rate. Roughly speaking, we’d be missing 40% of the people who otherwise ought to be dying from COVID-19 once infected.
That said, as I mention in the data analysis below, if that were true, the case mortality rate for the elderly ought to fall quite a bit. If it really is true that we’re now missing 40% of their post-infection deaths, because of the vaccine, that should have a profound impact on the case mortality rate for that group. And that doesn’t show up in the data.
So, again, while this is the most plausible story out there, I don’t think the data back it up.
In this section, I calculate an age-standardized COVID-19 mortality rate for the U.S. This is just a simple and standard way of figuring out what the mortality rate would be, if the age mix of the population stayed constant over time. The idea is to see whether or not this falling mortality rate is due to a shift in the age mix.
A lot of things make it tough to analyze the death rates. Among other things, deaths lag cases by about two weeks — roughly the average time that it takes for a COVID-19 decedent to die, and have that death certificate tabulated. So you have to account for that.
But what made this a particularly difficult exercise is that nobody has the right national data. All you need is cases and deaths, by week, broken out for the same age groups. What you can find is a mix of highly-incomplete CDC COVID monitoring data, CDC/NCHS death certificate counts, and then national totals for cases and deaths that aren’t a very good match to either. All with different age breakouts.
Nobody cares about methods. I finally was able to calculate an age-adjusted mortality rate by cobbling together several different national datasets. That is, I calculated what the U.S. COVID-19 case mortality rate would have been if the age mix of cases hadn’t shifted over time.
To obtain the age distribution of cases, the CDC’s grossly incomplete person-level COVID-19 case surveillance file. I couldn’t use that for anything but percent-of-cases by age, because the file and most of the file data fields are grossly incomplete. For deaths by age, I used CDC Vital Statistics data, which is tabulations of death certificates. I used Census population counts by single-digits-of-age to translate from the age groups used in the CDC surveillance file to the ones used in the CDC deaths file. (I apportioned the cases to the deaths age brackets based in portion of population). I also shifted the deaths by two weeks, relative to the cases, to account for the typical lag between the two.
It’s absolutely true that a shift toward younger cases ought to drop the case mortality rate. Virtually all the deaths occur in the older population.
Source: Calculated from a variety of CDC data sources.
The problem is that the shift in the age mix occurred half a year ago. Here’s a headline from May 2021 noting that more than one-fifth of COVID-19 cases were children. And it’s been that way since. Here’s a chart contrasting the share of under-15, under-25, versus over-65, over-85 cases.
The majority of the shift to young people occurred by the middle of 2020. (In fact, the first time I used the phase “young person’s pandemic was July 2020 (in this post about Virginia). The only truly new change in 2021 has been the steadily increasing share of the youngest — shown here as those up to age 15. But if you refer to the chart of mortality rates above, there’s scarcely any difference between the under-15 and under-25 populations in terms of COVID-19 case rate mortality.
When I put that all together, age-adjusting the mortality rate has an impact on the estimate mortality in the very earliest part of the pandemic. That’s when it was (e.g.) sweeping through nursing homes. But of late, the adjustment to a standardized set of ages just doesn’t much matter.
So this remains a puzzle. The low current case mortality rate is not explained by any strong shift in age mix of new cases during the Delta wave. That shift occurred well before the Delta wave got started.
Some of the observed change in the case mortality rate may be purely random variation. The case-mortality rate seems to wander around a bit, for no obvious reason. So, for example, we can see earlier episodes where the case mortality rate was around 1%. So there is some seemingly random variation, and some of that may be occurring now.
It’s clear that hospitalizations have shifted toward younger people.
The Virginia Department of Health has documented the shift in the age mix of hospitalizations for the Delta wave. They attribute this to the very high vaccination rates of the oldest old, which, by and large, is keeping them out of hospitals. January — pre-vaccine — the oldest old accounted for almost half of COVID-19 hospitalizations. Now they account for a little over 20%.
Source: Virginia Department of Health.
In Virginia, the drop in hospitalizations was accompanied by a shift in the age mix of new cases. In January, about 10% of all new cases were age 70 and up. By August, that had fallen to less than 6%. So that all appears consistent. Higher vaccination rate, fewer cases, and fewer hospitalizations among the elderly. (And I also note that, at present, Virginia has a very low case mortality rate, which I calculate as less than 0.7%, also consistent with that shift of disease away from the elderly).
Data from the CDC show essentially the same thing, for hospitalizations. Starting in early 2021, the elderly’s share of COVID-19 hospitalizations has fallen. This would be consistent with the generally higher vaccination rate of the elderly.
What’s not consistent is that the CDC’s surveillance data don’t show a shift in the age distribution of all infections. Which is NOT consistent with the higher rates of vaccination in the elderly. And is why the age-adjusted mortality rate hardly changes, above, in the 2021 portion of the pandemic.
And so, I end up by adding a caveat about the data. I derived the age mix of new cases from CDC COVID-19 case surveillance data. (But I normalize the totals to match the NY Time data for total cases.) The CDC data are extracts of brief reports voluntarily submitted by public health departments across the country, summarizing each COVID-19 case. The CDC case surveillance data file is hugely incomplete, both as to count of cases (about 25M in the file, about 38M in reality) and for information missing on the individual data items (such as hospitalization or death). Plausibly, there actually has been a shift in the true mix of cases, but this file might not have picked that up. It’s hard to tell how trustworthy that data source is.
The bottom line is that I can’t really say why the case mortality rate appears to be in decline while case hospitalization rate is rising. You’d think they’d go hand-in-hand. I think that some clue is in the shifting age mix of those hospitalized. But, based on CDC data at least, there’s been no corresponding national shift in the age mix of those infected. When all is said and done, I can’t really make head or tail out of the publicly available data.
Ah, FWIW, here’s my estimate of the age-specific COVID-19 mortality rate for the elderly. This is calculated from the same mix of CDC and other data files as was used for the rest of this section. The age-specific mortality rates are pretty much exactly where they were 11 weeks ago, just before the start of the U.S. Delta wave.