No real change in new case trend. The U.S. is now at 22.2 cases per 100K per day.
Currently the second-worst U.S. COVID-19 wave ever
The U.S. new-case rate has now topped the peaks of every U.S. wave except the third (winter) wave. Best guess, we’re on track to beat that in the third week of August.
Increased deadliness confirmed
The increased deadliness of the Delta variant, relative to prior variants, now seems to be confirmed by multiple independent studies. You can’t see it straightaway in the raw numbers because we’ve vaccinated a larger proportion of the most vulnerable (Post #1178, then Post #1186).
When I last looked at this, there was one formal study that did the apples-to-apples (“risk adjusted”) comparison of Delta versus prior variants. Now, based on reporting in the L.A. Times, there are at least three scholarly studies available, all of which confirm that Delta carries a materially higher risk of severe illness and death. Based on the median of those odd ratios (2.21), in round numbers, for a given individual, the risk of severe illness from a Delta infection is about twice the risk of from the prior variants.
The evidence on likelihood of breakthrough infection (infection after full vaccination) seems oddly mixed, to me. But I suspect that’s largely due to reporting bias (you only see bad news reported) and observational data (what you see reflects a combination of vaccination, self-selection, and behavior). And once I start breaking it down, I see nothing in European or American results to suggest that the clinical trials data is wrong.
The gold standard is a randomized controlled clinical trial. I summarized those results for the Pfizer vaccine in Post #1192. Those show that immunity declines slowly (about 6% every two months, for the first six months), and that overall protection against any symptomatic Delta infection averaged 91% over the first six months. Protection against severe disease (e.g., ending up in the ICU) was 97% over the first six months.
But the reported “observational data” estimates are all over the map. And almost all of the “observational data” estimates that get reported are worse than that, often far worse.
By “observational”, I mean the rate of infection observed among people who chose to (were observed to) get vaccinated. And so, not based on the randomized clinical trial, but, in effect, based on whoever showed up to get vaccinated.
Those measures are never as reliable as those from randomized trials, because people choosing vaccination are going to be systematically different from those who don’t. They might have different levels of risk, different levels of exposure (through their occupations, say), and different behavior regarding risk-taking and use of other COVID-19 hygiene measures.
Let me summarize those factors as saying that the vaccinated may differ from the un-vaccinated due to differences in self-selection, exposure, and hygiene.
Now combine that with “publication bias”. In this case, that means standard fear-based journalism. The news media seek out and promote bad news, because that’s what gets your attention, and capturing your attention is how they make money. The upshot is that any time there is a particularly awful-looking infection rate among the vaccinated, anywhere in the world, you can be sure you’ll see it.
But where the infection rate appears extremely low, you won’t. “Things Aren’t Going to Hell in a Hand-Basket” is a headline that you will never read in the pandemic.
That is, in part, why I keep returning to the reliable Virginia public-use data files, and to calculating my own estimates. Last time I looked, your chance of picking up an infection, if vaccinated, was less than 1 percent of the risk for those who were unvaccinated (Post #1191). That’s far better than clinical trials suggest. And as a result, your chance of seeing that analysis reported in the newspapers is nil.
I can only guess why we see better-than-clinical-trial observed results in Virginia.
One possibility is self-selection. If the virus is largely circulating among the young, but vaccination is more common among the old, then the vaccinated population has less-than-average exposure to infection. You’ll see fewer infections there than you’d naively expect, not because the vaccine works exceptionally well in Virginia, but because the vaccinated population has exceptionally low exposure to infection.
A second possibility is behavior. Plausibly, those choosing vaccination are more cautious about all aspects of avoiding infection, compared to those who refuse to get vaccinated. And so you see fewer-than-expected infections among the vaccinated because they have chosen to reduce their exposure and to implement normal COVID-19 hygiene measures.
But those factors could easily go in the other direction. For example, vaccinating health care workers first means that those initial subsets of vaccinated individuals will face far greater exposure than average. Vaccinating the elderly first means that those initial subsets face far greater risk of hospitalization, if infected, compared to the average. And so on.
So I’m a little hesitant even to tabulate the estimates based on observational data. Certainly not what gets reported in the news, because that will have been selected for bad results only.
The problem is, I think Virginia is the only state that is showing infections by vaccination status. (Even though, as I discussed earlier, it ought to be a piece of cake for states to do this.) At least, when I search by the term-of-art Virginia uses (infections by vaccination status), the only state that appears is Virginia.
At this point, all I can do is punt and refer to the official CDC line. That’s published on this web page. In addition to the clinical trial results, the CDC shows an extensive summary of “real world” (observational) results. There’s nothing that I, as an armchair scientist, can add to that. And their conclusion is that the protection provided by the vaccines against new variants is good, particularly the protection against severe illness. But, again per that review, vaccinated individuals will get infected sometimes, and they can spread COVID-19 when infected, just at a lower rate than the un-vaccinated.
Finally, the newspapers love to report on the raw proportion of new infections that are breakthrough infections, ignoring the high fraction of the population that is vaccinated.
Let me take some recent reporting in Forbes, and do some arithmetic.
In the Netherlands, 9% of COVID-19 hospital admissions were from fully-vaccinated non-immunocompromised individuals. Currently, half the population of the Netherlands is fully vaccinated. The raw numbers suggest:
Relative risk of hospitalization = (9%/50%) / (91%/50%) = ~ 10%. So the raw numbers there suggest that the vaccine is only 90% effective in preventing hospitalization. To me, that’s well within what you might plausibly see, at one point in time, with observational data, even if the true relative risk is actually 3%.
For Great Britain, it has been reported that 40% of recent hospital admissions were vaccinated individuals. But while that gets reported, what is often omitted is that the 40% figure includes the partially-vaccinated (those with just one shot). Note that one shot is not very effective against the Delta variant, and that Great Britain chose early in the pandemic to prioritize getting one shot into more arms, rather than to get two shots into fewer arms, so they have a higher proportion of their population that is partially vaccinated, relative to other countries.
Apparently nobody in the British Ministry of Health every broke that down further, so let me take a guess, based on the measured effectiveness of one and two shots at preventing hospitalization. Eyeballing all the numbers, I’d guess that about half those admissions among the vaccinated are for the fully-vaccinated population, a population that now accounts for 72% of the British population.
Relative risk of hospitalization = (20%/72%) / (80%/28%) = ~ 10%. So, again, the rate based on the observational data is well within what I might plausibly expect given that the true relative risk is 3%.
Detailed reporting by the British government also notes that vaccinated individuals who are hospitalized have much milder cases than those who are not. And so the simple risk of hospitalization overstates the likelihood of having a severe (e.g., ventilator-dependent) hospitalization. Again, that’s good news, so it’s something that you won’t see reported in the popular press.
The upshot is that you’ll see a lot of sensational numbers presented, but if you do the math, and understand that these are observational data, there’s nothing that suggests that the original clinical trials estimates were wrong. Near as I can tell, the COVID-19 vaccines provide good protection against symptomatic infection (estimated 88% effective), and very good protection against severe disease (estimated 97% effective).
Better late than never?
I will note that it’s far too late for new vaccinations to have any effect on the course of this U.S. Delta wave. That’s almost 100% true for an individual, and certainly true for the U.S. as a whole.
Or at least, I sincerely hope it’s too late. If not, we are in for a rough ride.
- One shot doesn’t do much to protect against Delta. I believe the most recent estimate is something like 30% effectiveness two weeks after the first shot.
- It takes about six weeks after the first shot to achieve peak two-shot immunity.
- We are only capable of vaccinating a few million people a day, at peak.
And so, given the current rate of growth of new cases, we all sincerely hope the Delta wave has peaked long before any new vaccination push could possibly have an effect.
And, separately, let’s face it. The people who haven’t gotten vaccinated yet aren’t exactly lining up to volunteer for it now. And the populations for whom vaccination might plausibly be mandated are a drop in the bucket.
And so, the British have managed achieve full vaccination in 72% of their adult population, and at least one shot in 88.4% of adults. The comparable figures for the U.S.A. are 60% and 69%.
In my best grumpy-old-man style, it’s a pity the vaccination isn’t an Olympic sport. Perhaps then more people would be concerned that we’re losing this race.
Technical note: State data reporting is now so degraded that I need a better gap-fill algorithm
Today the trend seems to have fallen a bit, but that’s mostly an artifact of spotty data reporting by the states.
Just a few weeks ago, states would either report their case count, or report nothing, which was then interpreted as zero new cases. I worked up a gap-fill to catch those zeros and gap-fill with the most recent “real” seven-day moving average.
That is, when a state reports no new data, I use that state’s trend from the last day on which they actually reported new data as my best guess for what the true number for that day would have been. And then, when the state finally does report new data, I overwrite that with the actual count.
But now, many states have taken to reporting trivial numbers of new cases on most days, instead of zero. And then having “catch up” days where they report all the cases that weren’t reported on the prior days. My algorithm doesn’t catch that (it only catches the zeros) and so that under-reporting passes directly into my estimates.
As a result, I have a bias toward showing too little growth on the days when a lot of states report either zero or trivial cases. I think I’ll fix that for the next round of estimates.