[conspire] COVID-19 update
Rick Moen
rick at linuxmafia.com
Tue Sep 7 00:36:09 PDT 2021
For those of us hoping to peer into the future without waiting for it,
https://calcat.covid19.ca.gov/ is an interesting time waster^W^W site.
It does modeling of COVID19 trends for California as a whole and
for individual counties. "Modeling" means using predictive models
to guesstimate the near future. I often visit the "Nowcast" pages,
to gaze into the crystal ball about progress, or lack of same, in
getting over the pandemic.
The graphs and figures presented highlight a composite/average of six
credible data models about case data, etc. All six have a decent track
record, and notion of averaging them is... well... reasonable. And, so,
a nod to prediction being difficult, especially of the future, having
been made, how do we seem to be doing?
In short: Wow, pretty well. Statewide, number of infections are
heading down. In the Bay Area, more so.
R0 in Contrast to R-eff:
-----------------------
For details, I need to explain two epidemology metrics, R0 (pronounced
"R naught") and R-effective ("R-eff").
R0 aka basic reproduction number is a measure of a pathogen's abstract
infectiveness when it hits a population having no disease resistance.
It's defined as the average number of secondary infections an infected
person would tend to cause, before that infected person ceases to be
infected, either through recovering or dying. Famously, about the most
infectious pathogen known is the viral disease measles, its R0 being
somewhere between 12 and 18.
It's important to note that R0 is the _potential_ infectiousness
of a pathogen absent any disease resistance or countermeasures, where
a population is unprotected and also does nothing to deal with the threat.
Disease R0 Herd Immunity Threshold (HIT)
------- -- -----------------------------
Measles 12-18 92-94%
Chickenpox 10-12 90-92%
Mumps 10-12 90-92%
Rubella 6-7 83-86%
COVID-19 Delta 5-8 80-88%
Polio 5-7 80-86$
Pertussis 5.5 82%
Smallpox 3.5-6 71-83%
COVID-19 Alpha 4-5 75-80%
COVID-19 original 2.39-3.44 58-71%
SARS 2-4 50-67&
Dipheria 1.7-4.3 31-44%
Common cold 2-3 50-67%
Ebola 2014 1.44-1.8 31-44%
Influenza 2009 1.34-2.04 25-51%
So, we now move on to R-eff, which is a real-world, contextual
measure, of how things _actually_ are going, in a particular place and
time, R-eff adds in the meliorating effect of both vaccine and
natural immunity and other countermeasures, such as (for respiratory
pathogens) social distancing, masks, reduced travel, etc.
So, for example, in a local community, if 95% of the local population
have measles immunity because of widespread application of the very
effective MMR vaccine, or prior infection, then R-eff will go below the
magic 1.0 line -- thus, herd immunity.
Whenever or wherever R-eff for a pathogen is below 1.0, by defintion
that means any outbreak is going to die out over time -- fewer and fewer
infected until nobody is.
Local R-eff:
------------
Good news: R-eff for the Delta variant is (recently) well under 1.0 in
California, and better than that in the Bay Area.
Quoting https://calcat.covid19.ca.gov/ "Nowcast" modeling:
Where R-eff estimate
----- --------------
California statewide 0.9
San Mateo Co. 0.88
San Francisco Co. 0.8
Santa Clara Co. 0.97
Alameda Co. 0.87
Contra Costa Co. 0.87
Marin Co. 0.92
Solano Co. 0.94
Napa Co. 0.99
Sonoma Co. 0.85
So, for example, the "Nowcast" graph's trend line for San Mateo County
(my county) is down, down, down, sharply -- which is exactly what we
want to see -- despite Delta's high infectiousness (R0 estimated about 6.5).
So, California as a _whole_, and major metro areas including the Bay
Area, is beating the infection. Rural counties are a mixed bag.
Real-World County Metrics:
--------------------------
That's modeling, though. How are the _real-world_ stats? Cases?
Hospitalisations? ICU beds? Deaths? Vaccination rates?
https://www.smchealth.org/ has... stuff. {sigh}
https://www.smchealth.org/data-dashboard/county-data-dashboard has cases
by date. The Delta wave in July is obvious, and peaked on August 4th
with 163 lab-confirmed cases for the county. It's been steeply
declining since then, the most recent daily total being 16 cases on
Sept. 1st. (Based on later readings, I think this is ICU only.)
That page states total cumulative county COVID-19 deaths to date (600),
but makes it difficult to track trend, so I'm punting on that.
https://www.smchealth.org/data-dashboard/hospital-data is a different
view of data, directly from hospitals. Again, things are trending down.
And there were 13 ICU beds with confirmed or suspected COVID-19 patients
on Sept. 2nd. Total hospitalisations of confirmed cases on Sept. 2nd:
13 in ICU, 24 in acute care. The trend is significantly _down_ since
late August.
Vaccination percentage:
-----------------------
For obvious reasons, this is key, and is our likely way out. At
https://www.smchealth.org/data-dashboard/vaccination-totals-locations-data
, the county trumpets the percentage of _eligible_ county residents
who've completed vaccination -- being target-focussed -- but that's not
what matters, but rather the percent of _everyone_ (including people
under age 12 who aren't eligible), who've gotten vaxxed. For that, we
have to calculate. Page says the county population is 774,990, and that
554,461 of us have completed a vaccination series. That's 71.5%.
That's good. And it's not enough for herd immunity. As shown in the
above table, for us to go around without masks and social distancing,
and still not have rising infections, we need 80-88% immunity (given
Delta having R0 = 5 to 8).
To be specific, having 80-88% immunity (either vaccine or natural;
to be discussed further) would keep R-eff significantly below 1.0
_without_ countermeasures like social distancing & masks, i.e., a
return to normal life.
Delta, Boosters, and the Limits of Knowledge:
---------------------------------------------
Here's where things get tricky. One of the troubling questions
is: How _long-lived_ is vaccine immunity from our three approved
vaccines? Five months and then a steep trail-off? A year? Two years?
Five years? Twenty? The data aren't in. How good, on average, is
"natural" immunity from prior infection, and how long does it last?
Those data aren't in, either.
Generically, we expect natural immunity to be somewhat hit-or-miss,
depending in part on whether the quick but rather generic "innate"
immune system sufficed to fight off the infection, or whther the
much slower but more formidable "adaptive" immune system got involved,
and, in the latter case, whether the adaptive system picked up a good
profile on the pathogen. A major reason for vaccines, completely aside
from the "you may get very ill and possibly die, and infect others" risk
of acquiring natural immunity, is that a good vaccine is,
metaphorically, an expert briefing for the adaptive immune system,
teaching exactly what to look for and respond to.
But along came an excellent, but possibly misleading, pair of Israeli preprint
(not yet peer reviewed) papers by physician Tal Patalon and colleagues at
Israel's 2nd largest HMO, Maccabi Healthcare Services (MHS), here:
https://www.medrxiv.org/content/10.1101/2021.07.29.21261317v1.full.pdf
https://www.medrxiv.org/content/10.1101/2021.08.24.21262415v1.full.pdf
Two very good layman's write-ups followed from AAAS, here:
https://www.science.org/news/2021/08/grim-warning-israel-vaccination-blunts-does-not-defeat-delta
https://www.science.org/content/article/having-sars-cov-2-once-confers-much-greater-immunity-vaccine-vaccination-remains-vital
These were very respectable and important studies, _but_ it's important to
know that not everything called a "study" is the same thing. More
specifically this was _retrospective data analysis_, published late
Aug., of computer records from MHS, looking for patterns. The authors'
preprints didn't involve any clinical work from the authors. They
crunched through database dumps using the Python programming language,
to produce their results.
Specifically, they searched MHS's dataset looking at people who'd either
had either Pfizer vaccine series or PCR-confirmed infection during
Jan-Feb 2021, and then searching for infection 6-7 months later, i.e.,
up to August 14, 2021. Of people having _recent_ PCR-confirmed
(July-Aug. 2021) infection, data-mining suggested about 13x incidence of
breakthrough infection after vaccination, compared to having had an
early infection instead of vaccination. Thus, in that sense,
infection-acquired natural immunity seemed markedly more effective than
was 6-7 month-old Pfizer vaccine immunity. It should be noted natural
immunity was _inferred_ from MHS members not getting re-infected.
Nobody got tested for antibodies (because the authors did no clinical
work, only data-crunching).
I mean no disrespect to Dr. Patalon and colleagues in saying that. It
was excellent work. The point, though, is: With no ability to do any
clinical work, what does it mean? The authors inherently had no chance
to check for hidden variables, for example, and they properly
acknowledge the fact, talking about "confounders" that could be muddying
the trail.
Dr. Patalon & colleagues' preprint got a _lot_ of attention, and, e.g.,
the Biden Administration immediately decided on a booster shot for
Americans, and took as _given_ that vaccine immunity was sharply
falling off about five months in. But hold on....
I was still pondering some of the context, such as the fact that Israel
started a major all-population vaccine campaign _early_ (Dec. 2020) with
major emphasis on vaccinating the very elderly and patients with multiple
co-morbidities (hence, weakest immune systems), which alone would
interfere with extrapolating to the USA, when expert critiques started
to arrive, fleshing out some of my doubts about guesstimating
epidemiological probabilities from _merely_ crunching historical data in
an HMO database. Specifically, this e-mailed piece by N.Y. Times's
David Leonhardt after talking to many top experts:
https://milled.com/nytimes/the-morning-is-vaccine-immunity-really-waning-UVKEBnAcnBj_BA4h
The Morning: Is vaccine immunity really waning?
By David Leonhardt
Aug 30, 2021 6:20am
I'll quote just a couple of bits, but the whole piece is worth reading.
At first glance, the Israeli data seems straightforward: People who
had been vaccinated in the winter [Jan-Feb 2021] were more likely to
contract the virus this summer than people who had been vaccinated
in the spring.
Yet it would truly be proof of waning immunity only if the two groups
-- the winter and spring vaccine recipients -- were otherwise similar to
each other. If not, the other differences between them might be the
real reason for the gap in the Covid rates.
As it turns out, the two groups _were_ different. The first Israelis
to have received the vaccine tended to be more affluent and educated.
By coincidence, these same groups later were among the first exposed to
the Delta variant, perhaps because they were more likely to travel.
Their higher infection rate may have stemmed from the new risks they
were taking, not any change in their vaccine protection.
Statisticians have a name for this possibility -- when topline
statistics point to a false conclusion that disappears when you examine
subgroups. It's called Simpson's Paradox.
This paradox may also explain some of the U.S. data that the C.D.C.
has cited to justify booster shots. Many Americans began to resume more
indoor activities this spring. That more were getting Covid may
reflect their newfound Covid exposure (as well as the arrival of
Delta), rather than any waning of immunity over time.
It's often, I've found, difficult to understand paradoxes and fallacies
involving probability sizes, such as "Simpson's Paradox" and "Base Rate
Fallacy", because the illustrative examples you find on, say, Wikipedia
tend to be clear as mud. So, I'll have a go at making Simpson's Paradox
clearer for non-statisticians:
You study the inferred probability of something in a big population, and
your results seem, frankly, weird, improbably high or improbably low
calculated probability. But, on looking more closely at the data, you
discover a hidden variable, a segmentation. There was a hidden reason
why you were actually looking at two very _different_ subgroups. This
is particularly confusing if it turns out that there was a big, systemic
difference between your study group population and your control group
population, but you were unaware of that difference, making a hash of
your comparison. These differences were acknowledged in advance by Dr.
Patalon and colleagues as "confounders".
It's sometimes the case in an experimental study that, once you identify
the hidden variable and control for it, the study's conclusion about to
overall effect disappears or is reversed (thus, paradox).
Example: Researchers studied 1973 U.C. Berkeley graduate admissions.
There seemed to be a dramatic difference in admissions by sex: 44% of
the men were admitted and only 34.5% of the women were admitted. So,
wait, was graduate admission sexist? People were worried, but the
problem with retrospective data-crunching "studies" is you can have
confounders. Researchers stratified the data by students' _majors_,
and, aha! It turned out, that semester, more men had applied to the
"easier to get into" majors and had high acceptance rates. At the same
time, more women had applied to the "difficult to get into" majors, and
had much lower acceptance rates. But, if you controlled for this
difference, the "bias" went away. Most individual majors actually
favoured women or had very similar acceptance rates.
This isn't a rare problem. Not hardly. Especially with no ability to
look at clinical data, because all you have is a database dump.
So: One problem with the Israeli retrospetive studies is what David
Leonhardt noted. Early vaccination correlated strongly with a different
risk and experience profile relative to later vaccination. Lack of
ability to easily control for that in an observational (and strictly
data-crunching) study is an obvious confounder.
Leonhardt also makes a separately telling point:
Sure enough, other data supports the notion that vaccine immunity is
not waning much.
The ratio of positive Covid tests among older adults and children, for
example, does not seem to be changing, [Johns Hopkins epidemiologist
Dr. David] Dowdy notes. If waning immunity were a major problem, we
should expect to see a faster rise in Covid cases among older people
(who were among the first to receive shots). And even the Israeli
analysis showed that the vaccines continued to prevent serious Covid
illness at essentially the same rate as before.
"If there's data proving the need for boosters, where is it?" Zeynep
Tufekci, the sociologist and Times columnist, has written.
Quite. If the Israeli dataset demonstrated Pfizer immunity falling
off rapidly after about five months, as many including the Biden
Administration and CDC have pessimistically concluded, then wouldn't
there have been more _serious_ COVID after five months among early
(Jan-Feb 2021) vaccine recipients than among patients who got vaccinated
later (in the spring), hence had vaccination _less_ than five months
ago? But that is exactly what is _not_ seen in the crunched data.
Leonhardt checked with experts for their best take on what the available
data really mean. He concludes:
1. Sure, vaccine immunity probably wanes modestly over a year. We'll
know for sure later. But a booster soon makes sense for the most
vulnerable and immunocompromised at least, since many have gone since
early 2021 since vaccination, and have impaired immune response.
2. For other folks, it's not clear a booster currently will help
significantly, but OTOH it's likely that a vaccine rejiggered to
better address newer variants _would_. (That's not yet available.)
Leonhardt has other points, so please see -- and he doesn't claim to
have had the last word, but I'd say he had some really good points.
Stuff I Didn't Cover:
--------------------
I said I hope to cover a comparison between natural immunity vs. vaccine
immunity effectiveness. Sorry, out of time, and that would be a long
and contentious topic, but suffice to say, for starters, that the
problem of "confounders" (hidden variables) haunts the famous "13x" claim
in the Israeli retrospective papers.
The most credible estimates of Pfizer/Moderna estimates of effectiveness
against Delta variant remain approximately 90%. Obviously, if this
changes, we go with the data.
About "natural immunity", the problem with setting out to acquire it
is _in part_ that you might not get any, or poor immunity, e.g., if the
innate immune system suffices to fight off an infection, you will get
literally zero antibodies from the experience. In part, the problem
is that, in the process of (trying to) acquiring that immunity, you
could ruin your health, or that of others, or someone die, in the
process. Not recommended. As AAAS put it:
The [Israeli] study demonstrates the power of the human immune system,
but infectious disease experts emphasized that this vaccine and others
for COVID-19 nonetheless remain highly protective against severe
disease and death. And they caution that intentional infection among
unvaccinated people would be extremely risky. "What we don't want
people to say is: 'All right, I should go out and get infected, I
should have an infection party'" says Michel Nussenzweig, an
immunologist at Rockefeller University who researches the immune
response to SARS-CoV-2 and was not involved in the study. "Because
somebody could die."
I didn't cover evidence that _if_ successfully infected by Delta,
a vaccinated person can have the same viral load (measured by
threshold cycle = Ct) as an infected unvaccinated person, because
Delta can hang around in the outer nasal passages for a while even
while having little effect on the vaccinated person's health.
https://www.nationalgeographic.com/science/article/evidence-mounts-that-people-with-breakthrough-infections-can-spread-delta-easily
Note in passing: Evidence is still preliminary; we'll know more
later. But it's a safe bet that the average duration of infectiousness
for an infected vaccinated person will be quite short, relatively, and
it's extremely likely that such persons are far less likely to pick up
an infection.
I didn't cover the reason why R0 for Delta was originally estimated at
8.5 (claimed to be same as chickenpox) on leaked CDC slideware about a
month ago, and got downrated later, but 5-8 is now claimed to be more
likely, and probably about 6.5.
I didn't cover how the Herd Immunity Threshold (HIT) gets calculated, but
it's from Plotkin's vaccines handbook, and described here:
https://www.businessinsider.com/delta-variant-herd-immunity-90-percent-2021-8
[(R0 - 1)/ R0] / % vaccine efficacy
Assume 90% vaccine efficiency, and R0 of 6.5. Then:
[(6.5 - 1) / 6.5] / .90 = 94%
If the vaccine were 100% effective, it would be 84%, as in the table
above. In a real-world population, you would be helped by some degree
of natural immunity over time, too. The "94%" figure would be the
pessimistic one, where you want to make sure there is herd immunity
_even if_ natural immunity isn't a significant help.
More information about the conspire
mailing list