10
This morning, after a trip to the restroom, the youngest shared with me an epiphany regarding the nutritional efficacy of corn. I complemented his observational skills, and suggested that those skills may also be of use determining covid vaccine efficacy out of the publicly available covid vaccine breakthrough data.
You know when a retriever looks at you and tilts its head with that retriever-ish confused look? Turns out kids can also do that.
Nevertheless, we’re going to see if we can extract any kernels of truth out of the existing state data we’ve uncovered so far. My intent actually here was to lay all 5 states we’ve looked at so far on top of each other, and try to make some sense of the ensuing mess. I ended up bogged down in Oregon again though. There are certainly worse places.
Part of what’s making this interesting is the data is changing underfoot. Now to be clear, states are quick to point out data is provisional and subject to change. They are providing it near real time, and in some cases correcting as they go. A very good illustration of this can now be found in the Oregon Breakthrough Reports1 over time. Below is a plot of the effective vaccine efficacy over time calculated from the data presented in the breakthrough reports from 1/13 to 2/27 (I think in all cases Table 1 in each).
There’s more description of the efficacy, Ev, in previous posts here2, but I always like to have the equation in front of me while looking at these:
So this is what this quantity looks like, calculated from each week-to-week set of Oregon numbers:
Each dated colored line presents the data up to that report. Obviously each week we progress, we add another point to the end of the plot, but we’re also seeing the prior dates get updated as we move forward. In Oregon, there is evidence they are following up and updating the data with their latest understanding each week. WE NEED MORE HONESTY LIKE THIS!
The earliest version of the data in the above plot is light blue, latest in magenta. Some observations/comments:
In the first 4 sets of points — 1/13 -2/3, “vaccinated” and “unvaccinated” case counts are the only two quantities provided. There are small changes in numbers from each week’s report, generally handfuls of people being added to each category.
The later two sets, 2/10 and 2/17 are quite different. In these last two reports Oregon is adding a new column “Cases with known vaccination status”.
Prior to this — we surmise the two categories really were “vaccinated”, and “unvaccinated AND unknown status”. This will bias Ev high, as some number of vaccinated cases get counted as unvaccinated. See California3…
For the latter two sets of data then, we are calculating Ev based on the KNOWN status in either case, now that we have them — vaccinated (status known) and (known status - vaccinated = known unvaccinated). <insert Rumsfeld quote here>
Further, we are seeing as time progresses, more people appear to move from the unknown category into vaccinated than unvaccinated. Look particularly at what happens on January 8:
Earlier versions of the data have Ev as high as 75%, and it progresses significantly downwards as the data updates. That means more vaccinated breakthrough cases are being identified out of the total as time progresses.
With the latest data release we appear to see that the vaccine efficacy goes negative on January 8. This could mean vaccinated people got covid at a higher rate than unvaccinated people.
There seems to be a recovery — we have discussed this in past posts4, that could mean covid — in this case omicron — is more infectious in vaccinated people, hitting them first harder, and the recovery comes from infections in the rest of the population “catching up”.
Rates of known status are in the 70% range in January, but improve to the 90’s in the last two weeks with the smaller case loads. We could actually use this to get some idea of an uncertainty on the Ev we calculate. Lets try this on this last set of data:
What I’ve done above now is use the latest data as given in the 2/17 report as the center blue line — it is the same as the magenta line in the top plot. But then what I have done is taken the “unknown vaccination status” population each week and place it entirely either in the unvaccinated (top dashed curve) or vaccinated (bottom dashed curve) groups. This gives an envelope that the true value could live within if all cases vaccination status were known (assuming there aren’t changes/errors in the already assigned vaccination statuses).
The envelope is wider in January, where only 70-ish% are known, and tighter in the last two weeks, where known fractions are in the 90%’s. There is no envelope here prior to 12/25, because we do not have “% known” data available to define it. Is the blue curve as high as it is there and earlier because a different variant was active (delta)? Or because, as the upper dotted line might hint at, all unknown statuses were counted as unvaccinated? Personally I’d lean more towards the difference being delta variant — case loads were quite a bit lower in fall/winter 2021 than during omicron, so if backlogs in following up and determining status were due to sheer numbers of cases, a backlog would seem less likely there. Actually no — some combination of both. It would be interesting to see “known status” numbers now going back further.
Statistically speaking, the dashed line envelope I have here is definitely an overestimate of the uncertainty on the center blue line, given we’re talking thousands of people in each of these points. Statistical errors (as in sqrt(N)/N) would be in the smaller single digit % range. On the other hand, there are definitely significant non-statistical, systematic, effects at work here. For example, looking at the trend in week to week changes between the different colored lines in the top plot, there is what looks like the unraveling of a significant bias. The trend appears to be moving from erring significantly on the side vaccine efficacy — unknown being lumped in entirely with unvaccinated — to perhaps now revealing a significant lack, perhaps even anti-efficacy.
We are seeing as more data is made available that efficacy of these vaccines is nowhere near what we expect when one says “vaccine”, or expected when these things were promised over a year ago. Whether the negative number we came up with today in one of these weeks holds out is unclear, uncertainties are still quite large on the data we have in hand. We’ll continue to poke around and find other sources to try to compare to and try to pick the kernels out of…
That the State of Oregon appears to be moving toward a more transparent presentation of their data is encouraging and to be applauded. That we were, and in I expect in most states are, in a place where data is presented with biases that artificially inflate the efficacy of these injections is a very sad place to be. To evaluate whether a medical practice is advisable or not requires accurate knowledge of both its safety and efficacy. You need to be able to weigh the risks and benefits. We though seem to be in a dark place where
These vaccines have been presented as more effective than they may be5.
These vaccinations are presented as being more safe than they may be6.
Don’t ask questions, take them or lose your job7.
Are we though starting to see the sunrise peeking through the shades?
The last of a series of these I’ve put together is here — earlier references are within that:
The original data in question is archived here:
https://www.oregon.gov/oha/erd/pages/covid-19-news.aspx
In the menu under “COVID-19 Data Reports and Projections” you find a menu, under which are the weekly Breakthrough reports — i.e. the latest as of this writing is “Breakthrough Cases Report 02.17.2022”. I’ve incorporated archived reports back to August, but am particularly illustrating the differences from 1/13 to 2/17.
Maybe something more complete is buried in here:
Ha ha — infinite loop! GOTO 10
I mean just today:
https://www.whitehouse.gov/briefing-room/speeches-remarks/2021/09/09/remarks-by-president-biden-on-fighting-the-covid-19-pandemic-3/
Can you be sure that the "efficacy" in the first few weeks isn't simply due to the misclassification of vax statuses? Over time, the misclassification disappears (as those vaxxed within 2 weeks) get moved from the unvaxxed to the vaxxed where they should have been all along). If you control for that, I doubt you'd get any efficacy at all. And for infections, you wouldn't expect to either since it is wedll established by now that the "vaccines" to do not prevent infections.