The NY Kids Paper

A partial glimpse into the pediatric hospitalization "with/for" distinction in December and January.

A paper looking at “case” and hospitalization rates in New York state provides separate counts for “with” and “for Covid-19” hospitalizations, revealing 2 - 5 times over-reporting.

A paper looking at rates of test-proxied “cases” and hospitalization mentioning “Covid-19” appears to show negative infection efficacy for children aged 5-11 in January.¹ If by “show,” you mean, make incredibly hard to see. The bars here actually correspond to the rate at which unvaccinated children in the same age-groups are experiencing “cases” compared to Covid-vaccinated children of the same age-group who are so-and-so many days away from their 2nd dose, per week:

Never mind. I’m not even going to show it; it’s too convoluted and confusing.² Anyway: the Pfizer / BioNTech “kid-size” Covid vaccine looks bad, though only for a specific slice of children whose 2nd dose was received at a specific time, and only when case rates were waning (so that false positives would be more likely).³

“Own goal,” if we assume that these authors set out to show how wonderfully the emergency-authorized biological experiment was working. It looked like it was causing harm.

Should this interest and / or worry us? I would almost say it’s interesting; but not really worrying.

There are many reasons which spring to mind why these results would simply be the result of selection bias or other factors. If tests were scarce in New York in January, as they were elsewhere, one could naturally see why the same parents who were crazy enough to subject their children to an experimental biological intervention with no proven benefit out of the fear of a virus which posed no threat to said children in the first place would be more likely to go out of their way to acquire a test. The study doesn’t give us any way to estimate or adjust for this bias.

What’s more, it remains unclear whether PCR-positivity is anything like a useful proxy for infection with BA.1 (“Omicron”), especially in children.

Given this and other obvious potential biases (see footnotes⁴), it’s curious that the authors chose to undertake a time-since-vaccination analysis at all. But they did; and the numbers came out “bad;” and there’s no reason to place much stock in them regardless. It was a badly-designed study perfunctorily slapped together from disparate data sets that may have no resemblance to reality.⁵

More interestingly, the authors used two different definitions for hospitalizations, and in so doing they allow a direct quantification of the with / for distinction during the study period.

The “Primary Reason” counts are provided in the supplementary materials, accompanied by an amusing footnote which laments the possibility that some excluded hospitalizations might be “secret” for-Covid-19 hospitalizations - ones that were, you know, for-Covid-19 on the inside:

Dorabawila, V. et al. eTable 3 (“122” in the unvaccinated 12-17 total excludes results before Dec 6)

As we can see, hospitalization rates with “Covid-19” as the primary reason are low for the Covid-vaccinated, but these represent a smaller portion of children in New York vs. the unvaccinated. And in both groups, absolute rates are too low to justify any society-wide fuss over the Virus to begin with. Nonetheless, the “efficacy” of the injections against this rare outcome in children appear decent; and it’s certainly reassuring that there isn’t a “negative” result in this one. No sign of the much-hyped “Covid Vaccine-Induced AIDS” here.

But our concern is not the comparison between the unvaccinated and the Covid-vaccinated, but between these values and the paper’s other hospitalization results, used for the efficacy findings in the main portion of the paper, which represent “with Covid” hospitalizations. The media and expert talking-heads, of course, finally acknowledged the “with” / “for” distinction around the same time as these results were being recorded; it’s still illuminating to be able to compare the raw numbers, for once.

While both datasets involve hospital admissions recorded by the NY Health Electronic Response Data System, the main results used figures for “all new admissions with laboratory-confirmed COVID-19.” The table above, as labeled, only includes admissions in the same system where “Covid-19” is the primary reason.

Overlaying the numbers above with the less precise table used for the main portion of the paper, we can neatly compare “with” (“mentioning”) and “for” (“primary”) pediatric admissions over the entire two-month period in New York state:

Whether this can be generalized to other American healthcare systems, or to the period before BA.1 / “Omicron,” is unclear.

It would be tempting to look at the surge in the over-reporting rate beginning in late December, and conclude that we have been laboring this entire “Pandemic” under a 200 to 250% exaggeration of the true number of pediatric hospitalizations on the part of the experts and the media - in other words, that it is only the increase in incidental positive tests in January that prompted the authorities to admit that the numbers do not correspond to realty. “Let them think it’s twice the real rate, it’s for their own good - but four times the real rate, well that’s too much.”

However, that actual timing of the media admission of the over-reporting - in late December⁶ - makes it seem like the admission preceded the worst of the surge (though it’s at least possible that the increase to 300% was noticed in the last week of December, triggering the media course-correction). Does this mean the 200 - 250% over-reporting in New York for early December was, itself, more than would show if we were looking at the older numbers?

That will have to remain a mystery.

1. Dorabawila, V. et al. “Effectiveness of the BNT162b2 vaccine among children 5-11 and 12-17 years in New York after the Emergence of the Omicron Variant.”

2. And the bars aren’t even labeled!

Dorabawila, V. et al. Fig. 2.

Blue bars are unvaccinated 5-11 year olds vs 2nd-dosed 5-11 year olds at given time-since-vaccination values; orange bars are same but for 12-17 year-olds. Orange bars have wider brackets as fewer youth in that age bracket were Covid-vaccinated in December or January (as the injections had been emergency-authorized for those age groups long before then) compared to 5-11 year-olds.

The supplemental materials present the corresponding raw numbers, but removes the infection rates for the unvaccinated. Note that above, rates are for the whole of January (though subjects spend portions of the month in different “time-since” groups). Below, rates are per-week. However, we can see that the “negative efficacy” cohort still ends up being limited to the end of the month, since none of the subjects in either age group includes 2nd-dose-recipients from before December 13.

Anyway, “bravo” to the authors for finding so many different bad ways to present the data.

Dorabawila, V. et al. eTable 1

This reveals another way in which the results may be biased, mentioned at the bottom of footnote 4.

3. See explanation in footnote 2. Going for a very choose-your-own adventure experience with this one.

4. A further bias which may apply here is the “Iceland Dashboard Bias,” so-called because if it has an official name, I don’t know it. But I first encountered it in a disclaimer appended to the Iceland Directory of Health dashboard shortly before they ceased updating it in February. So that’s what I calls it.

See https://www.covid.is/data

The data should be interpreted with caution. The number of unvaccinated individuals is calculated by subtracting the number of individuals who have received at least two doses of vaccine from the size of population according to Statistics Iceland. However, it is unknown whether everyone in the unvaccinated group currently resides in Iceland [here, imagine “New York state”] despite being registered as residents. Those people would not be vaccinated here or tested but would still be counted as unvaccinated. The number of unvaccinated would then be overestimated and the the calculated incidence of infection among the unvaccinated underestimated.

Applying the Iceland Dashboard Bias here, children 5 to 11 might only appear as “fully vaccinated” in the New York State and NYC vaccination databases (NYSIIS/CIR) if they were in the state to receive their second dose in December or January (as these products were only emergency-authorized for that age group in November) - in other words, children who had moved to another state, or were out of state for a long period, and received their Covid vaccine during this time would not likely be registered in the New York databases.

As the “unvaccinated” denominator is calculated based on the census (minus the number of first or second-dosed kids), it includes children who may be out of town during the entire study period. Obviously, some of the Covid-vaccinated kids will have been out of town after registration as well; but the unvaccinated likely have the less realistic (more over-estimated) denominator.

(Partially Covid-vaccinated children were subtracted from both the numerator and denominator for the “unvaccinated” set.)

The time-since-vaccination plot in eTable 1 offers another possible source of selection bias. Already presented in footnote 2, here it is again:

Dorabawila, V. et al. eTable 1 (again)

Rates are higher for the 5 to 11 Dec 13-19 second-dose recipients at every time point in January, vs recipients in the other two weeks. Is this because the protection from the injections is waning? But how did the injections offer protection against BA.1 to begin with?

Perhaps it is instead the case that the Dec 13-19 and Dec 20-26 second-dosed are older than those in the Dec 27 - Jan 2 group (though eTable 2 suggests uptake was staggeringly even across every age value), or otherwise more likely to test positive, and time-since dose itself has nothing to do with the difference in rates. Thus, the values for 35-41 and 42-48 days for the whole of January in fact (mostly) reflects the values for this group alone (whereas all three groups contribute to the other time-point outcomes), and the “negative infection efficacy” could simply be an artifact of whatever confounder separates this group from 5 to 11 year-olds as a whole.

5. In one table, the authors randomly drop the “100,000” from several population values, implying that for some reason New York state has almost no Covid-vaccinated 14 year-olds in the beginning of January, and that an additional ~185,000 12 and 15 year-olds were successfully purged by the end of the month. From the supplemental materials:

Dorabawila, V. et al. eTable 2 (with annotations)

Did any of their calculated infection rate and efficacy values include a similar data entry error? Eh, who cares.

6. See Jacobs, Andrew. “Omicron Is Not More Severe for Children, Despite Rising Hospitalizations.” (2021, December 28.) New York Times.