Nestschutz Nonsense

Matthew 7:9

Winter is coming

Another day, another Covid vaccine study. German (state) media immediately went on board (h/t

Fabian Spieker

): pregnant women, get your booster during pregnancy! Previous vaccinations are useless! Care about “Nestschutz” (maternal passive immunity; I don’t know if there is a better translation of this beautiful German word) for your unborn babies!

This time, the good tidings are in from Singapore. Say what you will about states bordering on the totalitarian, but they have got the data. It seems that almost all information you might want about all babies born is in the system. Which does not mean, unfortunately, that this information will be made available unfiltered for us to study.

What, then, are we being offered? Altogether, 21,927 children were born in Singapore between January 1, 2022, and September 30, 2022. Of these, the authors excluded 1,024 for various reasons (unusual type of vaccine; number of mothers’ vaccinations not 0, 2, or 3; premature births; incomplete data), ending up with 20,903. Note that Figure 1 in the paper mentions 21,903, and this is all you need to know about peer review.

And then the authors go wild: “To isolate the estimated effect of in-utero vaccination and minimize bias in varying infant exposure to SARS-CoV-2, only those whose parents had a confirmed SARS-CoV-2 infection from the infant’s date of birth up to 6 months of age were included. By selecting only infants with definite infant exposure to the virus due to the close contact between parents and newborn infants, we limited the possibility of the healthy vaccinee bias and overestimation of estimated vaccine effectiveness for infants, an important limitation of observational studies investigating maternal vaccine effectiveness. This selection was done independent of information regarding maternal vaccination status.”

I beg your pardon? You are throwing away almost two thirds of your sample (13,611 births), ending up with only 7,292, and that is your explanation? Is your choice supported by the data? For example, what was the rate of infection among babies whose parents were not infected? What can you say about the timing of infections? What if it is the babies getting it during secret baby parties, and then infecting their parents (note that regarding exposure to respiratory viruses, I am sympathetic to

Clare Craig

‘s hypothesis – read her book, or watch her recent discussion with John Campbell)? If this selection was done independent of information regarding maternal vaccination status, and if the criterion is infection, what on earth does this have to do with healthy vaccinee bias? Why do you go to great lengths comparing the included and excluded data points by everything but size of shoe, vaccination status, and infection?

We might stop here because there is no point anymore in trusting the paper’s results, but there is more fun to be had. For 7,120 of the 7,292 data points, the mother was vaccinated – and got Covid anyway. That’s a whopping 97.6%, and definitely larger than the vaccination rate in the population. However, since the criterion was not infection of the mother but infection of the parents, this does not necessarily indicate negative vaccine effectiveness regarding infection. For example, vaccination of the mother might be negatively correlated to vaccination of the father. Another piece of data that is surely available to the chosen, but not to us. We can only ask questions, and questions ask we will. Do I have to mention that I contacted the authors, of course, and got no response so far, of course?

From here it only gets worse, because models are added to the mix. According to the raw numbers, there were 44 baby infections for 172 unvaccinated mothers (25.6%), 68 infections for 578 mothers vaccinated before pregnancy (11.8%), and 1,160 infections for 6,542 mothers vaccinated during pregnancy (17.7%). Translated into vaccine effectiveness (silly as this is due to the restriction to infected parents) this means 1 – 11.8/25.6 = 54% effectiveness for vaccination before pregnancy, and 31% for vaccination during pregnancy. Sounds pretty decent for the vaccines, but not to the authors. What good is higher effectiveness connected to earlier vaccination if you want to sell more vaccines, or justify your government’s vaccine craze?

Weighting (which is a kind of modeling, of course) rushes to the authors’ help. “To adjust for confounding by demographic and socioeconomic factors and stage of the pandemic, inverse probability weighting was performed.” The end result are not numbers of babies but days-at-risk, quite a standard quantity in hazard-rate models. Babies were monitored for 180 days after birth. Neglecting censoring (e.g., babies dying, or dropping out of the study for other reasons), the maximum raw number of days-at-risk is number of babies times 180. Covid infection takes babies out of the “at-risk” group, so for the “my adjustment” column, I assumed infection at halftime on average, and subtracted 90 times the number of infected babies:

The two remaining columns are from Tables 2 and 1 of the paper, and I have absolutely no idea how they came into being, why there have to be two versions, and why they are so different from each other. In both versions, however, the vaccinated-before-pregnancy group is made smaller in comparison to the unvaccinated group. My main suspect among the control variables is timing of infection.

Table 2 seems to be the springboard to further modeling adventures. The “crude incidence rates” (which are not crude due to the previous weighting) can indeed be reproduced (number of infections divided by person-days at risk). Vaccine effectiveness computed from these would amount to 1 – 122.2/174.3 = 30% for vaccination before pregnancy, and 26% for vaccination during pregnancy. Worse than before, and still vaccination before pregnancy trumps vaccination during pregnancy? Who Ya Gonna Call? Cox regression!

This often used and much less often understood model spits out 15.4% vaccine effectiveness (95% confidence interval –17.6% to 39.1%) for vaccination before pregnancy, and 41.5% (95% confidence interval 22.8% to 55.7%) for vaccination during pregnancy. Finally, the authors have something they can crawl to their overlords with: vaccination during pregnancy is the only thing that helps, and its “true” (i.e., modeled) effectiveness is even higher than suggested by the raw numbers. Who cares if the raw effectiveness for vaccination before pregnancy is outside the 95% confidence interval for the “true” effectiveness!

Just when you think that more confusion would hardly be possible, you notice that there is also a Figure 2 (here, decorated with my remarks in red).

Cumulative proportion of infants is plotted against time since birth of infant. The graphs end up at 32.9% (unvaccinated), 29.2% (vaccinated before pregnancy), and 20.6% (vaccinated during pregnancy). Based on these figures, vaccine effectiveness would amount to around 11.5% and 37.5%, so whatever is on display here seems to have run through the Cox regression meat grinder.

For the unvaccinated, the vertical axis appears to be pretty close to what actually happened: counting the number of steps, and considering that some seem to be too large for a single step, we end up around 44. This does not work for the vaccinated-before-pregnancy group.

And how can we explain the numbers-at-risk below the diagram? Yes, they correspond to the proportions on display (1 – 1,357/2,023 = 32.9%), but what is their meaning? If there are 172 unvaccinated mothers in the sample, how can you sleep at night after model-boosting these to 2,023? And what is the connection between the 2,023 (or the 1,357, or any other number in that row) to the figures in the remainder of the paper?

The authors claim that “this report follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.” In this, they are not even wrong! I invite you to go through the STROBE guideline and think about how they would claim to have assessed all the different recommendations.

The authors conclude that their findings “may suggest a need for maternal SARS-CoV-2 vaccination at each pregnancy similar to current recommendations for maternal influenza and pertussis vaccination.” I conclude that vaccines that do not prevent infection of the mother seem to be able to work wonders regarding infection of the baby, that science is in a sorry state, that peer review does not help, and that limited data in a free society are better than complete data in an unfree society.