Will v-safe fail an audit of free text entries?
LLMs can be used to accelerate the audit
Summary:
CDC's safety reports on mRNA vaccines during pregnancy were incomplete
For example, CDC reported no neonatal deaths in their v-safe pregnancy registry study in early 2021
But there are clear examples of neonatal death after Pfizer vaccination in the v-safe free text which happened after the paper got published
Why didn’t the CDC do a followup study and publish results for the entire v-safe monitoring period?
The fact that v-safe data is used as the primary citation to establish safety in pregnancy for the mRNA vaccines, is all the more reason why it is very strange that there is no Wikipedia page about v-safe
There is a lot of incomplete reporting when it comes to v-safe, and there are at least three things you can audit using v-safe free text:
Verifying if symptoms were properly recorded based on free text entries
Accurately mapping MedDRA codes to free text entries
Examining if CDC safety claims extend to the entire v-safe data collection period
Why are health regulators not using LLMs to audit v-safe data?
In a previous article I mentioned why LLMs are going to become the worst nightmare for Big Pharma companies
I will provide another example in this article - an audit of v-safe free text entries.
Why do we need an audit of v-safe free text?
The main paper that discusses the safety of mRNA vaccines during pregnancy was written by the CDC based on the “v-safe pregnancy registry”1.
The paper was written by the “CDC v-safe COVID-19 Pregnancy Registry Team”
The paper states this (emphasis mine):
Among 3958 participants enrolled in the v-safe pregnancy registry, 827 had a completed pregnancy, of which 115 (13.9%) were pregnancy losses and 712 (86.1%) were live births (mostly among participants vaccinated in the third trimester). Adverse neonatal outcomes included preterm birth (in 9.4%) and small size for gestational age (in 3.2%); no neonatal deaths were reported. Although not directly comparable, calculated proportions of adverse pregnancy and neonatal outcomes in persons vaccinated against Covid-19 who had a completed pregnancy were similar to incidences reported in studies involving pregnant women that were conducted before the Covid-19 pandemic. Among 221 pregnancy-related adverse events reported to the VAERS, the most frequently reported event was spontaneous abortion (46 cases).
If you wanted to genuinely “follow the science”, a reasonable follow up question would be: Was this true for the entirety of the v-safe data collection?
Answer: it wasn’t
Here is an example of a neonatal death after the Pfizer vaccine:
And there were probably more.
Why didn’t the reviewers demand follow up over a longer period of time? If the data was all good news, the CDC researchers could have in fact published multiple papers based on v-safe, which would have been good for their careers.
In fact there is no mention of Death in any of the “Symptom Description” MedDRA codes for this registrant. Why? And how often did this kind of (incomplete) MedDRA coding happen?
Were there other cases of death which weren’t coded as Death?
As an example
found that if the relative of the registrant wrote “I died” which can only be reasonably interpreted as “death of registrant” it was still coded as an “Unevaluable event”2.
Three things to audit in the v-safe free text
Here are at least three things I can think of when auditing the v-safe free text information
Were all the symptoms properly recorded based on the entirety of the free text?
For example, we can start by looking at all the “Unevaluable event” codes and ask an LLM to identify if it is a death
Can we accurately map each MedDRA code to the correct free text entry for all the v-safe registrants? This is necessary for any aggregate statistical calculation
For example, we can ask an LLM to map free text entries to the corresponding checkin by looking at the group of MedDRA codes assigned by the CDC3
Do the CDC claims about mRNA vaccine safety extend to the entirety of the v-safe data collection period? (At first glance, the answer seems to be No)
This could be much more involved, and will require us to write some code which parses each section of text in v-safe papers published by CDC and see if the pattern extrapolates over the entire v-safe dataset.
This also leads to a fourth question - why isn’t any health regulator anywhere in the world using LLMs to do a rigorous audit of the v-safe free text information?
Do they actually expect people to believe the health regulators and the wider medical community has already rigorously studied v-safe adverse events?
The fact that v-safe data is used as the primary citation for safety in pregnancy for the mRNA vaccines is all the more reason why it is very strange that there is no Wikipedia page about v-safe
However there were also many people who were just sending in prank notes, so you cannot really be sure without reading the actual text or sometimes looking at the full timeline
Given that the 12th installment of free text entries has not been published, we cannot do this for the incomplete cases. But now we have enough free text that we can do this mapping for a very large chunk of the free text entries
Please see Dr Kevin McCairn findings in a 3yo vaxxed in utero. Terrifying!