The "Grand Debunk" is wrong about VAERS

It is also a grand example of skillset deficit

Summary

• The “Grand Debunk” of the “Turtles” book on the Science Based Medicine (SBM) website is written by people who seem to lack basic data science skills

• v-safe free text information was not actively monitored, and you will know this if you had basic knowledge of Natural Language Processing

• “Crazy” reports are a small fraction of VAERS and can be effectively ignored, and you will know this if you have used domain specific Machine Learning models in spaCy

• Followup reports are not added to public facing VAERS, which ends up underreporting the true vaccine danger. You can analyze this using basic data science techniques to map the deleted report to the original report

• VAERS search tools like OpenVAERS provide additional functionality not provided by CDC Wonder. Instead of dismissing these tools, the SBM writers should be asking the CDC why their search tool is incomplete

• Discussion of the Pfizer indemnity contracts shows the appalling double standards of the SBM writers when it comes to criticizing regular people vs criticizing large corporations

There is now a series of articles on the Science Based Medicine (SBM) website dedicated to a “Grand Debunk” of the “Turtles All the Way Down” book.

You might remember my article where I mention that David Gorski (chief poobah of SBM) is wrong about VAERS:

If you read the article, you can see that when I refer to David Gorski, I am also referring to the folks who write articles on the SBM website.

There is a major skillset deficit amongst the people who write articles on that website.

And the “Grand Debunk” article about VAERS is actually a grand example.

I will just take a single paragraph from the article, because it perfectly highlights the skillset deficit.

v-safe is not actively monitored

The article says (emphasis mine)

This is a double-edged sword. It is possible for dishonest actors to encourage false reporting to VAERS at any time, although Dr Shimabukuro, the CDC deputy director, advises us that there is no current evidence of widespread systematic gaming of the system. It is also accurate that this is a “passive surveillance” system, in that anyone is invited to report. However, the system does not actively solicit the recording of adverse events, in the same way V-Safe, a system implemented during the rollout of the COVID vaccines, solicits reports.

But the v-safe system actually scrubbed out the more serious potential vaccine injuries from its check-the-box list.

V-Safe Part 2: Evidence the CDC Purposely Did Not Include Check-The-Box Selections for Myocarditis, Pericarditis, Seizures, Stroke, and Other Known Potential Serious Harms in V-Safe

That could have been OK, because they had a free form text field in v-safe where users could still type in their symptoms.

Except, the CDC did NOT do any text analysis on those free-form entries either.

I vaguely remember that the CDC also claimed that there are too many entries and it would be too hard to anonymize the free text data, and that is why they did not allow outside researchers to access it¹. How convenient!

Skillset deficit 1: SBM writers do not seem to know that it is possible to use text analytics to infer patterns in the free-text v-safe responses.

Crazy reports are a very small fraction of VAERS

The article then says:

Prospective researchers who want to use VAERS should acknowledge that its “open door policy” does allow as many reports as the public is willing to write up, but that appropriate work is necessary to filter out reports that do not make biological sense, are clearly not vaccine related, or are clearly fraudulent — for example the report of a vaccine turning a man into the Hulk.

This is an incomplete summary which leaves out some important details.

First of all, some external researchers did a VAERS audit and found that there is a fairly involved vetting process for VAERS now.

On top of that, you can now use Machine Learning to automatically filter out the crazy VAERS reports. It will take only a few hours to write the code, and a few more hours to actually get a filtered list of genuine reports.

I have covered this topic in some depth already, but the basic insight is this - you can use Machine Learning models like scispacy to identify medical jargon in the text writeup. In NLP terminology, these are called “entities”. By looking at the number of unique entities in the text writeup, you can calculate a “score” for the writeup and decide whether or not it is a genuine or a fake report. Fake/crazy reports often have near-zero entity scores.

Once you do the analysis, you will realize that these crazy/fake reports are such a small fraction of VAERS that you can ignore them.

Skillset deficit 2: SBM writers do not seem to have any knowledge of how ML models can be used to automatically identify genuine VAERS reports²

Interestingly, the first sentence of the cited paragraph is also wrong in the exact opposite direction³!

In fact, I could rewrite the sentence like this:

Prospective researchers who want to use VAERS should acknowledge that its “ignore all followup reports policy” does allow many reports to slip through the cracks, and that appropriate work is necessary to account for these ghosted reports

And this policy of ignoring followup reports sometimes also excludes death reports.

Funnily enough, David Gorski has written “This is a more difficult conspiracy theory to look at because there could be any of a number of reasons why VAERS ID entries are deleted.“

Skillset deficit 3: Between all of them, SBM writers seem to be lacking the very basic data science skills necessary to investigate VAERS deletions. I have been able to single-handedly map thousands of deleted reports to their original reports because I know how to write some Python code to do this. Why don’t they get some help from their tech friends?

VAERS mirrors like OpenVAERS provide features which are not available in CDC Wonder search

The next paragraph displays a comical lack of awareness of the important of user interface design

People who run databases that mirror VAERS (like OpenVAERS) know this – you have to acknowledge that you’ve read the warning on the official VAERS website before you can access the data. But again, I remind the reader that the main goal of the antivax ecosystem is to stir up as much fear as possible about vaccines, without any thought as to whether that fear may have any validity.

So providing a better user interface for search is somehow the same as stirring up as much fear as possible?

Compare the two (CDC Wonder vs OpenVAERS search):

Which one would you prefer to use to search VAERS reports?

Skillset deficit 4: SBM writers do not seem to understand the importance of user interface design. Maybe they also preferred to use MSN Search over Google in the early 2000s⁴?

Also, by mentioning that OpenVAERS is a mirror (meaning an exact replica), the author concedes that the OpenVAERS website does not misrepresent VAERS in any way.

But OpenVAERS is more than just a mirror.

You can actually search for deaths reported FROM a given country using OpenVAERS, and you cannot do that using CDC Wonder.

For example, there are over 2000 reports of death after COVID19 vaccination from Japan.

Yes, I am very much aware that this does not imply all these deaths were caused by the vaccine, but did you know that over 100 deaths were considered to be related to the mRNA vaccine by actual health care professionals in Japan?

This would also be a good time to mention that the CDC stopped publishing VAERS writeups for the EU region (which has high quality writeups because a large majority of the EU reports are filed by healthcare professionals) in Nov 2022.

In turn, this means that you cannot find VAERS death reports for countries in the EU using the OpenVAERS search tool.

The foreign data set was gutted this week in VAERS and the cancer signal was halved, the myocarditis dose 3 response signal was lost and 994 spontaneous abortions/still births were dropped

In any case, back to my main point.

How would you do a similar search using the CDC Wonder tool?

(Hint: it is not possible)

In other words, even as it stands, OpenVAERS search is already an improvement over the CDC Wonder tool.

Does it matter if a search tool has a simple user interface?

Yes, it does.

There are some actual Doctors who are using the CDC Wonder VAERS search, and not realizing that it does not return results for the foreign dataset by country⁵.

Link to article

So how do they expect laypeople to look up this information? Or have the SBM writers just unilaterally decided country-level information is not important?

Instead of dismissing these search tools, the SBM writers should instead be asking the CDC why the “official” search tool is incomplete.

Skillset deficit 5: The SBM writers do not even have a thorough enough understanding of VAERS to realize that it is possible to search on country level data

Why should vaccine skeptics ignore indemnity clauses?

Then the article says this:

Liz Willner, the author of OpenVAERS, theoretically knows to not regard VAERS reports as stand-alone proof of causality, but she and other like-minded people prefer to de-emphasize this. The number of safe vaccines for a dedicated antivaxxer is, has, and always will be zero. These people face absolutely no consequences when someone listens to their advice and contracts a vaccine-preventable disease.

This is so ironic.

It is now well known that the mRNA vaccine manufacturers were given overly generous liability shields from vaccine injuries.

In fact, the liability shield was so horrendously one-sided, India refused to sign a contract with Pfizer. As you can see, Pfizer tried to use the indemnity shield it procured from the US/UK to try and coerce the entire world to accept a similar contract⁶.

Source

In fact, things got so idiotic, at some point Pfizer demanded Argentina hand over military bases as collateral!

Source

If anything, the SBM writers should first protest all the liability shields offered to the vaccine manufacturers instead of complaining about OpenVAERS and the like.

However, the word “indemnity” is not even mentioned on the SBM website⁷

1. I remember reading this on some website or in a published article, but cannot find the actual source. If you know the link to the source document, please let me know in the comments.

2. This strategy can cause false negatives. That is, there are definitely some VAERS reports which are genuine, but have a very short writeup and a correspondingly small entity score. But that only proves my point more emphatically.

3. In fact, I will now go a step further and speculate that the underreporting caused by followup reports is at least an order of magnitude larger than the overreporting caused by fake/prank/crazy VAERS reports

4. How far Google has fallen! Now they are actively censoring articles about vaccine injuries.

5. You can simply select “Foreign” and do the search, but this is problematic because the quality of VAERS writeups varies drastically between different countries

6. In hindsight, looking at the explosion in VAERS reports, it is not surprising that Pfizer wanted ironclad contracts and was even willing to sacrifice a billion-people market. I think the poor SBM writers are all color blind, because they seem to have a very hard time identifying red flags.

7. I suppose it is possible they use a different term for it on their site. But this is now a pretty famous term which also has very specific meaning in the context of the COVID19 vaccines, so it would be a bit strange if they used a different term to describe the same thing.