Why deleting followup reports automatically understates vaccine danger signals

Assessing the quality and seriousness of deleted VAERS reports

Some recent comments by readers prompted me to look into the quality of deleted VAERS reports.

We now know that the reason for the deletion is not nefarious (at least I don’t have any reason to think so till now) - they keep the original reports but delete followup reports.

Do deleted reports simply follow "Keep the first" policy?

But this is not a good idea.

For one, this means all outside researchers using VAERS data are not getting the complete picture.

Why all research papers based on VAERS are wrong about vaccine safety

And in the worst case, this includes death reports.

How the VAERS followup process hides death reports

But I had an intuition in a previous article that the deleted reports are also qualitatively different from the original reports.

It is my view that if you move from looking at only the deleted death reports, to also taking into account serious adverse events, and then also taking into account the full narrative text of both the original and the followup report, the sum total analysis you can potentially do will end up being drastically different.

The hidden underreporting factor in VAERS

The reasons are pretty obvious

a) VAERS is not easy to report to - despite what the pro-vaccine community wants us to believe, it is actually vetted. You first get assigned a temporary VAERS_ID, and only get a permanent one after your information is at least nominally verified.

Source

b) if someone still filed a followup report, they were likely already familiar with the process and were still willing to go through the whole thing once again. The additional motivation alone should probably be a factor in the overall report being better.

This also tells us that the more serious a report, the more likely there will be a followup (if needed). Is this intuition correct?

Are followup reports more serious?

There were 882863 US VAERS reports related to COVID19 as of first week of Feb 2023 when I downloaded the dataset.

We will define a report to be serious if one of the following is true:

DIED, HOSPITAL, DISABLE, L_THREAT

Click here to see the definition of these fields

There are 94766 such reports.

So that is 94766/882863 = 10.73% of all reports are serious. Note that this matches the expected estimate.

Source

There were 31098 deleted reports as of first week of Feb 2023 when I downloaded the dataset.

There are 6173 reports which can be classified as serious using the exact same query as before.

That is, 6173/31098 = 19.85% of deleted reports can be classified as serious.

To summarize, 10.73% of original VAERS reports can be classified as serious, while 19.85% of followup VAERS reports can be classified as serious.

In other words, nearly twice as many deleted reports are serious reports when compared to the original reports.

Are followup reports more descriptive?

The average length of original reports is 703 characters.

The average length of followup reports is 1103 characters, meaning the followup reports are on average more than 50% longer ((1103-703)/703=56.8%).

In other words, going through the followup report is likely to be much more useful if you are trying to figure out if the vaccine caused an injury.

Of course, it is not an either/or scenario. It makes a lot more sense to merge the information together. However, I am doing these comparisons to show what we could be missing out when we ignore followup reports.

Should CDC keep the original report or the followup report?

Interestingly, the average length of the narrative text for original death reports is only 604 characters, while the average length of the narrative text for followup death reports is more than double at 1407 characters.

But these are averages over the whole dataset and not based on matched reports.

I also tried to look at the difference in the average length of the SYMPTOM_TEXT in the reports which I was able to match (remember I could only match about 50% of the deleted death reports). There is also a jump here - from an average length of 701 characters to an average length of 910 characters. This is not as pronounced. I will look into this in more detail when I get a chance.

There are probably other ways to prove that deleting (and thus ignoring) followup reports is a bad idea, and if someone has any suggestions please let me know in the comments.