Thread by @USMortality

Why vaccine efficacy (VE) MUST be calculated by taking into account 'total number of tests', and why calculations based on population are likely misleading/incorrect.

Thread: ⬇️⬇️⬇️

Please like/share/RT!
#Covid #Covid19 #Corona #Coronavirus

@RWMaloneMD @BretWeinstein @stkirsch

I've recently come across @jsm2334's calculation of VE based on epidemiological data, and share of population.

While the calculations by Prof Morris are correct, it does not address several underlying issues.

To make this analysis easier, I'm only going to focus on the 'All ages >12' & 'Vax (NB)' group.

I get the same results, as the Prof, using population:
- VE against death with/from Covid19 of 69.4%

It appears to be stat. significant; However there might be Sample Ratio Mismatch.

My main criticism about this method is that it does not account for several different con-founders, such as age, co-morbidities, natural immunity, etc - the two groups are likely not representing an average population.

In this analysis however, I want to focus on "testing bias".

"Testing bias" occurs, when the two different groups are tested differently.

We have real world indications that, that is what's happening already:

https://twitter.com/justin_hart/status/1418683541580173312?s=20

Recently the German government, also confirmed that vaccinated ICU patients are LIKELY not tested anymore.

Hypothesis: Difference in testing protocols for vaccinated and unvaccinated people might lead to artificially inflated VE.

The correct way of calculating VE, according to CDC is to take into account non-cases. I interpret this as negative tests.

https://twitter.com/USMortality/status/1438916471187329025

However, none of the countries publish 'total number of tests' (or negative tests) by vaccination status, which means we CANNOT calculate VE by the CDC defined method.

But, we can at least see, how different levels of testing for each group would affect the VE calculation.

Based on OWID data, I'm estimating about 3.1M tests performed in the given time frame 8/10 - 9/8.

So assuming all groups (vax/unvax/boost) are tested equal, we arrive at the same results, as before (obviously!)

But are both groups tested equally though?

Let's look what happens, assuming unvaxxed people are test twice as often, as vaccinated:

--> Result: VE drops down to 21.9%

Break even point is at ~2.3x. So if unvaxxed people are tested 2.3x more often, the VE is exactly 0%.

More than 2.3x and VE would be negative!

As I said before, unfortunately this is all hypothetic, since none of the countries publish all this required data. US, UK, Germany, Israel.... None!

I've submitted:

https://twitter.com/USMortality/status/1438742716943257603

So can we be certain that the vaccines are effective, given the limited amount of data?

An alternative way to check for testing bias would be to look at all cause hospitalizations/deaths by vaccine status. If those are similar to the background rate of each group before vaccination, then this would indicate a VE of 0 as well.
Hence I've requested this data too.

You can find the sheet with calculations and sources here:

docs.google.com/spreadsheets/d…

Please share/like/RT and let me know what you think in the comments.