Do COVID Vaccines Work? (part 10)

I utilise data from an unknown NHS Trust in the development of a staged multivariate logistic regression model in the assessment of 14-day efficacy by dose

In part 9 of this series I ended with the following suggestion:

I’m going to refine the model by considering dosing and the time delay between dosing and death. Right now we have a hotch-potch of unvaccinated in-hospital death, dose 1 in-hospital death, and dose 2 in-hospital death with no regard to days elapsed between vaccination and vaccine-acquired immunity.

Yes, we are going to have to suffer even more mind-numbing logistic regression output, but at least this should now be familiar and I’m hoping readers understand where I’m going with all this. In this particular series I’m not trying to provide evidence of increased risk of early death following vaccination. Nope, no, nein, and нет! What I am trying to do is the exact opposite and that is to weedle out evidence of benefit in terms of reduced risk of COVID-induced respiratory conditions. The logic here is if the jabs don’t work then why bother to take them even if it can be shown that they pose minimal risk? But they don’t pose minimal risk, do they? It’s been quite clear for some time that they pose considerable risk that the manufacturers, authorities and media are doing their very best to ignore. Hence heavy censorship across all of social media instead of heavyweight debate.

In all honesty I shouldn’t be hunting and scraping for evidence of vaccine benefit like this, for the situation should be self-evident if claims of 95% or even 80% efficacy are to be believed. By ‘self-evident’ I mean data that comes at us singing a song, doing a jiggy dance and shouting-from-the-rooftops sort of evident. Admittedly, it’s a bit weird looking for benefit amongst inpatients who went on to die but the reason why I use this sombre cohort is because of the fine level of diagnostic detail in the EPR.

Plat Du Jour

For today’s feast I am going to swallow the official narrative hook, line and sinker and give the vaccines 14 days to (allegedly) assemble immunity against ACE2 receptor attack in the lungs by intramuscular injection in the arm. In doing so I am ignoring the curveball thrown by some experts (you can’t solve lung vulnerability via injectables in the arm), and I shall filter the data to permit separate models for those who received just the initial dose, as well as those who received both doses prior to death. Very few boosted deaths were present in the data sample owing to the time frame. The modelling strategy will be precisely as described in part 9.

What this refinement will hopefully do is address any grumbling from the pro-vaccine cult who will insist we shouldn’t be analysing benefit based on the initial dose alone, and that we shouldn’t be analysing benefit after the second dose unless 14 days have passed. I hear you. You can hide a lot of vaccine harm in those 14 days but pro-vaxxers can’t seem to hear that message which, if we think about it, is a decent definition of cult-like behaviour.

An assumption here that has always sat awkwardly with me is that a second dose of the same stuff is going to magically save the day and suddenly yield a fabulous benefit even if the first dose of the same stuff failed to do so. Another decent definition of cult-like behaviour, methinks! It is tempting to place a bet but I invariably find that the best bet is always to see what the data says when the handle turned as honestly as possible. Regular subscribers will know the drill by now… it’s time to get the coffee on the stove and something tasty on that plate!

Logistic Model For Initial Dose Only (14 day delay)

Just to recap on the sample used… I’m looking at adult in-hospital deaths (aged 18 years and above) for the period 2021/w10 – w36 for those: a) who were never vaccinated, or b) who received an initial dose only at least 14 days prior to death (n=2,715). Here’s the tray bake…

This time round I’ve included the results of the Hosmer & Lemeshow test for goodness of fit for those who like extra geek with their nerds. For those who don’t please forgive the intrusion and simply bask in the knowledge that the model passes muster!

A classification table is always a good thing to throw down on the dinner table because it’s the ultimate acid test as to whether the model is doing anything useful. As we may see the model is very good at predicting absence of negative test results (99.0% true negative rate) but is pretty ropy at predicting presence of positive test results (15.6% true positive rate). The lesson we learn here is that COVID Dx, as a dependent variable, is a pretty tricky customer to predict and it’s worth cogitating on this, for it reveals just how ambiguous the real world is.

The bottom table contains the model coefficients, and we need to zoom right in on Vaccinated (OR = 0.54, p<0.001). This is the overall risk of a positive test result for those receiving their first dose at least 14 days prior to death from any cause compared to the risk for the unvaccinated. An odds ratio less than unity indicates lowered risk for the vaccinated cohort, and we’re roughly looking at halving of that risk. Only shareholders and/or the numerically naïve will jump to the conclusion that this is evidence of vaccine benefit. It may indeed be, but it may also arise in many other ways (e.g. inequality in test frequency across groups).

If we want to judge genuine vaccine benefit then this must be done in relation to alleviation of respiratory illness. This is where the interactive terms Chronic Respiratory by Vaccinated (OR = 0.75, p=0.383) and Acute Respiratory by Vaccinated (OR = 1.06, p=0.860) come into play, and we once again fail to find statistically significant results. What is most intriguing are the contrary odds ratios, with chronic diseases falling on the side of vaccine benefit and acute conditions falling on the side of vaccine harm. Quite how folk suffering from asthma, emphysema, COPD etc can benefit from a highly specific mRNA therapy for SARS-COV-2 infection is beyond me!

This doesn’t make clinical sense and I suspect a hefty dollop of sample bias is going unaccounted for. The sort of thing we’re looking at here is reluctance to undertake testing of vaccinated in-patients with a lengthy medical history of respiratory disease coupled with unwarranted enthusiasm to test the daylights out of unvaccinated inpatients with the same. Such bias would be sufficient to sink any battleship and there’s no way of adjusting for this. Questions need to be raised as to what protocols were employed and who devised them. I’m not so certain this smells of fragrantly innocent nursing.

Logistic Model For Second Dose (14 day delay)

Just to recap on the sample used… I’m looking at adult in-hospital deaths (aged 18 years and above) for the period 2021/w10 – w36 for those: a) who were never vaccinated, or b) who received a second dose at least 14 days prior to death (n=3,879). Here’s the second tray bake of the day…

The Hosmer & Lemeshow results indicate this model also passes muster, and once again prediction of presence of a positive test result (true positive) is poor at 15.8%, revealing to the world that the model sits on a vast foundational sponge of essential ambiguity.

This time round the main effect labelled Vaccinated pops out at OR = 0.08 (p<0.001), indicating a thirteen-fold reduction in the likelihood of a positive test result for the vaccinated cohort. What this means is anyone’s guess because we’re talking positive test results only, which have absolutely nothing to say about presence of the (whole) active virus, infectivity, transmissibility or severity of respiratory illness. ‘Tis strange that such a pointless non-entity of a variable should be the yardstick by which all is judged in what is supposedly the realm of healthcare.

Again, if we want to judge genuine vaccine benefit then this must be done in relation to alleviation of respiratory illness. This is where the interactive terms Chronic Respiratory by Vaccinated (OR = 1.13, p=0.698) and Acute Respiratory by Vaccinated (OR = 1.59, p=0.140) pop into play, and we once again fail to find statistically significant results. However, what should catch everyone’s eye at this point is that both odds ratios are now on the ‘wrong side’ of unity; that is, they are both indicating vaccine harm. Whoops!

Not only that but the p-value for Acute Respiratory by Vaccinated is approaching the 90% level of confidence that is often used in medical science owing to the messiness of the field. Hand me another sample of 3,879 deaths and we might just find a statistically significant effect.

This should be raising eyebrows because if that odds ratio of 1.59 is a reasonable point estimate for the real world then we’re looking at a 60% increase in the risk of COVID-associated acute respiratory conditions following the second dose. What we now have is evidence that is not just leaning toward generalised vaccine harm but evidence that suggests that dosing up on vaccines is inducing the very conditions they are supposed to alleviate!

The Next Move

My next move is to throw away the rather ambiguous meringue of a dependent variable (COVID Dx) and replace it with Symptomatic COVID, then revise the model structure accordingly. To pull in the largest sample possible my definition of symptomatic COVID will simply be any in-patient who received both a positive test result and any respiratory diagnosis prior to death. It will be interesting to see how this simplified approach will swing things. Until then…

Kettle On!