Do COVID Vaccines Work? (part 13)

I utilise data from an unknown NHS Trust in the development of a staged multivariate logistic regression model in the prediction of acute respiratory conditions

In part 12 of this series I presented a staged multivariate logistic regression model for the prediction of acute respiratory conditions using a bunch of demographic data and clinical indicators for a sample of 19,457 in-hospital deaths over the period 2020/w11 – 2021/w36 using an EPR data dump supplied by an unknown NHS Trust.

The headline result was discovery of vaccine harm, which emerged from the illusion of vaccine benefit once several confounding biases had been accounted for. I then went in pursuit of correcting bias inherent in the coding of COVID status at death as well as COVID vaccination status at death using machine learning techniques to forge what I am calling probable COVID and probable vaccination. I finished my wild foray by developing a variable called PROD (prior risk of death – a proxy for case commonality) that is used in conjunction with the total diagnoses made (a proxy for case complexity).

With the advent of these three new ingredients, along with a proxy for disease prevalence based on the pillar 1 (clinical need) testing scheme, I am ready to roll my sleeves, scrub the table and roll out the pastry for yet another model bake for the prediction of acute respiratory conditions.

A Bit Of Weeding

Before I start turning the handle again there are two loose ends I wish to tidy. I’ll start by excluding those 1,222 cases exceeding 94 years of age at death that were mentioned in my previous article, and I shall exclude the 35 DOA cases with zero diagnosis as discussed in Prior Risk Of Death (part 1). This will keep things clean and reduce my sample size for in-hospital deaths to a nice, round 18,200 for the period 2020/w11 – 2021/w36.

The ingredients

Herewith summary statistics for the 9 independent variables along with the dependent variable (Acute respiratory) for the period 2020/w11 – 2021/w36. We may note that the sample mean for acute respiratory conditions is 0.22, hence 22% of deaths during this period possessed at least one appropriate diagnosis, this representing a modest sample of 3,930 cases. For definitions and explanations as to what these variables mean and how they were derived please see earlier articles in this series.