Reprise: Can a covid vax save a life? Not significantly...
...in the frail elderly - the people who most need it - the people we locked down for.
A quick note!
I discussed the Yap et al paywalled paper in my last substack “Can a covid vaccine save a life"?” The Yap paper examined vaccination status of frail elderly and looked for associations with death. The time period of focus was 9 Sep - 9 Dec 2021, when the circulating variant was predominantly Delta.
Recap on my cause: There was Victorian data showing that the oldest age group was less vaccinated than the younger age groups.
I had hypothesised this would either be
a decision for health reasons, given the Norwegian experience of the very frail dying early due to the vaccines,
or, in the case of vaccine believers,
an act of mercy to help someone out of a difficult life - effectively a “Do Not Resuscitate”.
A Queensland paper on a similar aged care theme reported that “50% of those with Dementia (and 42% of people without cognitive impairment) had documented wishes to not receive cardiopulmonary resuscitation”.
As such, in the Yap et al paper (which contained data on the Clinical Frailty Scale), I was hoping to find an empirical value of extreme frailty associated with the choice of aged care residents (or their powers of attorney) to remain unvaccinated.
I was statistically disappointed that they didn’t include this scale in their multivariable analysis to assess it against vaccination status.
Afterwards
In later discussion with lovely commenter “Diana” I dashed around the paper to find a bit more data, and in my hurry, was misled by a mis-statement.
The first big red arrow in the pic below points to “Unvaccinated vs partially vaccinated” which I quickly interpreted as a 0 Dose vs 1 Dose comparison. The Odds Ratio at 0.34 was favourable for the people with only 1 Dose. That surprised me - I’d hypothesised that the 1 Dose were the vax-injured or suddenly-too-unwell to have a second dose, and were likely to be a greater risk for death.
However, the second red arrow points to the words “Minimum of first dose” which actually means 1 or 2 doses, and thus includes all the people well enough and with sufficient life prospects to have had 2 doses (and to have survived them).
But there is quite a lot more detail. We are so used to every published paper reporting negatively against the unvaccinated. In this same form, in written words, Yap et al only reported positively on vaccination, which, as we see by the two cases above, could only be done by adding the 11 Partially Vaxd (1 dose) people to either the Unvaxd (0 doses) or to the Fully Vaxd (2 doses) - a statistician’s fiddle.
But they did also report doing tests for 0 Dose vs 1 Dose (see below). The Odds Ratio of 1.71 implies an unfavourable outcome for the 1 Dose people as originally hypothesised, but the confidence interval was 0.35 - 7.46 so it was not significant.
And they ALSO (our dream), reported testing for 0 Doses vs 2 Doses, found an an Odds Ratio of 0.42 (favourable for the 2 Doses), but with a confidence interval of 0.15-1.12 and was consequently also Not Significant.
Thus we can say that Yap et al found that the Fully Vaccinated in this aged care cohort were not significantly advantaged on death<28 days over the Unvaccinated. Not Effective. Without any control for frailness or ‘do not resuscitate’ orders that may associated with an unvaccinated status.
Now, I can take this at face value and feel very happy on this rare published report, but the stats had been a bit suspect through the paper and I wanted to reproduce these simple Odds Ratios. However, I could only reproduce the first result (green ticks).
I have written to Yap et al for explanation and clarification - no reply yet. If they don’t reply will I write to the Journal? Probably not, lest the Journal remove the paper - I actually like those numbers being published. Maybe the Journal did the fiddle and are waiting for a complaint. But maybe it’s just me… ?
If you’re not statistical we might part here, but just a heads up, there’s some really good stuff to come on this subject. It gets better. Catch you round!
Can any statistical readers advise/clarify…?
I could reproduce the first Odds Ratio and confidence interval, for which I was given all of the data (see green ticks above and the numbers and method below).
I wasn’t given enough data to do the other comparisons directly, but I thought I should be able to reverse engineer the data from the Odds Ratio, here with the 0 Dose vs 1 Dose comparison. But not so…
No integer “x” satisfied those conditions.
Thus I couldn’t check the
0 Dose vs 1 Dose statistics, nor the
0 Dose vs 2 Dose statistics, nor the
0 Dose vs 1 or 2 Dose statistics.
Can anyone see anything?