You quoted Fenton et al. as writing that the HVE "could not possibly have explained those spikes in mortality rates". But in Barry's NZ data people with n-1 doses also had huge spikes in ASMR when the nth dose was rolled out (even though there were no cheap tricks involved): sars2.net/moar.html#Excess_ASMR_compared_to_reported_mortality_data_in_New_Zealand. I have tried to get Neil and Fenton to calculate ASMR by month and dose in the NZ data, but I don't think I have been successful.
You observed that "HVE is abnormal in nearly all age groups and for all doses". However there is also a strong HVE in the ONS dataset for mortality by vaccination status.'
You pointed out that the youngest age groups had a fairly strong HVE in the first couple of weeks after vaccination. However in old versions of the ONS dataset for mortality by vaccination status up to the version published in July 2022, there was an extra table which showed the number of COVID and non-COVID deaths by weeks after vaccination and age group. It showed that ages 10-39 had only about 60% the normal number of deaths during the first week after vaccination, even though the number of deaths was already close to the normal range by the second week: sars2.net/stat.html#Plot_deaths_by_weeks_after_vaccination_and_age_group.
You were also asking if there would be HVE for the second to fourth doses. But for example in the ONS dataset for mortality by vaccination status, October 2021 was the first month when a large number of people were included under the third dose. But in October 2021 compared to the general population of England, people with 3 doses had about -50% to -70% excess CMR depending on the age group (apart from ages 18-39 which had about -10% excess CMR): sars2.net/stat.html#Plot_excess_mortality_by_dose. In Barry's dataset there's also a strong HVE for the second, third, and fourth doses: sars2.net/i/moar-excess-week-age.png.
Yes, I think the NZ and the UK data suffer from the same problem. Namely that the vaccination status of an individual does not necessarily propagate through the system if they die shortly after injection.
I'm not asserting this is intentional, (even though it could be) but can easily imagine this happening in busy administrative environments : "Oh look this person is already dead... I don't need to waste my time updating their vaccination status then... next."
----
My biggest concern is really about (data) transparency. None of this is hard if we are given the right data.
The fact that government agencies make it is so difficult (or straight out refuse) to hand over fine grained (anonymized) data, really tells me everything I need to know.
At least in the case of the ONS data and Barry's NZ data, the effect where people have reduced mortality for a few weeks after vaccination is less strong in young age groups than old age groups. So in your scenario, why would younger people who died soon after vaccination be less likely to be missing vaccination records than older people?
BTW I think the reason for your different phases is that different age groups got vaccinated at different times. For example the percentage of vaccination dates in the first 10 days of the month was about 54% in ages 65-69+ but about 25% in ages 80+ (where the date of vaccination is the date of the Monday of the week of vaccination): https://i.ibb.co/yRntsqt/kalev-phases.png.
You can probably get rid of the monthly periodic up-and-down pattern if you plot the mortality rate instead of the raw number of deaths. You can get the weekly number of new vaccine doses by age group from the zip file here: ukhsa-dashboard.data.gov.uk/covid-19-archive-data-download (file vaccinations/2021/vaccinationsAgeDemographics_nation_2021.csv).
> So in your scenario, why would younger people who died soon after vaccination be less likely to be missing vaccination records than older people?
In the ONS data the reduced mortality for the first few weeks in the younger age groups is about the same as in the older age groups. (i.e. 6-7 weeks).
I can't specifically remember looking at this in Barry's NZ data but will have another look soon.
There well could be differences between the younger and older age groups in the NZ data. Each country would probably have different administrative procedures.
---
> I think the reason for your different phases is that different age groups got vaccinated at different times
Are you suggesting the different phases could be the clearest indication of a link between (time of) vaccination and deaths?
However actually in the ONS dataset for mortality by vaccination status, the excess mortality percentages in the people who got vaccinated less than 21 days ago don't seem to be that different between age groups: https://i.ibb.co/84BbTr6/ons-excess-cmr-heatmap-by-month-and-age-and-dose-number.png. For example for the category "Second dose, less than 21 days ago", I got -49% total excess CMR in ages 18-39, -50% in 40-49, -59% in 50-59, -58% in 60-69, -58% in 70-79, -54% in 80-89, and -49% in 90+. I remembered wrong that younger age groups would've been clearly closer to zero like in Barry's data.
Your plots appear to show that even young age groups have a very strong HVE in the first few weeks after vaccination. But I think it's partially because you used the last day of the month as the day of death, so only people who died near the end of the month would be counted as deaths within the first week from vaccination.
So if for example someone got vaccinated on February 2nd 2021 and they died on February 3rd, their date of vaccination in the CSV file would be on February 1st (which was the first day of the week) and the date of death would be February 28th in your system. I don't know how you converted the difference in dates to weeks, but it would presumably be categorized as a death on week 4 or 3.
You quoted Fenton et al. as writing that the HVE "could not possibly have explained those spikes in mortality rates". But in Barry's NZ data people with n-1 doses also had huge spikes in ASMR when the nth dose was rolled out (even though there were no cheap tricks involved): sars2.net/moar.html#Excess_ASMR_compared_to_reported_mortality_data_in_New_Zealand. I have tried to get Neil and Fenton to calculate ASMR by month and dose in the NZ data, but I don't think I have been successful.
You observed that "HVE is abnormal in nearly all age groups and for all doses". However there is also a strong HVE in the ONS dataset for mortality by vaccination status.'
You pointed out that the youngest age groups had a fairly strong HVE in the first couple of weeks after vaccination. However in old versions of the ONS dataset for mortality by vaccination status up to the version published in July 2022, there was an extra table which showed the number of COVID and non-COVID deaths by weeks after vaccination and age group. It showed that ages 10-39 had only about 60% the normal number of deaths during the first week after vaccination, even though the number of deaths was already close to the normal range by the second week: sars2.net/stat.html#Plot_deaths_by_weeks_after_vaccination_and_age_group.
You were also asking if there would be HVE for the second to fourth doses. But for example in the ONS dataset for mortality by vaccination status, October 2021 was the first month when a large number of people were included under the third dose. But in October 2021 compared to the general population of England, people with 3 doses had about -50% to -70% excess CMR depending on the age group (apart from ages 18-39 which had about -10% excess CMR): sars2.net/stat.html#Plot_excess_mortality_by_dose. In Barry's dataset there's also a strong HVE for the second, third, and fourth doses: sars2.net/i/moar-excess-week-age.png.
Yes, I think the NZ and the UK data suffer from the same problem. Namely that the vaccination status of an individual does not necessarily propagate through the system if they die shortly after injection.
I'm not asserting this is intentional, (even though it could be) but can easily imagine this happening in busy administrative environments : "Oh look this person is already dead... I don't need to waste my time updating their vaccination status then... next."
----
My biggest concern is really about (data) transparency. None of this is hard if we are given the right data.
The fact that government agencies make it is so difficult (or straight out refuse) to hand over fine grained (anonymized) data, really tells me everything I need to know.
At least in the case of the ONS data and Barry's NZ data, the effect where people have reduced mortality for a few weeks after vaccination is less strong in young age groups than old age groups. So in your scenario, why would younger people who died soon after vaccination be less likely to be missing vaccination records than older people?
BTW I think the reason for your different phases is that different age groups got vaccinated at different times. For example the percentage of vaccination dates in the first 10 days of the month was about 54% in ages 65-69+ but about 25% in ages 80+ (where the date of vaccination is the date of the Monday of the week of vaccination): https://i.ibb.co/yRntsqt/kalev-phases.png.
You can probably get rid of the monthly periodic up-and-down pattern if you plot the mortality rate instead of the raw number of deaths. You can get the weekly number of new vaccine doses by age group from the zip file here: ukhsa-dashboard.data.gov.uk/covid-19-archive-data-download (file vaccinations/2021/vaccinationsAgeDemographics_nation_2021.csv).
> So in your scenario, why would younger people who died soon after vaccination be less likely to be missing vaccination records than older people?
In the ONS data the reduced mortality for the first few weeks in the younger age groups is about the same as in the older age groups. (i.e. 6-7 weeks).
I can't specifically remember looking at this in Barry's NZ data but will have another look soon.
There well could be differences between the younger and older age groups in the NZ data. Each country would probably have different administrative procedures.
---
> I think the reason for your different phases is that different age groups got vaccinated at different times
Are you suggesting the different phases could be the clearest indication of a link between (time of) vaccination and deaths?
In Barry's dataset during weeks 0-3 from vaccination, I got -52% excess mortality in ages 80-89, -46% in 70-79, -43% in 60-69, -33% in 50-59, -26% in 40-49, and -19% in 30-39. So it got progressively closer to zero in each age group: https://i.ibb.co/VLF2JyY/nz-ppd-excess-mortality-by-age-and-weeks-after-vaccination-heatmap.png (R code: sars2.net/moar.html#Heatmap_for_excess_mortality_by_weeks_after_vaccination_and_age_group). The trend breaks in younger age groups and age 90+ though.
However actually in the ONS dataset for mortality by vaccination status, the excess mortality percentages in the people who got vaccinated less than 21 days ago don't seem to be that different between age groups: https://i.ibb.co/84BbTr6/ons-excess-cmr-heatmap-by-month-and-age-and-dose-number.png. For example for the category "Second dose, less than 21 days ago", I got -49% total excess CMR in ages 18-39, -50% in 40-49, -59% in 50-59, -58% in 60-69, -58% in 70-79, -54% in 80-89, and -49% in 90+. I remembered wrong that younger age groups would've been clearly closer to zero like in Barry's data.
Your plots appear to show that even young age groups have a very strong HVE in the first few weeks after vaccination. But I think it's partially because you used the last day of the month as the day of death, so only people who died near the end of the month would be counted as deaths within the first week from vaccination.
So if for example someone got vaccinated on February 2nd 2021 and they died on February 3rd, their date of vaccination in the CSV file would be on February 1st (which was the first day of the week) and the date of death would be February 28th in your system. I don't know how you converted the difference in dates to weeks, but it would presumably be categorized as a death on week 4 or 3.
> But I think it's partially because you used the last day of the month
True.
Lumping deaths into a monthly bucket is an insult to those who died.
It may make the HVE period look longer, but it also guarantees that no obvious spikes in death immediately following the injection can be found.
As I mentioned in the article, we really need finer grained data.
> but it would presumably be categorized as a death on week 4 or 3
Correct. (I essentially assume all deaths happen on the first day of the following month)