Content uploaded by André Redert
Author content
All content in this area was uploaded by André Redert on Jul 07, 2022
Content may be subject to copyright.
Covid-19 vaccinations and all-cause mortality - a long-term differential
analysis among municipalities"
André Redert, PhD"
Independent researcher"
Rodotti, Netherlands"
Abstract
We analyse the relation between covid-19 vaccinations and all-cause-mortality in N=340 Dutch
municipalities (17.3M people, ~99% of population), during the entire pandemic period. We do not
use covid-19-attributed mortality, mortality predictions and excess mortality, thereby bypassing
the ambiguities of case-identification and mortality-modeling."
#Municipal demographics such as age, culture and population density are strong confounders
of mortality and vaccine-uptake. We account for these by normalizing results to prepandemic year
2019, where covid was absent but demographics were highly representative for later years.
Normalized to 2019, we found no correlation between municipal mortality in 2020 with vaccination
uptake in 2021, which shows the effectiveness of our confounder accounting."
#We could not observe a mortality-reducing effect of vaccination in Dutch municipalities after
vaccination and booster campaigns. We did find a 4-sigma-significant mortality-enhancing effect
during the two periods of high unexplained excess mortality. Our results add to other recent
findings of zero mRna-vaccine effectiveness on all-cause mortality, calling for more research on
this topic."
Introduction
After the first covid vaccination campaign in The Netherlands, high excess mortality rates were
observed in the second half of 2021. It was known that covid variants could escape vaccine
protection [1], but the excess mortality could not be related to covid. Based on the sparse publicly
available Dutch data on excess mortality and vaccination rates, an early analysis was made
indicating a possible increased mortality in the few weeks following vaccination [2]. This inspired
Dutch parliament in December 2021 to require a more thorough analysis to be performed by the
academic world [3]. As the Dutch government did not provide medical data for this purpose, no
such analysis could be performed. In June 2022, the Dutch central bureau of statistics (CBS)
published a report [4] finding no relation between excess mortality and covid or vaccines, while
still offering no alternative explanation for the excess mortality. The academic world’s response
was critical [5]."
#In continued lack of better datasets, we performed a vaccine/mortality analysis using a dataset
that to our knowledge has not been used yet: the weekly all-cause mortality for each Dutch
municipality reported by CBS [6]. There are approximately 350 municipalities in The Netherlands,
cohorts that are very similar and clearly not suited to isolate phenomena related to age, gender or
health-status. Their usefullness comes from being similar while having slight variations in
vaccination coverage that are random/unrelated to health, e.g. due to vaccination logistics. "
#Figure 1 shows an earlier result that motivated this short report, a strong positive correlation
between vaccination coverage and mortality rate in the 2nd half of 2021. Demographics differ
across municipalities in various ways strongly affecting vaccin-uptake, see Figure 2. For covid,
age is the strongest indicator for mortality [7], which subsequently causes vaccination-
preparedness in the elderly. We will take these confounders into account by extending the
of 1 8
analysis to prepandemic year 2019, in which covid and vaccines were absent but demographics
were obviously very similar."
"
Figure 1: Preliminary result that motivated this report. Municipal vaccination-coverage and
mortality-rate correlate strongly at the end of 2021.
Figure 2: Municipal demographics provide strong counfounders such as average age, culture
(immigration) and population density, all of which affect vaccination uptake and/or mortality. All
correlation coëfficients between quantities shown are in the order of ±50%.
We focus on longer-term effects of vaccination on mortality during the entire pandemic, as by now
it is clear that protective effects of vaccines against covid wane quickly and can even become
negative after about half a year, see e.g. [8,9]. We base our results only on all-cause mortality,
without reference to covid mortality, expected mortality and excess mortality. That way, we
bypass the inevitable ambiguities in covid diagnostics (death with or because of covid) and
mortality-modeling during a developing pandemic. Further, vaccines are known to have effects
beyond their target disease, recently shown specifically for covid vaccins [10], advocating for all-
cause mortality analyses."
Method
We use the following quantities, with the municipality (group), and t a time period:"
1≤g≤340
of 2 8
#All-cause mortality rate (% of municipal population) [6]"
## End-level of vaccination coverage (% of municipal population) [11]"
## Relative municipal population size (% of analysed national population) [12]"
#Proportionality scalar, our main outcome: municipal differental of M per differental of V
During 2019-2022, several municipalities have split and merged. In our analysis, we use only
N#=#340 municipalities that existed during the entire analysis period. This brings the total number
of people in the analysis at 17.3M, about 99% of The Netherlands. The analysis will weigh every
municipality by relative population size (the ’s sum up to exactly 1)."
#We use vaccination coverage at saturated-level reached in November 2021, for full-vaccination
of people aged 18+, rounded off to integer-valued percentages of the municipal population [11]. It
has municipal-weighted (national) mean and standard deviation of 82±5%. Non-rounded
data and/or coverage at earlier dates are not publicly available. The later booster uptake (as of
end of May 2022 [13]) correlates 96% with full-vaccination, see Figure 3, and is not additionally
informative in our analysis."
"
Figure 3: Booster coverage correlates very strongly with full-vaccination coverage, and is not
further informative in our analysis.
The essence of our approach is to regard all-cause-mortality as an N-dimensional vector
with a component for each municipality. One is free to transform any vector by rewriting it as a
weighted sum of N orthogonal basis-vectors:"
$(1)
Each of the base-vectors can be seen as a “pattern” of mortality across all municipalities, with
each the amount of that pattern occurring in or contributing to the total mortality pattern .
The first base-vector is 1g which has every component equal to 1, or equivalently, a uniform equal
Mg(t)
Vg
wg
K(t)
wg
wg
Vμ
Vσ
Mg(t)
Mg(t) = K1(t)⋅1g+K2(t)⋅ ΔVg+K3(t)⋅. . . + KN(t)⋅Nth$base$vector
K
M
of 3 8
spread of mortality-rate across all municipalities. We choose the 2nd base-vector as , the
municipality’s deviation in vaccination coverage compared to the national weighted mean :"
$(2)
Vectors 1g and are orthogonal (their weighted inner product is zero)."
#The rationale of this approach is that the mortality rate is expected to be a quite uniform
pattern over all municipalities (first base vector 1g ), and that the remaining mortality variations will
be distributed among all other N-1 base vectors. If there is a signal present in mortality correlating
with vaccine coverage, it will manifest specifically in while other influences on mortality are
thinned out over all remaining N-2 (= 338) base vectors, improving the odds of detecting relevant
events."
#We do not have to define any of the other base-vectors 3 to N, as they are not needed to
compute the first two ’s. Solving (1) gives trivially with and defined
similarly as in (2). The is our main and only focus, from now on we will omit the 2. We obtain
its value by:"
$(3)"
The source data for [6] is given in weekly t periods. As is linear in , we can easily
recombine weekly computed values to longer or smoothed periods. As weekly data is highly
volatile, we will smooth values to a month-period by a centered, weighted moving average of 5
weeks (weights are 0.1, 0.25, 0.3, 0.25, 0.1). This lowers volatility but leaves the unit of
unchanged at mortality-rate/vaccination-coverage per week."
#We will compute from Jan 2019 to April 2022. Values before vaccination refer to correlation
of mortality with vaccination-preparedness, while values in weeks after vaccination refer to
correlation of mortality with vaccination-coverage. In the absence of confounders, an effective
vaccine will result in < 0 in the period after vaccination. Before vaccination, however, the
expectation of an effective vaccine will result in > 0, as vulnerable people are more vaccination-
prepared."
#Age and other demographics are strong confounders for mortality, covid, and vaccin-uptake,
which all bias . To detect significant values/events in , we compute mean and standard
deviation over all weeks in prepandemic year 2019, and use that as z-score :"
$(4)"
Clearly, in 2019 covid was absent but municipal demographics were very representative for the
next few years."
ΔVg
Vμ
ΔVg=Vg−VμVμ=∑
γ
wγVγ
ΔVg
∑1γwγΔVγ
M
K2(t)
K
K1(t) = Mμ(t)
Mμ
ΔMg
K2(t)
K(t) =
∑γΔMγ(t)wγΔVγ
∑γΔVγwγΔVγ
Mg(t)
K
ΔM
K
K
K
K
K
K
K
K(t)
Kμ
Kσ
Z(t)
Z(t) = K(t)−Kμ(2019wk1..52)
Kσ(2019wk1..52)
of 4 8
#One must bear in mind that refers to a ratio of differential mortality-rate and differential
vaccination-coverage, and one cannot extend it to an absolute relation . The differential
effect is probably caused by a limited part of the population for which the vaccine has a
substantial effect (elderly, people with comorbidities, youngsters/adults with unlucky susceptibility
to adverse effects). When this part is accounted for, a non-linear saturation effect on mortality may
occur, for which our linear model (1) does not account."
#Our method is mathematically equivalent with computing weighted trendlines between
and , with K the resulting regression coefficient. It could very well happen that at some periods
covariances between and are relatively small (<20%) and that so-called modeling-
strength R2 is in the order of a few percent. In such cases, it would be false to conclude that the
trendlines K are thus unreliable. In fact a low R2 would just reflect the relative size of a mortality-
vaccination effect amidst a much larger overall mortality rate. We will not compute R2 or similar,
instead we will use the z-score with and to determine event significance."
Results
Figure 4 shows , and an example for the period of week 50 of 2021 to get an impression
of the municipal variation."
"
Figure 4: Municipal differences in population size, vaccination coverage and mortality rate
(example for week 50 of 2021). A subset of all 340 municipalities’ names appear on the left.
The weights have (unweighted) average 1/N (~0.29%). As there are clearly a few very large
municipalities (Amsterdam, Rotterdam, The Hague, etc) one may wonder if these will dominate the
results, effectively lowering N. The effective number of same-sized-groups in our analysis,
however, is still in the order of hundreds:"
$(5)"
K
M=K V
ΔM
ΔV
ΔM
ΔV
Kμ
Kσ
wg
Vg
Mg
wg
Neffective =e−∑gwgln wg≈209
of 5 8
For the example week 50 of 2021 we find:"
$(6)"
This municipal-average mortality rate corresponds to the ~4k deaths registered by the Dutch
CBS [14] (0.023% times the full Dutch population of 17.5M people)."
#The value in this period after vaccination is positive, while a negative value is expected for an
effective vaccine, in the absence of confounders. It is even about the same size as reflecting
at least a huge impact of confounders such as age. Further, the value of refers to differential
mortality per differential vaccination coverage, with the latter having of just 5%. The net
result is that vaccination-correlated mortality-rate is 5% of average mortality rate , or approx.
200 people in this one week. These are not 200 additional deaths related to vaccination, it means
that 200 out of 4k deaths were distributed over municipalities in the same pattern as vaccination
coverage."
#Figure 5 shows our main result, weekly values (moving monthly averages) during the
whole analysis period from Jan 2019 to Apr 2022, with our z-score using and computed
over prepandemic year 2019:"
$(7)"
For context, national prognosed, actual all-cause, and covid mortality-rates are shown, plus
primary-vaccination and booster campaigns."
"
Figure 5: Main result, K from Jan 2019 to Apr 2022 by z-score Z defined over prepandemic year
2019. Added for context are all-cause mortality, mortality prognosis (from 2020 onwards),
reported covid deaths and vaccination-coverage over time.
Mμ(2021wk50) ≈0.023 %
K(2021wk 50) ≈0.024 %
Mμ
K
Mμ
K
ΔVσ
Mμ
Z(t)
Kμ
Kσ
Kμ(2019wk1−52) ≈1.68x10−4≈0.0017 %
Kσ(2019wk1−52) ≈3.25x10−5≈0.0003 %
of 6 8
From this graphic, we observe:"
•Mortality in the 1st Wuhan wave correlates extremely strong (>10 sigma) with vaccination-
preparedness, in contrast with the 2nd Alpha wave (Z ~ 0 on average, short peak at 2 sigma)
and the seasonal mortality-peak at the beginning of 2019 (-1.5 sigma)."
•After the Wuhan peak, Z is significantly below zero (approx -3 sigma) for many months and
similar for Alpha wave (at -1 sigma). The average of Z over 2020 is very close to zero (~ -0.04) ,
1
meaning mortality over the entire covid-year before vaccination does not correlate with
vaccination-preparedness."
•After the vaccination campaign, Z rises to +4σ at the 3rd Delta wave, then briefly dips between
Delta and Omicron waves during the booster campaign, to rise directly again to +4σ during the
much milder Omicron wave. The rises coincide with both unexplained excess mortality periods."
Conclusions
In the absence of publicly available high-quality covid/mortality/vaccination datasets in the
Netherlands, we examined the available data of municipal vaccination-coverage and weekly all-
cause mortality in a differential approach. Although municipal variations are relatively small, the
amount of municipalities is sufficiently high to find significant correlations. We accounted for
strong demographic confounders (e.g. age, culture, population density) by normalizing all results
to prepandemic year 2019."
#Municipalities that suffered the highest mortality during the 1st covid wave beginning of 2020
were most willing to be vaccinated later in 2021 (10 sigma). The 2nd wave in 2020 as well as the
seasonal mortality peak at beginning of 2019 did not contribute to vaccine-preparedness. Directly
after the 1st and 2nd waves a prolonged municipal mortality-rate reversal was observed, with a
net result that vaccination coverage is not correlated with total mortality from pre-vaccination year
2020. This absence of correlation reflects the effectiveness of our approach to account for
mortality-vaccination confounders, enabling the following observation to be significant."
#After both vaccination and booster campaigns, we did not observe the negative correlation
between mortality and vaccination expected for an effective vaccine. Instead, during Delta and
milder Omicron waves, correlation was significantly positive (4 sigma), coinciding exactly with the
two periods of excess mortality in The Netherlands peaking in Nov 2021 and Mar/Apr 2022."
Discussion
We could not observe a mortality-reducing effect of vaccines in Dutch municipalities, while we did
find a 4-sigma-significant mortality-enhancing effect during the two periods of high unexplained
excess mortality. These results add to recent findings of zero mRna-vaccine effectiveness on all-
cause mortality [10]."
#Clearly, our study has many shortcomings: we made use of very limited publicly available data.
Our several requests for improved data (non-rounded vaccination coverages at more time
instances) were unfortunately not granted by the Dutch government. The variability of our dataset
Proper calculation of the 2020 overall Z-value involves multiplying the found average of -0.04 by
1
the number of independent Z values in 2020, which is 52 divided by (5) applied to our weekly-to-
monthly-moving-average weights 0.1, 0.25, 0.35, 0.25, 0.1, which is 52 / 4.55 = 11.43. The
resulting Z over 2020 is approx -0.47, well below the significance limit of ±1.
of 7 8
was also limited, e.g. municipal vaccination-coverages have a spread of only ±5%. Our simple
linear model between mortality and vaccines does not capture nonlinear effects or non-uniform
populations. Especially with respect to the latter, our approach is cohortless, while covid and thus
vaccine effectiveness against mortality are highly age-dependent [7]."
#Our main result remains alarming and calls for more research on the effect of current covid
vaccines on all-cause mortality."
References
[1]#T. Kustin et al, “Evidence for increased breakthrough rates of SARS-CoV-2 variants of
concern in BNT162b2-mRNA-vaccinated individuals”, Nature, https://www.nature.com/
articles/s41591-021-01413-7"
[2]#R. Meester, W. Aukema and T. Schetters, “COVID-19 vaccinations and mortality - a
Bayesian analysis”, https://www.researchgate.net/publication/
357032975_COVID-19_vaccinations_and_mortality_-_a_Bayesian_analysis"
[3]#Tweede Kamer (Dutch parliament), https://www.tweedekamer.nl/kamerstukken/
brieven_regering/detail?id=2021Z22246&did=2021D47333"
[4]#CBS (Dutch central bureau of statistics), https://www.cbs.nl/nl-nl/nieuws/2022/25/
oversterfte-in-tweede-helft-2021-hoger-dan-covid-19-sterfte"
[5]#NU.nl, “RIVM en CBS stellen dat coronaprikken niet tot grotere sterftekans leiden”, https://
www.nu.nl/coronavirus/6208120/rivm-en-cbs-stellen-dat-coronaprikken-niet-tot-grotere-
sterftekans-leiden.html"
[6]#CBS (Dutch central bureau of statistics), "Overledenen per week, provincie en gemeente,
week 16, 2022”, https://www.cbs.nl/nl-nl/maatwerk/2022/17/overledenen-per-week-
provincie-en-gemeente-week-16-2022"
[7]#S. Ghisolfi et al., “Predicted COVID-19 fatality rates based on age, sex, comorbidities and
health system capacity”, British Medical Journal, Global Health, https://gh.bmj.com/
content/bmjgh/5/9/e003094.full.pdf"
[8]#H. Chemaitelly et al., “Duration of mRNA vaccine protection against SARS-CoV-2 Omicron
BA.1 and BA.2 subvariants in Qatar”, Nature. https://www.nature.com/articles/
s41467-022-30895-3"
[9]#Y. Goldberg, “Protection and Waning of Natural and Hybrid Immunity to SARS-CoV-2”, New
England Medical Journal, https://www.nejm.org/doi/full/10.1056/NEJMoa2118946"
[10]#C.S. Benn and F. Schaltz-Buchholzer, “Randomised Clinical Trials of COVID-19 Vaccines: Do
Adenovirus-Vector Vaccines Have Beneficial Non-Specific Effects?”, The Lancet preprint,
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4072489"
[11]#Fully-vaccinated coverage in Dutch municipalities, https://data.rivm.nl/data/covid-19/
COVID-19_vaccinatiegraad_per_gemeente_per_week_leeftijd.csv, downloaded November
2021"
[12]#Population size of Dutch municipalities, CBS, o.a. https://www.cbs.nl/nl-nl/maatwerk/
2022/10/voorlopige-bevolkingsaantallen-per-gemeente-1-1-2022"
[13]#Booster coverage in Dutch municipalities, https://coronadashboard.rijksoverheid.nl/
landelijk/vaccinaties, downloaded May 2022"
[14]#CBS (Dutch central bureau of statistics), “In mei oversterfte, behalve in de laatste week”,
https://www.cbs.nl/nl-nl/nieuws/2022/22/in-mei-oversterfte-behalve-in-de-laatste-week"
of 8 8