COVID vaccination and age-stratified all-cause mortality risk
Spiro P. Pantazatos1,* and HervéSeligmann2
1Molecular Imaging and Neuropathology Area, New York State Psychiatric Institute and
Department of Psychiatry, Columbia University Irving Medical Center, New York, NY;
2Independent Research Scientist, Jerusalem, Israel
*To whom correspondence should be addressed:
This manuscript contains:
Abstract (256 words)
Main text (4536 words, with Methods and Materials in Supplement)
Running title: COVID vaccination and age-stratified mortality
Keywords: Public health, medical ethics, risk-benefit ratio, epidemiology, COVID-19,
SARS‑CoV‑2, vaccine adverse events
Accurate estimates of COVID vaccine-induced severe adverse event and death rates are
critical for risk-benefit ratio analyses of vaccination and boosters against SARS-CoV-2
coronavirus in different age groups. However, existing surveillance studies are not designed to
reliably estimate life-threatening event or vaccine-induced mortality risk (VMR). Here, regional
variation in vaccination rates was used to predict all-cause mortality and non-COVID deaths in
subsequent time periods using two independent, publicly available datasets from the US and
Europe (month- and week-level resolutions, respectively). Vaccination correlated negatively
with mortality 6-20 weeks post-injection, while vaccination predicted all-cause mortality 0-5
weeks post-injection in almost all age groups and with an age-related temporal pattern
consistent with the US vaccine rollout. Results from fitted regression slopes (p<0.05 FDR
corrected) suggest a US national average VMR of 0.04% (0.0244, 0.0474 95% CI) and higher
VMR with age (lower bound estimates of VMR=0.005% (0.0028, 0.0080 95% CI) in ages 0-17
increasing to 0.06% (0.0108, 0.0859 95% CI) in ages >75 years), and 146K to 187K
vaccine-associated US deaths between February and August, 2021. Notably, adult vaccination
increased ulterior mortality of unvaccinated young (<18, US; <15, Europe). Comparing our
estimate with the CDC-reported VMR (0.002%) suggests VAERS deaths are underreported by
a factor of 20, consistent with known VAERS under-ascertainment bias. Comparing our
age-stratified VMRs with published age-stratified coronavirus infection fatality rates (IFR)
suggests the risks of COVID vaccines and boosters outweigh the benefits in children, young
adults, and older adults with low occupational risk or previous coronavirus exposure. Our
findings raise important questions about current COVID mass vaccination strategies and
warrant further investigation and review.
In June, 2021 the US FDA added a warning to Fact Sheets for Healthcare Providers
Administering Vaccines, noting that “reports of adverse events suggest increased risks of
myocarditis and pericarditis, particularly following the second dose and with onset of symptoms
within a few days after vaccination (1).” Subclinical myocarditis may be a partial explanation
for vaccine-induced deaths in men up to age 50 (2–6). A leading cause of immediate death
following COVID vaccination may be thromboembolic events as all the vaccines have been
associated with forms of venous and arterial thrombosis (7–12). The Pfizer post-marketing
safety data which FDA relied on to approve the Pfizer vaccine (marketed as Comirnaty) was
recently released in March, 2022 following a federal court order. It shows that 42,086 adverse
events against the Pfizer vaccine were reported in the first 3 months of the Pfizer vaccine
rollout, including 1,223 deaths and life-threatening adverse events (i.e. 932 hematological and
1,403 cardiovascular events) occurring within a median of 1 day or <24 hours post-injection,
evidencing a causal link between vaccination, death and other severe AEs.1Data-driven
estimates of severe vaccine adverse event rates as well as all-cause mortality risk are critical
for cost-benefit ratio analyses of COVID vaccination in various age groups.
The vaccine clinical trials (~15-20K participants in each arm) and safety surveillance
studies (13) are either underpowered or did not include adequate safety assessments and
follow-up with respect to severe adverse events and death (see Discussion for brief review). In
the US, real-world vaccine safety signals and mortality incidence rates have relied on the
Center for Disease Control (CDC) Vaccine Adverse Events Reporting System (VAERS)
database (14) and the Vaccine Safety Datalink (15). The CDC has used VAERS data to report
a vaccine mortality risk (VMR) of 0.002%2, estimated by dividing the number of reported
VAERS deaths by the total number of vaccine doses administered in the US. However, the
1https://phmpt.org/wp-content/uploads/2022/04/reissue_5.3.6-postmarketing-experience.pdf (see Table 2 and 7).
VAERS has several limitations, including 1) reported incidents are not independently verified or
confirmed to results from vaccination, and 2) it only receives, not collects, reports from
individuals and/or health professionals and organizations and likely suffers from
under-ascertainment/underreporting bias (16). A key limitation of the Vaccine Safety Datalink, a
multisite vaccine safety dataset based on millions of US medical records (15), is that previous
or raw versions of published datasets are not publicly available to outside researchers (17),
limiting transparency, reproducibility, and reliability of published findings (13,18). This is
especially important given that previous VSD studies may have under- or misreported risk for
adverse events including acute myocardial infarction, pulmonary embolism and death (13,18)
(see Discussion for more details). A recent re-analysis of VSD data (subset from Kaiser
Permanente) used more sensitive methods to detect myopericarditis cases following COVID
vaccination and reported its risk may be one to two orders of magnitude higher than previously
reported in the US (i.e. 195 cases per million second doses in males ages 12-39, or about 1 in
5K) (3). These findings are consistent with and supported by a recently published study that
reported a 25% increase in cardiac event emergency calls among ages <40 yrs following
vaccination campaigns using Israeli National Emergency Medical Services data (19).
Here, two independent, publicly available data sources from the US and Europe were
used to test whether region-to-region variation in vaccination rates predicts or correlates with
region-to-region variation in future (following weeks or month) mortality rates. We focused on
mortality risk as data for other severe adverse events such as myopericarditis, myocardial
infarction and thromboembolism etc. are not publicly available. Using the European data, we
asked whether COVID vaccination correlates with deaths at short and long intervals
post-injection stratified by 6 age groups (0-14, 15-44, 45-64, 65-74, 75-84, and 85+). With the
US data, multiple linear regression was used to test whether we could observe similar short
term effects seen in the European data. The US data was stratified by 8 age groups (0-17,
18-29, 30-39, 40-49, 50-64, 64-74, 75-84, and 85+). These models adjusted for COVID deaths
as well as seasonality effects and interregional variation in mortality due to other factors by
adjusting for same-month 2020 deaths. Using same month deaths from 2020 (as opposed to
2019 or earlier) also helped control for interregional differences in pandemic public health
measures before the vaccination campaigns began.
Our second aim was to estimate a US national average VMR and age-stratified rates
using significant regression slopes for the vaccination term in the regression model. The
European data reports age-stratified mortality rates on a weekly basis and allows for higher
temporal resolution analyses, but mortality rates are z-scored normalized and hence effect size
estimates in real units are not possible. The units of the US data allow for such estimates since
it records raw numbers of administered vaccine doses and death counts in each jurisdiction,
but at a lower (monthly) temporal resolution. Finally, we compared our estimates with
previously published US national average and age-stratified SARS‑CoV‑2 infection fatality
rates for risk-benefit ratio analysis of vaccination against COVID-19 stratified by age.
Associations between weekly vaccination and mortality data from Europe and Israel
For each week since the start of 2021 for 22 European countries, weekly increases in
percentages of the total population who received at least one injection were extracted from
Coronavirus (COVID) Vaccinations - Statistics and Research - Our World in Data, and
correlated with varying time lags (0-28 weeks post vaccination) with weekly age-stratified
mortality data extracted from euromomo.eu (see Supplementary Materials and Methods). The
overall description of results requires distinguishing between the age group 0-14 which were
unvaccinated during the time period analyzed, and ages above 14. For ages above 14, there is
a positive association (correlation) between vaccination and mortality rates during the first few
weeks of vaccination (Table 1, lags 0-5 and Figure 1 leftmost yellow peaks). Overall, mortality
above age 14 associates near zero or negatively with vaccination for mortality later than 5-6
weeks after vaccination (Table 1, lags 5-20 and Figure 1 middle blue troughs). These results
coincide with known clinical developments of vaccination, as found in the VAERS data: most
deaths occur within the first weeks after vaccination, and vaccine protection occurs after the
sixth week after first dose injection. For age groups 15-44 and 45-64, the overall tendency is
that protective vaccine effects (meaning negative associations between mortality and
vaccination) disappear about 20 weeks after first injection. After week 20, there might be a
tendency for adverse effects of vaccination, meaning positive r values between mortality
beyond week 20 and vaccination at least 20 weeks before (Table 1, lags >20 and Figure 1
rightmost yellow peaks).
For the unvaccinated age group 0-14, most associations between mortality and
vaccination in adults are positive (among 39 r values with unadjusted two-tailed P < 0.05, 32
are positive and 7 are negative r's). This tendency for positive correlations increases from the
week of vaccination until week 18 after vaccination, then disappears. It indicates indirect
adverse effects of adult vaccination on mortality of children of ages 0-14 during the first 18
weeks after vaccination.
The following analyses used publicly available US data on vaccination, mortality and
age-stratified population size in each US state. The data were obtained from either the CDC or
US Census Bureau (see Data Sources section in Supplementary Materials and Methods). Our
analyses focused on whether we could replicate the higher mortality within the first 5 weeks of
vaccination observed in the euromomo.eu data. Since US mortality data were limited to
month-level resolution, we tested whether monthly vaccination rates predicted mortality during
the same month or during the next month. Multiple linear regression was used to predict the
total # of deaths among 8 age groups (0-17, 18-29, 30-39, 40-49, 50-64, 65-74, 75-84, >85
years) for 7 months (February, March April, May, June, July and August 2021). For each month
and age group, the following equation was fitted:
Where and are the number of total deaths for that month in
year 2021 and 2020, respectively, and is the number of vaccine doses administered in
the previous month (or current month). An additional analysis using
log(Y21_deaths)-log(Y20_deaths) instead of log(Y21_deaths) as the dependent Y variable
was confirmed to yield the same results as the above models. See Supplementary Methods
and Materials for more information and details about other analyses to rule out potential
confounding factors such as COVID case rates and COVID deaths.
Prior month or current month vaccinations (# of administered doses) predicted monthly
total deaths in most age groups. The beta coefficient for the vaccine term was significant in 15
regression models (p<0.05 FDR corrected, see yellow boxes in Table 2). Most vaccination
regression slopes terms were positive while no terms with negative slopes survived p<0.05
corrected nor a more liberal threshold of p<0.05 uncorrected. In older age groups (>75 years)
the beta weights were highest in the beginning of the year, while in younger ages th later in the
year. This is the expected result since the vaccination campaign first targeted nursing homes
and older age groups before younger age groups became eligible for vaccination. When using
vaccination counts from the same month (instead of previous month) as deaths, 7 models
survived the applied significance threshold where the original models did not, all in younger
(<50 years old) age groups (Table 3, light grey boxes). Using age-specific vaccination rates
also increased detection of significant effects for 2 models (Table 3, dark gray boxes) where
effects were not detected in the previous 2 models. Adjusting for the number of new COVID
cases during the previous month did not significantly alter these results (see Supplementary
Table 5). Moreover, the results were similar when predicting non-COVID associated deaths
(Supplementary Table S6). Note that because COVID-associated deaths are rarer in younger
age groups, the latter analyses had much less power because few states had available data to
compute non-COVID deaths in ages 0-49 (see Supplementary Table S6). Scatter plots and
best fit lines of significant vaccination terms (p<0.05 FDR corrected, yellow boxes in Table 2
and cells with numbers in Table 3) for each month and age group are shown in Figure 3.
Higher resolution versions of the same are shown in Supplementary Figure S1.
Cumulating the monthly model-estimated deaths across all significant results from the
original models and from an additional 9 results from the two model variations mentioned
above yielded a total of 146,988 deaths attributed to COVID vaccinations between February
and August of 2021 (lower right cell of “Estimated Deaths” in Table 3). Applying the same
procedure while thresholding the results at a more liberal threshold (p<0.05 uncorrected)
yielded an estimated 168,908 vaccine-related deaths (Supplementary Table S6). The same
procedure applied using standard linear regression with a stringent threshold (p<0.05
corrected) yielded 133,382 deaths attributed to vaccination (Supplementary Table S7), while
thresholding these regression weights more liberally (p<0.05 uncorrected) yielded 187,402
vaccine associated deaths (Supplementary Table S8).
Results from the robustfit regression models thresholded at p<0.05 FDR corrected were
used to estimate VMR (see Figure 4 and Supplementary Materials and Methods). Dividing the
total number of model-estimated deaths by the total number of vaccine doses administered
between January and August yielded an estimated US national average VMR of 0.04%
(bottom of Table 2). Lower bound estimates of age-stratified VMRs were estimated by
averaging the aVMR estimates (see eq 5 in Supplementary Materials and Methods) across all
months and for all 3 models when thresholding regression slopes at p<0.05 uncorrected (see
methods). These yielded estimated aVMRs of 0.0045% for ages 0-17 years, 0.0065% for
18-29 years, 0.0091% for 30-39, 0.0165% for 40-49, 0.0157% for 50-64, 0.0445% for 65-74,
0.0604% for 75-84, and 0.0577% for 85-plus (see Table 3 for 95% CI). Note we consider these
to be lower bound estimates since the denominator (eq 5 in Supplementary Materials and
Methods) is vaccine doses administered in a given month across all ages (vaccine doses
stratified to the same age groups as the mortality data was not available, see limitations).
Using age-stratified (vs. total population) vaccine doses as the independent variable would
make the denominator smaller, but presumably leave the estimated regression slopes for
vaccination unaffected (or they might increase). Note that the results for age group 0-17 during
are presumed to reflect vaccinations in ages >12 years (20) as well as indirect effects in ages
<1 years (see Supplement for analyses restricted to ages <1 years) since no COVID vaccine
was authorized in ages 0-11 over the time periods examined.
In this study we find that regional variation in vaccination rates predicts mortality in
subsequent time periods. The mortality data from euromomo.eu confirm previously known
patterns in the vaccinated: a positive association with adverse events, including death, up to
5-6 weeks after the first injection, followed by a decrease in mortality associated with
vaccination 6-20 weeks post-injection. The decrease is presumably due to the protective effect
of vaccination, which is known to start 6 weeks after the first injection. The end of the
protective vaccine period as observed in our data, about 20 weeks, corresponds approximately
with the end of the protective vaccine period as generally accepted, 4-6 months (21). The
euromomo.eu data also reveal an unexpected increase in mortality in children (which are
unvaccinated) with adult vaccination rates in the previous period. It is notable that this indirect
adverse vaccination effect was independently observed in both CDC and euromomo.eu
datasets. The majority of deaths <18 years age occur in infants <1 years, and a significant
effect of vaccination on infant mortality was detected when the US CDC data was restricted to
that age group (see Supplementary Results). It is unclear to what extent the observed effects
relate to abnormally high mortalities around delivery, and/or infants, and/or in older children
and/or young adolescents. Note that several important concerns and errors have been raised
in response to previously published studies supporting safety of vaccination in pregnant
women (see Supplementary Discussion for a brief review).
The increased mortality in the first 0-6 weeks post-injection may be due in part to
increased COVID infectivity before vaccine protection takes effect. A re-analysis of a large real
world study of vaccine effectiveness (Dagan et al 2021 (22)) suggests infectivity in vaccinated
persons increases 3-fold approximately 7 days following the 1st dose of the Pfizer vaccine
(17). Figure 2 in (24) suggests a similar pattern with the CoronaVac vaccine. Likewise, the
euromomo.eu data also suggest a tendency for adverse effects caused by the vaccine in those
above age 14 beginning 20 weeks after first injection, potentially indicating that
antibody-dependent enhancement (ADE) (25–27) or another related effect kicks in after
protective vaccine effects dissipate. Alternatively, the increase in adverse effects observed
after week 20 may instead be due to short term mortality arising from booster campaigns
which began in late summer or fall. Further analyses are required to disentangle and
understand the causes of this effect.
The US CDC data allowed for estimation of VMR and vaccine-induced deaths.
Importantly, our calculations do not rely on VAERS and its associated limitations. Our
estimated US national average VMR of 0.04% is 20-fold greater than the CDC reported VMR
of 0.002%3, suggesting vaccine-associated deaths are underreported by at least a factor of 20
in VAERS. The estimate is based only on significant effects detected in our analysis, and
hence likely represents a lower bound on the actual underreporting factor.
Interestingly, our estimates of 133K to 187K vaccine-related deaths are very similar to
recent, independent estimates based off of US VAERS data through August 28th, 2021 by
Rose and Crawford (28). The authors report a range of estimates depending on different
credible assumptions about the VAERS underreporting factor and percentages of VAERS
deaths definitely caused by vaccination based on pathologists’ autopsy findings. The authors
compared a previously reported incidence rate of anaphylaxis in reaction to mRNA COVID
vaccine (~2.5 per 10,000 vaccinated) (29) to the number of events reported to VAERS to
estimate an underreporting factor for anaphylaxis (41x). This factor, multiplied by the number
of reported VAERS deaths and the percentage of VAERS deaths believed to be caused by
vaccination based on pathologists’ estimates, yields various estimates with an average around
180K deaths. Our estimate does not rely on VAERS data and uses independent and publicly
available data, and thus contributes additional convergent evidence for the above estimate of
The striking similarity of our estimates with those based on VAERS with data-driven
estimates of its underascertainment bias (28) suggests our results evidence a causal, not just
statistical association between vaccination and mortality. The combination of anecdotal
evidence (see Supplementary Materials and Methods) and concerns and limitations with the
vaccine safety trials (30) further lend credence to our interpretation. If such causal relation
should exist, it would manifest itself as a statistically significant increase in all-cause mortality
associated with time-lagged vaccination rates. We identify statistically significant associations
between vaccination and increased mortality post-vaccination that do not appear to be
explainable by other factors. See Supplementary Discussion for more reasons arguing why our
results evidence a causal link (not just an association) between vaccination and death.
Death and severe adverse events to the COVID vaccines appear to be mediated in part
by cytotoxicity of the spike protein and its (unintended) cleaving from transfected cells and
biodistribution in organs outside the injection site (7,9,31–34). Vaccination may also contribute
to higher COVID IFR before vaccination protection kicks in (and after full protection wears off)
due to antibody dependent enhancement (ADE) (25,27,35). The effect may be related to
enhanced respiratory disease observed in preclinical studies of SARS and MERS vaccines
(36,37). An additional or alternative mechanism may stem from quality control issues related to
production, handling and distribution of the vaccines. A recent analysis of VAERS data
suggests only ~5% of the vaccine batches account for the majority (>90%) of adverse
reactions, those batches were the most widely distributed (more than 13 states), and reported
adverse event rates appear to vary across jurisdictions an order of magnitude (38). The
website https://www.howbadismybatch.com allows users to identify specific batch numbers of
Pfizer, Moderna and Janssen vaccines that are associated with the most adverse reactions.
Existing safety and surveillance studies are not designed to reliably estimate COVID
vaccine-induced death risk
A recent safety surveillance analysis of mRNA vaccines against COVID using the
Vaccine Safety Datalink (15) found event rates for 23 serious health outcomes were not
significantly higher for individuals 1 to 21 days after vaccination compared with similar
individuals at 22 to 42 days after vaccination. This is not very informative as the main
comparison of interest is the background rate of adverse events in the unvaccinated. If the
severe adverse event rate is similar 1-21 days post-vaccination as it is 22-42 days
post-vaccination, then no difference in risk (safety signal) will be detected. The authors include
an analysis using an unvaccinated comparator group in Supplementary eTable 6. Surprisingly,
the table reports significantly reduced risk of thrombosis with thrombocytopenia syndrome
(p=0.004), hemorrhagic stroke (p<0.001), pulmonary embolism (p<0.001), and acute
myocardial infarction (p<0.001) in the vaccinated 1-21 days post injection compared to the
unvaccinated comparator group. This is intriguing because these adverse events are precisely
the events known to be associated with both the viral vector-based and mRNA COVID
vaccines based on CDC VAERS data (749 results for “acute myocardial infarction”, 4,579
results for “thrombosis” or “thrombocytopenia”, 98 results for “hemorrhagic stroke”, and 2,395
results for “pulmonary embolism” for mRNA vaccines as of Oct 22nd, 2021) and published
case reports (7,10,39,40). The authors do not devote any discussion on how or why their results
provide strong evidence that COVID vaccination appears to protect against the very adverse
events that were previously associated with vaccination. We speculate it is more likely the
groups were mislabeled due to human or technical error.
A recent paper by Xu et al., also based on the Vaccine Safety Datalink (VSD) cohorts
used in Klein et al., reported significantly reduced mortality risk in vaccinated vs. unvaccinated
(18). As with Klein et al. that found significantly reduced risk for severe adverse events in
vaccinated people (discussed above), the finding of reduced standardized mortality rates
(p<0.001) in the vaccinated compared with unvaccined is unexpected, especially since the
groups were matched for “similar characteristics” and standardized mortality rates were
adjusted for age, sex, race and ethnicity. The authors suggest “This finding might be because
of differences in risk factors, such as underlying health status and risk behaviors among
recipients of mRNA and Janssen vaccines that might also be associated with mortality risk”
(18). However, this does not comport with recent findings from a large survey study that found
PhD-holders are among the most vaccine hesitant groups (41,42), as are women looking to
become pregnant, religious people, and people who practice yoga/“wellness” culture (43).
Given that the study is based on the same sites/cohorts used Klein et al. (13), which found
significantly reduced risk in the vaccinated for the same severe adverse events that have
associated with COVID vaccination in VAERS data and published case reports (see discussion
above), we speculate their findings may be due to a technical or human error involving group
labeling or coding. Note that the data used for their study is not publicly accessible (in contrast
to our study), and two authors report receiving funding from Pfizer.
Vaccine cost-benefit ratio
According to a recent meta-analysis of IFR studies, up to 90% of the variation in
population-wide coronavirus infection fatality rate (IFR) is explained by age composition and
the extent to which older age groups are exposed to the virus (44). The study reports the IFR
for age 10 is 0.002%, age 18 years is 0.005%, 25 years is about 0.01%, 45 years 0.1%, 55
years 0.4%, 65 years 1.4%, 75 years 5%, and 15% >85 years (44). Calculations based on 61
studies (74 estimates) and eight preliminary national estimates by Ioannides suggest a median
of 0.05% and upper bound IFR of 0.3% for ages <70 (45). This latter estimate is similar to an
estimated US national average IFR of 0.35% based on a Bayesian evidence synthesis model
that averaged age-specific IFRs weighted by the fraction of the population in each age group
across US states (46). A comparison of previously published age-stratified IFR (44) with our
age-stratified VMRs shows they are similar orders of magnitude below age 45, strongly
suggesting the benefits of vaccination do not outweigh the risks in anyone ages 45 and
younger (Figure 5).
An individual’s overall risk of dying from COVID is also a function of infection risk, which
varies based on lifestyle, location, time, occupation, and behavior (i.e. social distancing,
effective masking with N95 etc.), as well as the presence of comorbidities. In the vaccine
clinical trials (when social distancing and masking measures were in place), ~1-2% of the
participants contracted symptomatic COVID in the placebo group over a period of a few
months (37). Infection risk calculators allow someone to estimate their risk of infection based
on attending an event of a certain size (47). For example, a 55 year old attending events over a
given time period with a 10% infection risk has a 0.1*0.4%=0.04% chance of dying from
COVID, which is similar to the odds of vaccine-induced death (VMR~0.01%).
In individuals with no previous exposure and natural immunity, the benefits of
vaccination may outweigh the risks in age groups >75 years, where the IFR (>1%) for earlier
variants is one or two orders of magnitude greater than the estimated VMR of 0.06% in this
age group (44). The benefits may also outweigh the risks in ages >45 with high COVID risk
(several or more comorbidities and no previous coronavirus exposure) where the IFR of 0.1%
is an order of magnitude higher than the estimated VMR of 0.01% (34). However, given that
more recent variants (Omicron) may be up to 90% less lethal than previous variants (Delta)
(48), and that there is a lack of sufficient safety data on boosters (49), we advise that boosters
be contraindicated in all age groups until and unless their safety has been well established.
Some may argue against publication of our data on the grounds they may cause panic
in vaccinated individuals. However, such panic would be greatly mitigated by the fact that the
vast majority of lethal and severe adverse events occur the first 6 weeks following vaccination.
The vaccine companies are already suggesting a 4th booster may be recommended or
required by the fall of 20224. We hope publication of our results will lead to a paradigm shift
that could prevent hundreds of thousands or more unnecessary vaccine deaths due to the
continued booster campaigns. It is morally unacceptable and unethical to suppress research
for the sake of protecting people’s feelings and mental health, if the research has a benefit that
far outweighs these costs by i.e. informing the medical research community of the true risks of
the vaccines so that future boosters can be contraindicated in low COVID risk populations.
Implications for public health policy
There is little to no evidence that vaccines reduce community spread and transmission.
The vaccine clinical trials used symptomatic, not asymptomatic COVID, as a clinical endpoint.
Since they did not require weekly coronavirus testing in their participants, they were not
designed to estimate vaccine efficacy in reducing infection and hence transmission of the virus
in pre- and/or asymptomatic persons. Indeed a recent July CDC study in Barnstable, MA
reported a majority (75%) of COVID infections were among fully vaccinated people in an area
with 69% vaccination coverage, with similar viral loads between vaccinated and unvaccinated
(50). The US CDC has officially recognized that the vaccines do not prevent transmission or
spread of the virus5.Given that vaccines do not reduce community spread and that the risks to
the individual outweigh the benefits for most age groups,vaccine mandates in workplaces,
colleges, schools and elsewhere are ill-advised. We do not see much benefit in vaccine
mandates other than increasing serviceable obtainable market (SOM) share for the vaccine
companies. See (30) and (34) for a more in-depth discussion and literature review on why the
mandates are not based on sound science given the relatively low COVID risk in healthy
middle-aged and young adults and growing evidence base for alternative prevention and early
treatment options for COVID. See Supplemental Discussion for more resources where readers
can learn about the nature and volume of life-altering COVID vaccine injuries.
Limitations and future directions
Future studies that include autopsies on VAERS-reported deaths are required to
identify mechanisms of vaccine-induced death. Ideally, our analyses would use age-stratified
vaccination to predict age-stratified mortality within the same age groups. However, the
European and Israel vaccination data are not age-stratified, and the US vaccination data only
provides some age-specific data starting in later months (i.e. vaccines administered to ages
>65, >18, and >12 years). In addition, while the US vaccination and COVID cases are updated
daily, the age-stratified death counts are per-month, thus preventing analyses using shorter
time windows. The additional information may have increased our sensitivity to detect
significant effects in more age groups and time periods. Such a scenario would increase our
mortality estimates, in which case the death estimates presented here based only on
significant effects (p<0.05 corrected) can be considered a lower bound on the estimated
deaths attributed to COVID vaccination. The current study focused on vaccine-attributed
deaths within 5-6 weeks of vaccination to estimate age-stratified VMR. Future work should
examine later periods to estimate lives saved from vaccination and also potential vaccine
associated mortality after protective effects wane.
In the European and Israeli data, we find that COVID vaccination correlates positively
with mortality 0-5 weeks from vaccination, before associating with lower mortality 6-20 weeks
from vaccination. The US data allowed us to estimate a US national average VMR of 0.04%
and age-stratified vaccine-induced mortality rates within 1 month post-vaccination. Significant
regression terms estimate 130K-180K US deaths can be attributed to vaccination between
February and August of 2021. The estimate converges with independent estimates based on
the Vaccine Adverse Events Reporting System (VAERS) and suggests VAERS deaths are
underreported by a factor of 20. Comparison of our age-stratified VMR and with age-stratified
IFR rates suggests the risks of COVID vaccination outweighs the benefits in children, young
and middle age adults, and in older age groups with low COVID risk, previous coronavirus
exposure, and access to alternative prophylaxis and early treatment options. Our findings
raise important questions about mass COVID vaccinations strategies that warrant further
investigation and review.
Data and Resource Sharing
All data used in this study is publicly available. See Data Sources subsection in the
Methods for links to the raw data. The extracted data (minimally preprocessed spreadsheets
and intermediate results) for both European and US datasets is available in the provided
Github repository which is publicly available. The repository also contains all MATLAB code
used for the US dataset analyses. Readers who would like to inspect and replicate the results
or reanalyze the data may find it easier to first double check the intermediate table files (in
Table subfolder of the Github repo at https://github.com/spiropan/CoVMR) against the original
CDC data and then work off of these tables with their software of choice. In addition, readers
are referred to the comments section on the preprint of this article which has functioned as an
open pre-publication peer review with responses from the authors (see
SPP analyzed US data and drafted the manuscript; HS analyzed European and Israeli data
and drafted relevant text.
Conflict of interest
HS has no relevant conflicts of interest to report. SPP holds a short position on Moderna stock.
We would like to thank Eileen Natuzzi, Carlos Oliver, and Pericles Philippopoulos for critical
comments and feedback on the manuscript.
Table 1. Correlations between COVID vaccination rates and mortality as a function of lag (# weeks post-injection)
and age group. Each cell summarizes the pearson correlation coefficients between weekly increase in percent
vaccinated and weekly mortality in 23 European countries. Top header row: lag=weeks between mortality and injection,
n=number of correlations summarized. Middle matrix (%) shows the percentage of positive correlations for that lag among
n correlation. *=P < 0.05 corrected, sign test. Bottom matrix (P<0.05) shows the number of negative and positive
correlation r’s with P < 0.05 uncorrected. Blue: overall protective effect (more injections->lower mortality); yellow: overall
adverse effect (more injections->higher mortality).
Table 2. Regression weights and p-values for the vaccination term predicting same or
next month deaths using US CDC data. For each month in 2021 and age group, beta
weights and uncorrected p-values are listed for the vaccination term ( ) in the fitted equation:
where Vax = vaccine doses administered previous or same month across all US states with
available data for that month and age group (~42-52 states for each cell/regression, see
Equation 1). Models were fitted using robust regression. Yellow indicates positive beta slopes
with p-values < 0.05 FDR corrected. No negative slopes were significant.
Table 3. Model-estimated deaths attributed to COVID vaccination for each age group and
month using US CDC data. Significant beta weight coefficients ( ) in Table 2 surviving
p<0.05 FDR corrected were used to estimate VMR and total deaths for each age group and
month. If a model using same (not previous) month vaccinations was significant and the
equivalent models using previous month was not, then death estimates from those models
were used instead (light gray boxes). Similarly, if a model using age-specific vaccination (i.e.
doses administered to people >65 yrs) was significant and the equivalent model using all
vaccine doses administered was not, then death estimates from those models were used
instead (dark gray boxes). See methods for VMR and aVMR definitions and calculations.
ns=not significant at p<0.05 FDR corrected. NA=Not available.
0.0045 [0.0028, 0.0080]
0.0065 [0.0040, 0.0095]
0.0091 [0.0040, 0.0136]
0.0165 [0.0090, 0.0231]
0.0157* [0.0018, 0.0370]
0.0445* [0.0167, 0.0743]
0.0604 [0.0108, 0.0859]
0.0577 [0.0298, 0.0802]
# Vaccine doses
0.0408 [0.0244, 0.0474]
Light gray indicates models estimated using same, not previous, month vaccinations
Dark gray indicates models estimated using vaccines administered > ages 65
Light blue indicates significant results when predicting deaths in ages <1 years. Model estimated
667 infant deaths (see Supplementary Results).
*Robust regression did not yield significant results in these age groups. Thus these estimates were
derived from results of standard least-squares regression.
1. Commissioner O of the. Coronavirus (COVID-19) Update: June 25, 2021 [Internet]. FDA. FDA;
2021 [cited 2022 Feb 8]. Available from:
2. TEMPORARY REMOVAL: A Report on Myocarditis Adverse Events in the U.S. Vaccine Adverse
Events Reporting System (VAERS) in Association with COVID-19 Injectable Biological Products -
ScienceDirect [Internet]. [cited 2022 Feb 7]. Available from:
3. Sharff KA, Dancoes DM, Longueil JL, Johnson ES, Lewis PF. Risk of Myopericarditis following
COVID-19 mRNA vaccination in a Large Integrated Health System: A Comparison of Completeness
and Timeliness of Two Methods [Internet]. medRxiv; 2021 [cited 2022 Feb 8]. p.
2021.12.21.21268209. Available from:
4. Oster ME, Shay DK, Su JR, Gee J, Creech CB, Broder KR, et al. Myocarditis Cases Reported After
mRNA-Based COVID-19 Vaccination in the US From December 2020 to August 2021. JAMA
[Internet]. 2022 Jan 25 [cited 2022 May 18];327(4):331–40. Available from:
5. Patone M, Mei XW, Handunnetthi L, Dixon S, Zaccardi F, Shankar-Hari M, et al. Risks of
myocarditis, pericarditis, and cardiac arrhythmias associated with COVID-19 vaccination or
SARS-CoV-2 infection. Nat Med [Internet]. 2022 Feb [cited 2022 Jun 3];28(2):410–22. Available
6. Patone M, Mei XW, Handunnetthi L, Dixon S, Zaccardi F, Shankar-Hari M, et al. Risk of myocarditis
following sequential COVID-19 vaccinations by age and sex [Internet]. medRxiv; 2021 [cited 2022
Jun 26]. p. 2021.12.23.21268276. Available from:
7. Hunter PR. Thrombosis after covid-19 vaccination. BMJ. 2021 Apr 14;373:n958.
8. Lai CC, Ko WC, Chen CJ, Chen PY, Huang YC, Lee PI, et al. COVID-19 vaccines and thrombosis
with thrombocytopenia syndrome. Expert Rev Vaccines. 2021 Aug;20(8):1027–35.
9. Merchant H. Might post-injection distribution of CoViD vaccines to the brain explain the rare fatal
events of cerebral venous sinus thrombosis (CVST)? BMJ (Clinical research ed) [Internet]. 2021
Apr 15 [cited 2021 Aug 24];373:n958. Available from:
10. Hippisley-Cox J, Patone M, Mei XW, Saatci D, Dixon S, Khunti K, et al. Risk of thrombocytopenia
and thromboembolism after covid-19 vaccination and SARS-CoV-2 positive testing: self-controlled
case series study. BMJ. 2021 Aug 26;374:n1931.
11. Andraska EA, Kulkarni R, Chaudhary M, Sachdev U. Three cases of acute venous
thromboembolism in females following vaccination for COVID-19. J Vasc Surg Venous Lymphat
Disord. 2021 Aug 2;S2213-333X(21)00392-9.
12. Al-Maqbali JS, Al Rasbi S, Kashoub MS, Al Hinaai AM, Farhan H, Al Rawahi B, et al. A 59-Year-Old
Woman with Extensive Deep Vein Thrombosis and Pulmonary Thromboembolism 7 Days Following
a First Dose of the Pfizer-BioNTech BNT162b2 mRNA COVID-19 Vaccine. Am J Case Rep. 2021
13. Klein NP, Lewis N, Goddard K, Fireman B, Zerbo O, Hanson KE, et al. Surveillance for Adverse
Events After COVID-19 mRNA Vaccination. JAMA [Internet]. 2021 Sep 3 [cited 2021 Oct 8];
Available from: https://doi.org/10.1001/jama.2021.15072
14. VAERS - Data [Internet]. [cited 2021 Aug 16]. Available from: https://vaers.hhs.gov/data.html
15. Baggs J, Gee J, Lewis E, Fowler G, Benson P, Lieu T, et al. The Vaccine Safety Datalink: a model
for monitoring immunization safety. Pediatrics. 2011 May;127 Suppl 1:S45-53.
16. Lazarus R, Klompas M. Electronic Support for Public Health–Vaccine Adverse Event Reporting
System (ESP:VAERS) [Internet]. Available from:
17. VSD Data Sharing Program Guidelines | Vaccine Safety | CDC [Internet]. 2020 [cited 2022 Feb 8].
18. Xu S. COVID-19 Vaccination and Non–COVID-19 Mortality Risk — Seven Integrated Health Care
Organizations, United States, December 14, 2020–July 31, 2021. MMWR Morb Mortal Wkly Rep
[Internet]. 2021 [cited 2021 Oct 22];70. Available from:
19. Sun CLF, Jaffe E, Levi R. Increased emergency cardiovascular events among under-40 population
in Israel during vaccine rollout and third COVID-19 wave. Sci Rep. 2022 Apr 28;12(1):6978.
20. Murthy BP. COVID-19 Vaccination Coverage Among Adolescents Aged 12–17 Years — United
States, December 14, 2020–July 31, 2021. MMWR Morb Mortal Wkly Rep [Internet]. 2021 [cited
2022 Feb 8];70. Available from: https://www.cdc.gov/mmwr/volumes/70/wr/mm7035e1.htm
21. Levin EG, Lustig Y, Cohen C, Fluss R, Indenbaum V, Amit S, et al. Waning Immune Humoral
Response to BNT162b2 Covid-19 Vaccine over 6 Months. New England Journal of Medicine
[Internet]. 2021 Oct 6 [cited 2021 Nov 5];0(0):null. Available from:
22. Dagan N, Barda N, Kepten E, Miron O, Perchik S, Katz MA, et al. BNT162b2 mRNA Covid-19
Vaccine in a Nationwide Mass Vaccination Setting. N Engl J Med. 2021 Apr 15;384(15):1412–23.
23. Seligmann H. Expert evaluation on adverse effects of the Pfizer-COVID-19 vaccination [Internet].
2021. Available from:
24. Jara A, Undurraga EA, González C, Paredes F, Fontecilla T, Jara G, et al. Effectiveness of an
Inactivated SARS-CoV-2 Vaccine in Chile. New England Journal of Medicine [Internet]. 2021 Sep 2
[cited 2021 Sep 24];385(10):875–84. Available from: https://doi.org/10.1056/NEJMoa2107715
25. Arvin AM, Fink K, Schmid MA, Cathcart A, Spreafico R, Havenar-Daughton C, et al. A perspective
on potential antibody-dependent enhancement of SARS-CoV-2. Nature [Internet]. 2020 Aug [cited
2021 Aug 30];584(7821):353–63. Available from:
26. Wan Y, Shang J, Sun S, Tai W, Chen J, Geng Q, et al. Molecular Mechanism for
Antibody-Dependent Enhancement of Coronavirus Entry. J Virol. 2020 Feb 14;94(5):e02015-19.
27. Wen J, Cheng Y, Ling R, Dai Y, Huang B, Huang W, et al. Antibody-dependent enhancement of
coronavirus. Int J Infect Dis. 2020 Nov;100:483–9.
28. Rose J, Crawford M. Estimating the number of COVID vaccine deaths in America [Internet].
Available from: https://downloads.regulations.gov/CDC-2021-0089-0024/attachment_1.pdf
29. Blumenthal KG, Robinson LB, Camargo CA, Shenoy ES, Banerji A, Landman AB, et al. Acute
Allergic Reactions to mRNA COVID-19 Vaccines. JAMA. 2021 Apr 20;325(15):1562–5.
30. Pantazatos S. Vaccine mandates are not based on sound science: they are harmful and should be
lifted as soon as possible. 2021 Aug 23 [cited 2021 Sep 9]; Available from: https://researchers.one/
31. European Medicines Agency. Assessment report: COVID-19 mRNA vaccine (nucleoside-modified)
[Internet]. Available from:
32. Pzifer Confidential. SARS-COV-2 mRNA Vaccine (BNT162, PF-07302048) 2.6.4 Overview of
Pharmacokinetic Test [Internet]. Available from: https://files.catbox.moe/0vwcmj.pdf
33. Ogata AF, Cheng CA, Desjardins M, Senussi Y, Sherman AC, Powell M, et al. Circulating Severe
Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) Vaccine Antigen Detected in the
Plasma of mRNA-1273 Vaccine Recipients. Clinical Infectious Diseases [Internet]. 2021 May 20
[cited 2021 Sep 3];(ciab465). Available from: https://doi.org/10.1093/cid/ciab465
34. Kostoff RN, Calina D, Kanduc D, Briggs MB, Vlachoyiannopoulos P, Svistunov AA, et al. Why are
we vaccinating children against COVID-19? Toxicol Rep. 2021;8:1665–84.
35. Seneff S, Nigh G. Worse Than the Disease? Reviewing Some Possible Unintended Consequences
of the mRNA Vaccines Against COVID-19. International Journal of Vaccine Theory, Practice, and
Research [Internet]. 2021 Jun 16 [cited 2021 Aug 16];2(1):38–79. Available from:
36. Agrawal AS, Tao X, Algaissi A, Garron T, Narayanan K, Peng BH, et al. Immunization with
inactivated Middle East Respiratory Syndrome coronavirus vaccine leads to lung immunopathology
on challenge with live virus. Hum Vaccin Immunother [Internet]. 2016 Jun 7 [cited 2021 Aug
22];12(9):2351–6. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5027702/
37. Baden LR, El Sahly HM, Essink B, Kotloff K, Frey S, Novak R, et al. Efficacy and Safety of the
mRNA-1273 SARS-CoV-2 Vaccine. N Engl J Med. 2021 Feb 4;384(5):403–16.
38. EXCLUSIVE – 100% of Covid-19 Vaccine Deaths were caused by just 5% of the batches produced
according to official Government data – The Expose [Internet]. [cited 2021 Nov 23]. Available from:
39. Flower L, Bares Z, Santiapillai G, Harris S. Acute ST-segment elevation myocardial infarction
secondary to vaccine-induced immune thrombosis with thrombocytopaenia (VITT). BMJ Case Rep.
2021 Sep 27;14(9):e245218.
40. Sung JG, Sobieszczyk PS, Bhatt DL. Acute Myocardial Infarction Within 24 Hours After COVID-19
Vaccination. Am J Cardiol. 2021 Oct 1;156:129–31.
41. Americans with PhDs are most reluctant to get vaccinated against COVID [Internet]. MSN. [cited
2021 Oct 22]. Available from:
42. King WC, Rubinstein M, Reinhart A, Mejia RJ. Time trends and factors related to COVID-19
vaccine hesitancy from January-May 2021 among US adults: Findings from a large-scale national
survey [Internet]. 2021 Jul [cited 2021 Oct 22] p. 2021.07.20.21260795. Available from:
43. Osborne-Crowley L. ‘There is a lot of distrust’: why women in their 30s are hesitant about the Covid
vaccine. The Guardian [Internet]. 2021 Jan 30 [cited 2021 Oct 22]; Available from:
44. Levin AT, Hanage WP, Owusu-Boaitey N, Cochran KB, Walsh SP, Meyerowitz-Katz G. Assessing
the age specificity of infection fatality rates for COVID-19: systematic review, meta-analysis, and
public policy implications. Eur J Epidemiol. 2020 Dec;35(12):1123–38.
45. Ioannidis JPA. Infection fatality rate of COVID-19 inferred from seroprevalence data. Bull World
Health Organ. 2021 Jan 1;99(1):19-33F.
46. Chitwood MH, Russi M, Gunasekera K, Havumaki J, Klaassen F, Pitzer VE, et al. Reconstructing
the course of the COVID-19 epidemic over 2020 for US states and counties: results of a Bayesian
evidence synthesis model [Internet]. 2021 Jul [cited 2021 Aug 16] p. 2020.06.17.20133983.
Available from: https://www.medrxiv.org/content/10.1101/2020.06.17.20133983v3
47. Chande A. COVID-19 Event Risk Assessment Planning Tool [Internet]. [cited 2021 Aug 16].
Available from: https://covid19risk.biosci.gatech.edu/
48. Lewnard JA, Hong VX, Patel MM, Kahn R, Lipsitch M, Tartof SY. Clinical outcomes among patients
infected with Omicron (B.1.1.529) SARS-CoV-2 variant in southern California [Internet]. medRxiv;
2022 [cited 2022 Mar 1]. p. 2022.01.11.22269045. Available from:
49. Arbel R, Hammerman A, Sergienko R, Friger M, Peretz A, Netzer D, et al. BNT162b2 Vaccine
Booster and Mortality Due to Covid-19. N Engl J Med. 2021 Dec 23;385(26):2413–20.
50. Brown CM, Vostok J, Johnson H, Burns M, Gharpure R, Sami S, et al. Outbreak of SARS-CoV-2
Infections, Including COVID-19 Vaccine Breakthrough Infections, Associated with Large Public
Gatherings - Barnstable County, Massachusetts, July 2021. MMWR Morb Mortal Wkly Rep. 2021
51. Pantazatos S, Seligmann H. COVID vaccination and age-stratified all-cause mortality risk [Internet].
2021 Oct. Available from: https://doi.org/10.13140/RG.2.2.28257.43366
52. Pantazatos S, Seligmann H. Supplementary Tables and Figures - Google Drive for “COVID
vaccination and age-stratified all-cause mortality risk” [Internet]. [cited 2022 Feb 20]. Available from:
Figure 1. Graphical representation of Table 1 European data results. Adverse effects in
yellow, above horizontal line, protective effects in blue, below horizontal line. Results of
correlation analyses for all age classes and all combinations of weeks, with mortality occurring
the same week or after the injection week are plotted. In a) the percent positive correlations
between vaccination rates and mortality is plotted against time since 1st injection for 6 age
groups (A - 0-14 years, B 15-14 years, C 45-64 years, D 65-74 years, E 75-84 years, and F
85+ years). Percentages >50% are shaded yellow, <50% shaded blue. Asterisk indicates
p<0.05 corrected for the sign test (see methods). Pearson correlation coefficients r from these
analyses are in Supplementary Table 3. In b) % positive correlations (left column) and
numbers of negative and positive r with p<0.05 uncorrected (middle and right columns).
Figure 2. Example correlation plots from the European dataset. Top: Z-score of weekly
mortality for ages 15-44 in 23 countries on week 21 of 2021 as a function of increase in
percent vaccinated in these countries, during week 13 of 2021. For this correlation, the time
lag in weeks between injection and mortality is 21-13=8 weeks. The association suggests
beneficial injection effects at two months weeks after injection.
Bottom: Z-score of weekly mortality for ages 15-44 in 23 countries on week 14 of 2021 as a
function of increase in percent vaccinated in these countries, during week 12 of 2021. For this
correlation, the time lag in weeks between injection and mortality is 14-12=3 weeks. The
association suggests adverse injection effects during the first weeks after injection.
Figure 3. Scatter plots of monthly vaccination doses vs. subsequent month deaths with
best fit regression lines from the US CDC dataset. The graph plots log(administered
vaccine doses) vs. log(residual 2021 deaths) after adjusting for log(2020 deaths) for each
month (top) and age group (right), for each regression model in which the term survived
p<0.05 FDR corrected (see Table 2 and methods) ns=not significant. For a higher resolution
image see Supplementary Figure S1, and for the highest resolution plots viewable in a web
browser see (52).
Figure 4. Method to estimate COVID vaccination mortality risk using publicly available
US CDC data. The cartoon plot shows a schematic of the method to estimate COVID vaccine
mortality risk using regional variation in vaccine doses administered and all-cause mortality.
Vaccine-induced mortality risk is expressed as the ratio of model-predicted deaths over
vaccine doses (i.e. “rise” over “run”). Predicted deaths are estimated as the difference between
and for a given increase (i.e. 10%) in vaccine doses at . The approach is completely
data-driven and does not rely on assumptions about reporting bias as with other methods.
Figure 5. Simple cost-benefit ratio analysis of COVID vaccination stratified by age. The
lower bound estimates of age-stratified vaccine mortality rates from the current study (aVMR,
right) have similar orders of magnitude as previously published coronavirus infection mortality
rates (IFR). * Left panel is adapted from a meta-analysis of 27 studies to estimate
age-stratified coronavirus infection fatality rates (IFR) (44).
Supplementary Material for “COVID-19 vaccination and age-stratified all-cause mortality risk”
Spiro P. Pantazatos1and Hervé Seligmann2
1Molecular Imaging and Neuropathology Area, New York State Psychiatric Institute and Department
of Psychiatry, Columbia University Irving Medical Center, New York, NY; 2Independent Research
Scientist, Jerusalem, Israel
To whom correspondence should be addressed:
This manuscript contains:
Supplementary Methods and Materials
Supplementary Discussion and References
8 Supplementary Tables S1-S8
1 Supplementary Figure S1
Running title: COVID-19 vaccination and age-stratified mortality
Keywords: Public health, medical ethics, risk-benefit ratio, COVID-19, SARS‑CoV‑2, vaccine adverse
The 0-17 age group is peculiar in that it includes infants <1 years old. Infant deaths comprise
the majority of deaths in this age group (1). Since infants are not vaccinated, we hypothesized this
effect could be attributed to vaccinations in the mother given a July, 2021 report that found 2,346
VAERS-reported cases were pregnant mothers at time of vaccination, 36% of whom experienced
some type of pregnancy disorder (2). To further test this possibility, an additional regression in the <1
years of age group was run, and results were significant for the August model (p<0.05 corrected). The
model estimated 667 infant deaths in the US during the month of August, 2021 may be attributed to
vaccinations in July, 2021, while 1,227 deaths were estimated overall in the 0-17 age group (see light
blue box, Table 2 of main text).
Methods and Materials
European dataset sources
Weekly age-stratified mortality data were extracted from euromomo.eu for each week since the
start of 2021 for 22 countries covered by euromomo (21 european countries (Austria, Belgium,
Cyprus, Danemark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Luxemburg,
Malta, Netherlands, Norway, Portugal, Slovenia, Spain, Sweden, Switzerland, UK (England)) and
Israel). These consist of weekly adjusted z-scores, for 6 age groups: 0-14, 15-44, 45-64, 65-74, 75-84
and 80+ years of age. Weekly increases in percentages of the total population who received at least
one injection are extracted from Coronavirus (COVID) Vaccinations - Statistics and Research - Our
World in Data (https://ourworldindata.org/covid-vaccinations). Supplementary Tables S1 and S2
contained the raw input mortality and vaccination data used in the analyses while Supplementary
Table S3 contains the intermediate results (r correlations at each lag before statistical inference, see
US Dataset sources
All US data used in the analyses are publicly available and were obtained from either the CDC
or US Census Bureau. Vaccination rates across time and US states were extracted from the “COVID
Vaccinations in the United States,Jurisdiction” spreadsheet (3). Total deaths per month by age group
and sex for each US state were extracted from “Provisional COVID Deaths by Sex and Age”
spreadsheet (4). Number of COVID cases per month in each US state were extracted from “United
States COVID Cases and Deaths by State over Time” (5). Age-stratified total populations per US
state in 2019 were obtained from “NC-EST2019-AGESEX” spreadsheet (6). Spreadsheets were
accessed during the first week of September 2021 and included data up through September 1st,
2021. To facilitate importing of data into MATLAB, spreadsheets were filtered (unused columns and
rows were removed) and sorted by US State abbreviation. The proprocessed table for April is
included as Supplementary Table S4. Similar tables for the other months are available on the Github
repository in the subfolder InputFiles (see https://github.com/spiropan/CoVFR). The MATLAB script
was used to extract relevant data and reorganize the data into tables for each month that list Y20 and
Y21 deaths for each age group (columns) and for each state (rows) (see Github subfolder Tables).
Analyses of euromomo.eu data
These mortality data were analysed using Pearson’s correlation coefficient (Pearson’s r)
between weekly increases in vaccination rates and mortality rates for the same week or subsequent
weeks, from week 1 to week 33 of 2021, separately for each of the six age groups. This results in
correlation matrices between vaccination rates and age-specific mortality, with lags between
vaccination and mortality week ranging from lag 0 to lag 32 (see Supplementary Table 3). The r
correlations were then combined into 9 sets of 3 or 4 week groupings as follows: lags [0,1,2], [3,4,5],
[6,7,8], [9,10,11], [12,13,14], [15,16,17], [18,19.20], [21,22,23,24], and [25,26,27,28] for each age
group. The last two sets contained 4 weeks to make up for the lower number of r correlations
available at much longer lags. Each set of r correlations was then subjected to a sign test for a total of
(9 time periods x 6 age-groups= 54) total sign tests and resulting p-values. These p-values were then
corrected for multiple comparisons correction using Benjamini-Hochberg False Discovery Rate (FDR)
Analyses of US CDC data
Our analyses focused on whether we could replicate the higher mortality within the first 5
weeks of vaccination observed in the euromomo.eu data. Since US mortality data were limited to
month-level resolution, we tested whether monthly vaccination rates predicted mortality during the
same month or during the next month. All the raw data used in the following procedures are available
as csv tables to facilitate reanalysis (see Resource and Data Sharing section). Multiple linear
regression was used to predict the total # of deaths among 8 age groups (0-17, 18-29, 30-39, 40-49,
50-64, 65-74, 75-84, >85 years) for 7 months (February, March April, May, June, July and August
2021). For each month and age group, the following equation was fitted:
Where and are the number of total deaths for that month in year 2021 and
2020, respectively, and is the number of vaccine doses administered in the previous month (or
Our model is an ANCOVA model that adjusts for baseline outcomes (here Y20 monthly
deaths) while predicting “post” intervention (i.e. vaccination) outcomes (here Y21 deaths). Such
models have higher power compared to analyses applied to change scores or percent change looking
at baseline (7–9).The models were fitted using robust regression using robustfit function in MATLAB
2014b. The algorithm uses iteratively reweighted least squares with the bisquare weighting function
and helps ensure results are not driven by outliers by deweighting them during fitting. Standard
least-squares linear regression using glmfit was also applied for comparison and for cases where
results from robust regression could not be determined. The beta weights and associated p-values for
the term of interest ( ) are reported for each regression in each set of models. Positive (negative)
beta weights mean higher vaccinations predict higher (lower) mortality. Each set of models consisted
of 8x7=56 regressions and resulting p-values, which were corrected for multiple comparisons using
Benjamini-Hochberg False Discovery Rate (FDR) correction (10).
Since most (VAERS reported) vaccine-related deaths events occur within 1-2 weeks of
vaccination, an additional set of models were estimated in which vaccinations were used to predict
total deaths in the same month. Significant effects from these models replaced results when
significant results using previous month vaccinations were not detected and were used instead when
estimating deaths for that particular age group and month (Tables 2 and 3). To potentially improve
sensitivity in detecting vaccination effects in older age groups, an additional model in which the
number of vaccines administered to ages >65 years (instead of all ages) predicted deaths among
ages >65 years was estimated. The significant results of these models replaced results from
equivalent models with non-significant results and when estimating deaths attributed to vaccination.
Since the number of vaccines administered by age is not recorded until the first week of March
(Administered_65Plus column in (3)) these models employing age-specific vaccinations are estimated
only for April-August.
Controlling for COVID-associated deaths
To control for deaths due to COVID in young adult, middle and older age groups, regressions
were also run to predict non-COVID deaths (i.e. the number of total deaths minus deaths due to
COVID, Influenza or Pneumonia which are provided in the same spreadsheet as total deaths). In
addition, another set of models similar to Eq. 1 were estimated in which log # COVID cases in the
previous month were included as an additional covariate.
Controlling for seasonality, population size, and mortality differences across states
Both the number of administered vaccine doses and number of deaths in a given time period is
highly correlated with population size of each state. However, age-stratified population size was not
adjusted for in the analyses. Rather, the analyses adjusted for the number of deaths in the same
month during the previous year. This effectively controls for age-stratified population size differences
across states while additionally controlling for seasonal effects on mortality and state-to-state
variability in mortality due to other factors. In a post-hoc analysis, age-stratified populations were
correlated with residual deaths (Y21 deaths after adjusting for Y20 deaths), which confirmed that
controlling for Y20 deaths effectively removes variance due to age-stratified population difference (all
p-values for Spearman rank correlation coefficients ranged between 0.15 and 0.9).
Regression Diagnostics and Specification Tests
An important consideration when dealing with geospatial data is ensuring that parameter
estimates are not influenced by spatial autocorrelations in the observed data that are not properly
accounted for in the model, which could result in model misspecification (i.e. residuals from the
models would cluster spatially). We examined potential misspecification by testing for statistically
significant spatial autocorrelation in the regression residuals. We determined whether regression
residuals were spatially autocorrelated in each of the 15 models with significant vaccination terms
(Table 2). A binarized (queen) contiguity map of each US state or territory and residuals for each
model were used to estimate the Moran’s I autocorrelation index and corresponding p-values
(probability of observing the calculated I given the null hypothesis of zero autocorrelation in the
residuals) using the ape v5.6-2 package (http://ape-package.ird.fr) in R v4.2.0. None of the resulting
p-values were significant after corrections for multiple comparisons (all FDR adjusted p-values >
0.05), and all but two uncorrected p-values ranged between 0.1 and 0.99. The above post hoc
diagnostics indicate the assumption of independent error terms was met.
Estimating number of deaths attributed to COVID vaccination
The estimated beta ( ) weights (regression slopes) that survived p<0.05 FDR corrected were
used to estimate death counts for that month and age group. Briefly, for each age group, the increase
in deaths attributed to a small (i.e. 10%) increase in vaccinations across all states was used to
estimate a vaccine mortality rate for each age group. The rate was then multiplied by the total count
of administered vaccination for that month to arrive at an estimated number of deaths attributed to
For each state with Y20 death data, increases in deaths due to 10% increase in vaccinations
was estimated by solving a system of equations for and , where is the estimated Y21 deaths
given the actual vaccine doses administered in that state, and is the predicted Y21 deaths given a
10% increase in the actual vaccine doses administered in that state:
Solving for yields:
The differences between and were summed across all N states with available Y20 data, and
then divided by 1/10th (10%) of the sum of vaccine doses administered across those states to
estimate an age-specific vaccine-attributed fatality rate (aVFR) for that age group and month:
Finally, the aVFR was multiplied by the total number of vaccinations in the US during the month used
in the regression model to arrive at an estimated death count attributed to vaccines for each month
and age group that survived the applied significance threshold. The values are then used to populate
the cells in Table 2.
Two approaches were used to estimate a US national average and age-stratified VFRs. The
US national average VFR was estimated by dividing the death counts estimated above, summed over
all age groups and months, by the total number of vaccine doses administered between January and
August 21st. A second approach averaged across all monthly aVFRs (equation 5) within each age
group, calculated based on thresholded regression weights (p<0.05 uncorrected), resulting from the 3
model variations (i.e. the primary model using previous month vaccination, a second model using
same month vaccinations, and a third using previous month vaccinations in ages >65 years). A liberal
threshold of p<0.05 uncorrected was used to increase the sample size of rates used for the average.
This yielded an estimated aVFR for each of the age groups analyzed in the study.
Parametric bootstrap resampling was applied to compute 95% confidence intervals about
aVFR and VFR estimates. Briefly, data for each regression model in which the vaccination beta
weight term survived p<0.05 uncorrected was sampled (with replacement) 1,000 times and outliers
greater or less than 3x standard deviation from the mean were removed. Regression models using
these resampled data were then estimated to generate a distribution of 1,000 beta weights for the
vaccination terms, which were used to compute a distribution of 1,000 aVFR (for each age category)
and VFR estimates from which 95% confidence intervals were derived.
Errors and concerns raised with vaccines safety studies of pregnant women
Although vaccination during pregnancy is reported as safe by the US CDC (11), a number of
issues and concerns have been raised with the studies supporting vaccine safety among pregnant
women. Brock and Thornley (12) and McCullough et al. (13) point out several errors in an early safety
study among pregnant women by Shimabukuro et al. (14). The original Shimabukuro et al. study
reported a spontaneous abortion rate <20 weeks gestation rate of 12.6% after vaccination, which is
similar to previously published background rates. However, their denominator includes ~700 women
who were vaccinated after the timeframe for recording the outcome had elapsed (up to 20 weeks of
pregnancy). Excluding those participants results in a spontaneous abortion incidence rate that 7-8
times higher (82%-91%) than the originally report rate. Note that the rate seems high because the
study only examined completed pregnancies and many participants were yet not followed up on at
the time of the report (at early stages the majority of completed pregnancies are expected to be
spontaneous abortions). Shimabukuro et al. has since issued correction which now states “No
denominator was available to calculate a risk estimate for spontaneous abortions” in the Table
footnotes. However, the article abstract, results and discussion still report and discuss the initial
findings of the study, including the 12.6% spontaneous abortion rate in those exposed to vaccines
before 20 weeks.
A related, more recent case-control study by Kharbanda et al. concluded “Among women with
spontaneous abortions, the odds of COVID-19 vaccine exposure were not increased in the prior 28
days compared with women with ongoing pregnancies” (15). However, contrary to the authors’
conclusions, a comment on the article by Cosentino points out that a reanalysis of the frequencies
reported in Table 1 shows the crude OR of vaccine exposure in women with spontaneous abortions is
1.07 (95% CI: 1.01-1.14, p = 0.025 by Fisher's exact test), a result that is apparently fully accounted
for by the maternal age group 16-24 y, where the crude OR is 1.37 (95% CI: 1.07-1.75, P = 0.017).
Cosentino also points out the arbitrariness of using 28 days as a window. Why not track and report
spontaneous abortion rates across all participants up through week 19 gestation? The response by
Kharbanda et al. to Cosentino states that their results differ because they controlled for confounding
variables, but they do not report statistics for the nuisance terms, making it difficult to assess which
nuisances variable accounted correlated with higher spontaneous abortions rates and why. Finally,
we note that the authors’ original analysis DOES report trend level evidence for increased risk of
spontaneous abortion (see Table 2, gestation weeks 9-13, OR 1.07, 95% CI 0.99-1.17), but the result
is not discussed by the authors elsewhere in the article.
Reasons arguing why our results evidence a causal link (not just an association) between vaccination
1. Vaccination predicts mortality in future time periods. Thus results do not reflect increases in
vaccination rates that are caused by increased mortality. Temporal precedence is a basis for
inferring causality in i.e. Granger causality analysis.
2. Our estimates for total deaths due to vaccination are strikingly similar to independent estimates
based on a fundamentally different dataset and approach based on the VAERS database that
uses data-driven, credible assumptions about the VAERS underreporting bias (16). Our
results provide independent, converging lines of evidence for vaccine-induced mortality risk,
lending further credence to their accuracy and credibility.
3. We are aware of only one variable, COVID cases, that could potentially confound our results.
This could happen IF more people get vaccinated as local COVID cases rise and COVID
deaths comprise a majority of the deaths in subsequent time periods. Below are the main
reasons why COVID case rates do NOT explain our findings:
a. An additional set of analyses that include COVID case numbers in previous month as a
nuisance regressor yielded largely similar results (Supplementary Table S5).
b. A secondary set of analyses that use non-COVID, Influenza, and Pneumonia deaths
(non-COVINFPNU) as the dependent variable yielded similar results to analyses that
use total deaths, but with larger p-values because there are substantially fewer
observations for each regression (Supplementary S6). Note that non-COVINFPNU
deaths were not used as the primary outcome because the COVID death variable is
missing for younger ages (sample size is cut in half for ages 40-49 and below 30 it is
about 10-25% of the full sample size when using Total Deaths), and for the younger age
groups it is zero for most states that do report a value.
c. Vaccination rates predict mortality in younger age groups (where COVID deaths are
much rarer), providing further support that the effects seen here are not due to COVID.
4. The existing COVID vaccine surveillance studies supporting vaccine safety contain critical
errors, issues and limitations (see Discussion, Supplementary Discussion and (17) ).
5. Our results comport with the volume and nature of responses to social media posts, the FDA
dockets for solicited public comments, and websites created to give voice to the
vaccine-injured (see below section for sample links and URLs).
6. Our US results show an age-related temporal pattern that is consistent with the mass
vaccination campaign that first targeted nursing homes and older age groups (i.e. vaccination
predicts total deaths in ages older than 75 in early 2021, and then in younger ages later in the
year). There appears to be no other explanation for this other than a causal link between
vaccination and mortality risk.
7. Given items 1-6 and the absence of other potential confounding variables, the most logical and
reasonable conclusion is that our results reflect a causal effect of COVID vaccination on
Life-altering COVID vaccine injuries: real-world evidence through personal testimonials
To help give a real-world sense of the risks and impact of COVID vaccines, readers are
encouraged to browse through some of the testimonials on c19vaxreactions.com and
nomoresilence.world, two websites dedicated to giving voice to those injured by COVID vaccines.
Readers are also encouraged to sift through the over 100K solicited comments submitted to the
public FDA advisory committee meeting held on Oct 26th, 2021 to discuss approval of the COVID
vaccines for children ages 5-11 at https://www.regulations.gov/document/FDA-2021-N-1088-0001.
Perusing through over 250K comments left on a Facebook post by WXZY-TV Channel 7 news
https://www.facebook.com/wxyzdetroit/photos/a.461583946134/10158207966696135 is also
illuminating. The post asked people who had lost an unvaccinated loved one to COVID to contact
them for a story, but instead received tens, if not hundreds, of thousands of stories of vaccine injuries
or deaths instead. The post is telling of how injured patients, or those who have lost friends or family
to vaccine-induced death, are often ignored by the same major news outlets that encouraged them to
be vaccinated. This is understandable, as no one, especially those with good intentions and high
hopes but who were misled by less-than-rigorous science, wants to acknowledge the possibility that
the COVID vaccines and their boosters may be causing more harm than good overall. The sooner the
taboo surrounding research and discussion of vaccine-induced injury and death is lifted, the sooner
public health policy can be adjusted and resources can be mobilized to identify and develop therapies
1. Death rate by age and sex in the U.S. 2018 [Internet]. Statista. [cited 2021 Oct 3]. Available from:
2. Cotton C. VAERS Data Analysis [Internet]. Available from:
3. COVID-19 Vaccinations in the United States,Jurisdiction | Data | Centers for Disease Control and
Prevention [Internet]. [cited 2021 Oct 1]. Available from:
4. Provisional COVID-19 Deaths by Sex and Age | Data | Centers for Disease Control and Prevention
[Internet]. [cited 2021 Oct 1]. Available from:
5. United States COVID-19 Cases and Deaths by State over Time | Data | Centers for Disease Control and
Prevention [Internet]. [cited 2021 Oct 1]. Available from:
6. Bureau UC. 2019 Population Estimates by Age, Sex, Race and Hispanic Origin [Internet]. The United
States Census Bureau. [cited 2021 Oct 1]. Available from:
7. Vickers AJ. The use of percentage change from baseline as an outcome in a controlled trial is statistically
inefficient: a simulation study. BMC Med Res Methodol. 2001;1:6.
8. Clifton L, Clifton DA. The correlation between baseline score and post-intervention score, and its
implications for statistical analysis. Trials. 2019 Jan 11;20(1):43.
9. Van Breukelen GJP. ANCOVA versus change from baseline had more power in randomized studies and
more bias in nonrandomized studies. Journal of Clinical Epidemiology. 2006 Sep 1;59(9):920–5.
10. Benjamini Y, Hochberg Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to
Multiple Testing. Journal of the Royal Statistical Society: Series B (Methodological). 1995;57(1):289–300.
11. CDC. Vaccination Considerations for People Pregnant or Breastfeeding [Internet]. Centers for Disease
Control and Prevention. 2021 [cited 2021 Nov 6]. Available from:
12. Brock AR, Thornley S. Spontaneous Abortions and Policies on COVID-19 mRNA Vaccine Use During
Pregnancy. Science, Public Health Policy, and the Law. 2021 Nov;4:130–43.
13. Lack of Compelling Safety data for mRNA COVID Vaccines in Pregnant Women [Internet]. TrialSiteNews.
2021 [cited 2021 Nov 6]. Available from:
14. Shimabukuro TT, Kim SY, Myers TR, Moro PL, Oduyebo T, Panagiotakopoulos L, et al. Preliminary
Findings of mRNA Covid-19 Vaccine Safety in Pregnant Persons. New England Journal of Medicine.
2021 Jun 17;384(24):2273–82.
15. Kharbanda EO, Haapala J, DeSilva M, Vazquez-Benitez G, Vesco KK, Naleway AL, et al. Spontaneous
Abortion Following COVID-19 Vaccination During Pregnancy. JAMA. 2021 Oct 26;326(16):1629–31.
16. Rose J, Crawford M. Estimating the number of COVID vaccine deaths in America [Internet]. Available
17. Pantazatos S. Vaccine mandates are not based on sound science: they are harmful and should be lifted
as soon as possible. 2021 Aug 23 [cited 2021 Sep 9]; Available from: https://researchers.one/
18. Pantazatos S, Seligmann H. Supplementary Tables and Figures - Google Drive for “COVID vaccination
and age-stratified all-cause mortality risk” [Internet]. [cited 2022 Feb 20]. Available from:
Supplementary Table S1. Weekly increases in percent vaccinated in 23 European countries,
Coronavirus (COVID-19) Vaccinations - Statistics and Research - Our World in Data. See
Supplementary Table 1 spreadsheet (18).
Supplementary Table S2. Weekly total mortality data for 6 age classes for the first 33 weeks of 2021
from 23 European countries (https://www.euromomo.eu/graphs-and-maps). See Supplementary Table
2 spreadsheet (18).
Supplementary Table S3. Pearson correlation coefficient r between weekly injection percentage and
weekly mortality for 6 age classes (appendices 1 and 2) for 23 European countries. See
Supplementary Table 3 spreadsheet (18).
Supplementary Table S4. COVID Cases, prior month vaccinations and age-stratified mortality for
April 2021. Cumulative number of vaccinations or COVID cases are as of April 1st, 2021. See
Supplementary Table 4 spreadsheet (18). For the same tables for all other months see the Tables
subfolder in the git repo (https://github.com/spiropan/CoVFR).
Supplementary Table S5. Same as main text Table 2, except models adjust for previous month
COVID cases. For each month in 2021, beta weights and uncorrected p-values are listed for the
vaccination (b3) term in the GLM equation: log(Total Deaths Y21) ~ b0+b1*log(Total Deaths Y20) +
b2*log(previous month COVID cases)+b3*log(vaccine doses administered previous month) across all
US states with available data for that month and age group (~42-52 states for each regression).
Yellow indicates positive slopes with p-values < 0.05 FDR corrected.
Supplementary Table S6. Same as main text Table 2, except the dependent variable is
Non-COVID-Influenza-Pneumonia (COVINFPNU) Deaths. For each month in 2021, beta weights
and uncorrected p-values are listed for the vaccination (b2) term in the GLM equation:
log(Non-COVINFPNU Deaths Y21) ~ b0+b1*log(Non-COVINFPNU Deaths Y20)+b2*log(vaccine
doses administered previous month) across all US states with available data for that month and age
group. Note that because COVID deaths are relatively rare among younger age groups, there are
much fewer states with available data for Non-COVINFPNU deaths, particularly for the ages 0-49
(denoted with an asterisk). There were <9 data points for ages 0-17, <15 for 18-29, <18 for 30-39,
and <28 for ages 40-49. Yellow (light peach) indicates positive slopes with p-values < 0.05 FDR
corrected (p<0.05 uncorrected).
Supplementary Table S7. Same as Table 3 of main text, except deaths were estimated based on
robust regression results thresholded at p<0.05 uncorrected.
Supplementary Table S8. Model-estimated deaths attributed to COVID-19 vaccination for each
age group and month. Same as Table 3 of main text, except deaths were estimated based on
standard linear regression (glmfit MATLAB function) thresholded at p<0.05 FDR corrected. Beta
weight coefficients estimated from Equation 1 and surviving p<0.05 FDR corrected were used to
estimate VFR and total deaths for each age group and month. If a model using same (not previous)
month vaccinations was significant and the equivalent models using previous month was not, then
death counts from those models were used instead (light gray boxes). Similarly, if a model using
age-specific (i.e. >65 yrs) vaccine dose administrations was significant and the equivalent models
using total vaccine doses administered was not, then death counts from those models were used
instead (dark gray boxes). See methods for VFR and aVFR definitions and calculations. ns=not
significant at p<0.05 FDR corrected. NA=Not available.
Estimated Deaths and aVFR
Supplementary Table S9. Model-estimated deaths attributed to COVID-19 vaccination for each
age group and month. Same as Supplementary Table S5, except deaths were estimated based on
standard linear regression (glmfit MATLAB function) thresholded at p<0.05 uncorrected. Beta weight
coefficients estimated from Equation 1 and surviving p<0.05 uncorrected were used to estimate VFR
and total deaths for each age group and month. If a model using same (not previous) month
vaccinations was significant and the equivalent models using previous month was not, then death
counts from those models were used instead (light gray boxes). Similarly, if a model using
age-specific (i.e. >65 yrs) vaccine dose administrations was significant and the equivalent models
using total vaccine doses administered was not, then death counts from those models were used
instead (dark gray boxes). See methods for VFR and aVFR definitions and calculations. ns=not
significant at p<0.05 uncorrected. NA=Not available.
Supplementary Figure S1. Plots of log transformed vaccination vs. monthly Y21 deaths adjusted for Y20 deaths. Results are plotted for
each model in which the vaccination terms was significant at p<0.05 FDR corrected (see Table 2 and Table 3 of main text). ns=not significant. For
higher resolution images see Supplementary Figure S1 (18).