Did the COVID-19 vaccines save millions of lives in the USA?
Quantitative assessment of published claims
By Prof Denis Rancourt and Dr. Joseph Hickey
Abstract
Many media and public-record statements, including Congressional statements and testimony, since 2022, have often asserted that COVID19 vaccination in the USA prevented some 100 million infections, saved some millions of lives, saved some tens of millions of hospitalizations, and saved some 1 trillion dollars in associated medical costs. These fantastic and unverifiable claims are based on theoretical models of so-called counterfactual scenarios, which are back predictions under hypothetical absence of COVID19 vaccination.
The said claims are reported in several scientific articles, often in leading scientific journals, however their authors sparingly show and essentially never examine the time evolution of the back predictions for plausibility. We calculate time evolutions corresponding to the back predictions.
We show that if one accepts the counterfactual models and their inputs to then calculate the corresponding excess all-cause mortality that would have occurred, then one graphically obtains excess all-cause mortality by time (by week) that is contrary to realistic behaviours.
By accepting the counterfactual models, we must believe that the two main COVID-19 vaccination campaigns (doses 1+2 and first-booster dose rollouts, in early and late 2021, respectively) coincidentally were each applied just in time prior to two staggering spontaneous many-fold increases in viral virulence.
In other words, we must believe that the massive and repeated COVID-19 vaccine rollouts did not significantly reduce mortality in 2021 and in 2022 compared to 2020 (they actually did not) because the virus became more virulent than ever in those years, twice, in early 2021 and in late 2021―early 2022, producing 5fold hypothetical increases in excess all-cause mortality by year.
The counterfactual scenarios are so improbable that they can, on the sole basis of the predictions themselves, be qualified as impossible.
Table of Contents
Abstract
Table of Contents
1. Context and Purpose
2. Nature of the counterfactual exercise
3. Data and method for testing plausibility of calculated number of deaths averted
3.1. Data
3.2. Calculation of counterfactual mortality by time
3.3. Actual measured excess all-cause mortality by time
4. Results
5. Discussion
6. Conclusion
References
Appendix
1. Context and Purpose
On 10 October 2024, Dr. Peter Hotez testified to Congress under legal obligation to tell the truth as follows (transcript, at pages 42-43):
Table 1. Extract of Congressional interview of Peter Hotez, 10 October 2024.
.
.
Here, “[t]he studies by my colleague and friend Alison Galvani at Yale” and all of this refer to a blogpost by Fitzpatrick et al. (2022) dated 13 December 2022 on the website of The Commonwealth Fund.
The said blogpost (Fitzpatrick et al., 2022) gives results entirely based on counterfactual modelling (that is, back predictions under hypothetical absence of COVID-19 vaccination), without giving sufficient detail to allow scientific verification of either the calculation itself or its inputs, whereas related methods have apparently been used by the authors elsewhere (Pandey et al., 2022; Sah et al., 2022; Vilches et al., 2022a). The claims and the explicit inputs and methods, to our knowledge, have not been explained in a follow-up publication.
The said claims (Fitzpatrick et al., 2022) have been uncritically and disproportionately covered in mainstream media, since its posting and recently, for example, as follows.
New York Times
“How many lives have been saved by Covid vaccines?
Scientists believe that the vaccines have prevented millions of deaths. A study in the journal Lancet Infectious Diseases estimated that the shots saved 14.4 million lives worldwide in the first year alone.
In the United States, they are thought to have prevented more than 18.5 million hospitalizations and 3.2 million deaths by the end of 2022.” (2025-09-02)
https://www.nytimes.com/2025/09/02/health/trump-covid-vaccines.html
***
CBC
“ “The mRNA technology has been proven to be highly effective,” Hotez said. “By some estimates, 3.2 million American lives were saved by COVID mRNA vaccines during the pandemic.” ” (2025-08-25)
https://www.cbc.ca/news/health/mrna-vaccine-barda-explainer-1.7602830
***
ABC News
““Here’s the bottom line: mRNA vaccines for COVID, according to estimates from Yale School of Public Health, saved 3.2 million lives,” Dr. Peter Hotez, a professor of pediatrics and molecular virology at Baylor College of Medicine in Houston, told ABC News.” (2025-05-23)
https://abcnews.go.com/Health/safety-efficacy-mrna-vaccines-amid-recent-scrutiny/story?id=122068940
***
CNN
“The Covid-19 vaccines have kept more than 18.5 million people in the US out of the hospital and saved more than 3.2 million lives, a new study says – and that estimate is most likely a conservative one, the researchers say.” (2022-12-13)
https://www.cnn.com/2022/12/13/health/covid-19-vaccines-study
***
The Hill
“The Commonwealth Fund estimated the vaccines prevented more than 18.5 million hospitalizations and 3.2 million deaths from December 2020 to last month.
Researchers added the vaccines also prevented 120 million more COVID-19 infections and saved the U.S. more than $1 trillion.” (2022-12-13)
***
Yale School of Public Health (Facebook account)
VIDEO: “A sampling of media coverage of the new Yale School of Public Health and The Commonwealth Fund analysis that found U.S. COVID vaccinations have saved 3 million lives.” (2022-12-19; accessed 2025-09-27)
https://www.facebook.com/YaleSPH/videos/554401139547193
***
It is important to rigorously assess influential claims ― both in the context of the body of relevant science and for logic and validity of the underlying assumptions ― because the said claims may be demonstrably incorrect. False claims accepted by government officials and their advisors can have a disastrous effect on public health policy and society. The analysis below shows that Dr. Peter Hotez, in particular, appears not to have critically ascertained the value of the claims in the blogpost of Fitzpatrick et al. (2022) but instead relied on the status of its senior author (his friend; and collaborator, as per Bartsch et al., 2021) and her institution.
2. Nature of the Counterfactual Exercise
The blogpost results of Fitzpatrick et al. (2022) were obtained by a counterfactual theoretical calculation. In counterfactual analysis (BGI Consulting, 2007) “the outcomes of the intervention are compared with the outcomes that would have been achieved if the intervention had not been implemented.” Here, the intervention is COVID-19 vaccination in the USA. Therefore, the counterfactual theoretical calculations involve back predictions of mortality and other outcomes under hypothetical absence of vaccination. It is impossible to actually know how many people would have died (or been saved) from not being vaccinated. Instead, the number of lives saved is calculated using elaborate theoretical hypotheses and inputs (such as disease characteristics and vaccine efficacy) presumed to be valid.
In order to estimate the number of lives saved, counterfactual modellers need to estimate how many SARS-CoV-2 infections would have occurred through time without vaccination, and how many of these infections would have caused death. Simply put, the vaccine cannot save you if you would not have been infected. This brings us to arguably the most tenuous part of the counterfactual calculation used by Fitzpatrick et al. (2022): The hypothetical prevalence (infections) by time is calculated by contagion dynamics modelling, which has its own assumptions, complexities and uncertainties; not to mention that it may not be applicable whatsoever (Hickey et al., 2025).
The other main difficulty is that the modellers assume in all current vaccine counterfactual calculations that the vaccine efficacy inputs are reliable, despite being produced via contrived, questionable and non-transparent clinical trials (Gøtzsche, 2013; Rancourt, 2025a; Siri, 2025).
Fitzpatrick et al. (2022) concluded that in the USA:
- 3.2 million deaths
- 18.5 million hospitalizations
- 120 million infections
- $1.15 trillion in medical costs
were averted up to 2022-11-30 (two year period) by COVID-19 vaccination.
Using the same kind of counterfactual approach (contagion dynamics modelling + presumed-valid vaccine efficacy estimates), Watson et al. (2022) obtained a similarly fantastic result for the USA (taking the average of their two similar scenarios):
- 1.83 million deaths
averted up to 2021-12-08 (one year period) by COVID-19 vaccination.
The Watson et al. (2022) study has been fundamentally criticized (Rancourt and Hickey, 2023; Ophir et al., 2025; Sorli, 2025).
The counterfactual analysis of Ioannidis et al. (2024, 2025) gives a number of lives saved by COVID-19 vaccination in the USA through 2024 (four year period) that is an order of magnitude (10 times) less than the numbers obtained by Fitzpatrick et al. (2022) and by Watson et al. (2022).
Ioannidis et al. (2024, 2025) obtain 2.5 million lives saved globally through 2024. Using the fractions USA / World of lives saved obtained by Watson et al. (scenario 1: 1.76/14.4 and scenario 2: 1.90/19.8; from their supplementary material), the Ioannidis et al. global value (2.5 million) corresponds for the USA to:
- 270 thousand deaths
averted through 2024 (four year period) by COVID-19 vaccination.
The results of Fitzpatrick et al. (2022) and Watson et al. (2022) on the one hand are irreconcilable on their face with the results of Ioannidis et al. (2024, 2025) on the other hand. Both Fitzpatrick et al. (2022) and Watson et al. (2022) used “contagion dynamics modelling + presumed-valid vaccine efficacy estimates” whereas Ioannidis et al. (2024, 2025) used “seroprevalence data and reported COVID-19 deaths + presumed-valid vaccine efficacy estimates” in the counterfactual method, thus avoiding contagion dynamics modelling.
Rancourt (2025b) critically assessed the papers of Ioannidis et al. (2024, 2025), and finds no reliable reason to belief that the COVID-19 vaccines saved any lives. Nonetheless, it appears that authors can misguide themselves far more with contagion dynamic modelling than with seroprevalence data, as most epidemiologists would expect.
A better way is to devise a method that avoids counterfactual models altogether, and relies solely on measured data, especially higher reliability data such as mortality. This was done by McNamara et al. (2022). They performed an ecological study in which they exploited the established fact that assigned COVID-19 mortality, like all-cause mortality, is highly dependent on age, in fact exponential with age. They cleverly used death in lower age groups to normalize death in the elderly age groups, prior to and during vaccination, in the USA in the period 2020-11-01 to 2021-04-10. In their extensive analysis of the data, McNamara et al. (2022) found a null result: “the magnitude of the impact of vaccination roll-out on deaths was unclear.”
Therefore, in this case the sequence of reliability is:
.
.
It is not surprising that using contagion dynamics modelling is unreliable for hindcasting anything about the declared COVID-19 pandemic in that Hickey et al. (2025) have demonstrated that high-resolution and global geotemporal variations in excess all-cause mortality are incompatible with viral spread of a respiratory disease.
However, the purpose of the present article is not merely to point out the unreliability of the Fitzpatrick et al. (2022) blogpost, and published articles using the same counterfactual approach (Ogden et al., 2022; Shoukat et al., 2022; Steele et al., 2022; Vilches et al., 2022a, 2022b; Watson et al., 2022; Yamada et al., 2023).
Rather, we provide a proof that the Fitzpatrick et al. (2022) method implies impossible consequences when their counterfactual mortality is plotted by time (deaths per week) over the period of application, irrespective of any criticism of the validity of their method and inputs.
We have previously applied our demonstration of implausibility to 95 countries studied by Watson et al. (2022) (Rancourt and Hickey, 2023). Here, we apply the said demonstration of implausibility in more detail to the USA for the blogpost results of Fitzpatrick et al. (2022). The same holds for all the studies using essentially the same counterfactual approach with comparable presumed COVID-19 vaccine efficacies, such as: Ogden et al. (2022), Sah et al. (2022), Shoukat et al. (2022), Steele et al. (2022), Vilches et al. (2022a, 2022b), Watson et al. (2022), and Yamada et al. (2023).
3. Data and Method for Testing Plausibility of Calculated Number of Deaths Averted
3.1. Data
All-cause mortality and vaccination data by time for the USA and its states are from the US Department of Health and Human Services (HHS, 2025a; HHS, 2025b).
3.2. Calculation of counterfactual mortality by time
For a given long period (many months), a formula for the counterfactual number of lives saved is as follows. One must derive the number of deaths, D0, that should occur from the presumed pathogen in the absence of the intervention (i.e., without vaccination) and use an estimate of the vaccine efficacy (taken to be real-world efficiency), Evd, in preventing deaths. Evd is the vaccine-attributed reduction of probability of death per person presumed to be fatally infected. Then, the number of lives saved, Ls, (or deaths averted) is the product of D0, vaccine coverage Cv (expressed as a fraction of the population considered to be vaccinated) and Evd, in the period considered:
Ls = D0 x Cv x Evd. (1)
Here, “coverage” implies boosters intended to combat both waning vaccine potency and the presumed emergence of new variants. Cv can be known with relative certainty, whereas D0 and Evd are disjunctively problematic. Estimates of D0 usually rely on contagion dynamics modelling or seroprevalence and mortality data, whereas estimates of Evd are inferred from limited clinical trial data.
If we want the counterfactual time series of lives saved (by week) during and after vaccination rollouts, then we calculate the number of lives saved at each given time “t” (at the dates of each given week), Ls(t), as follows.
First, the number of infections saved (by vaccination) at each given time “t”, ISv(t), is calculated as:
ISv(t) = nI(t) x mCv(t) (2)
where
- nI(t) is the number of infections that would hypothetically have occurred at time “t” (in the given week occurring at or corresponding to time “t”) in the absence of the intervention (vaccination).
- mCv(t) is the vaccine-produced-immunity coverage at time “t” against infection, by definition between 0 and 1.
Here, mCv(t) is given by the mathematical convolution
mCv(t) = (Cv⊗Ev)(t) (3)
where
- Cv(τ) is the function by time of vaccine coverage, expressed as a fraction from 0 to 1 per time.
- Ev(τ) is the function by time (since vaccination) of vaccine efficiency against infection (0 to 1 per time), which includes both a delay since vaccination and efficiency waning.
Therefore, the number of infections saved (by vaccination) at each given time “t”, ISv(t), is calculated as:
ISv(t) = nI(t) x (Cv⊗Ev)(t) = nI(t)(Cv⊗Ev)(t) (4)
Next, this in turn is used to calculate the number of lives saved (by vaccination) at time “t”, Ls(t), as:
Ls(t) = (ISv⊗IFP)(t) = (nI(Cv⊗Ev)⊗IFP)(t) (5)
where
- IFP(τ) is the function by time (since infection) of the likelihood (0 to 1 per time) of death following infection, which is the time-dependent infection fatality probability.
Equation 5 is a probabilistic estimate rather than one that follows individuals and their personal vaccination and survival histories. Also, it is for average individuals, not age or otherwise stratified.
In application, for simplicity, we approximate IFP(τ) to be a mathematical Delta function (a unitary pulse centered at zero) multiplied by the infection fatality ratio, IFR,
IFP(τ) = IFR δ(τ) (6)
which is equivalent to assuming the death to occur at the time of infection with probability equal to the IFR (from 0 to 1). This introduces a shift of the corresponding mortality towards earlier time, with a shift magnitude equal to the mean time between infection and death for fatal infections. (This shift is cancelled as explained below, by our use of the actual excess all-cause mortality as a proxy for nI(t).)
With this simplification (Equation 6), Equation 5 for Ls(t) becomes:
Ls(t) = IFR x ISv(t) = IFR nI(t) (Cv⊗Ev)(t) (7)
again, where the constant IFR is the infection fatality ratio.
Also, in application, for simplicity, we take Ev(τ) to be a zero-to-constant rectangular response function of 182-day (26-week) width, delayed by 14 days (2 weeks) following vaccination. Likewise, we take Cv(τ) to be proportional to total vaccine doses delivered per unit time (by week), where the Ev(τ) response function automatically prevents double counting of injections; and we take nI(t) to be proportional to actual measured excess all-cause mortality (Section 3.3) at “t”.
The latter application to obtain nI(t) assumes that actual measured excess all-cause mortality is a valid proxy for potentially lethal prevalence of the virus, rather than using measured new cases data. However, this choice of proxy shifts the thus obtained nI(t) towards later time, the mean time between infection and death for fatal infections, which is cancelled by the shift arising from our application of Equation 6 (delta function). In any case, these shifts are individually small compared to the duration of the declared pandemic.
The proportionality constants (for the proxy and including IFR) are automatically obtained by normalization to the claimed number of lives saved being tested.
We verified that the results are insensitive to our simplifications on the illustrated time scale of the graphical displays. In particular, we tested rectangular response function widths (vaccine non-waning durations) of 91 days (13 weeks) and 364 days (52 weeks).
Rancourt and Hickey (2023) used the same calculation method, with the additional simplification that nI(t) was assumed to be a time-independent constant in the period of interest, in testing counterfactual scenarios of Watson et al. (2022).
In this way, we calculate the number of lives saved by time increment (by week) and normalize the result to the number of lives saved reported by authors of any given counterfactual estimate, over the time period of the reported counterfactual estimate.
We thus obtain the all-cause mortality by time that corresponds to the counterfactual scenario being tested and add it to the actual measured excess all-cause mortality by time. In other words, we calculate what total all-cause mortality by time would have been if the reported counterfactual calculation was a correct representation of reality and no COVID-19 vaccines were administered.
3.3. Actual measured excess all-cause mortality by time
Actual measured excess all-cause mortality by time (week) and its one-standard-deviation uncertainty are calculated as follows. The method has been explained and amply illustrated by Rancourt et al. (2024).
The excess all-cause mortality at a given time (week) is the difference (positive or negative) between the reported all-cause mortality for the given time and the expected all-cause mortality for the given time, which is ascertained from the historic allcause mortality in a reference period immediately preceding the Covid period (prior to the 11 March 2020 World Health Organization declaration of a pandemic).
In practice, our reference period is 2015 through 2019. We least-squares fit a straight line to the same week in each of the five reference years as the week of interest, where the slope of this fitted line is constrained to always (for every week of interest) be equal to the slope of a least-squares fitted line to all of the all-cause mortality data (all weeks) in the full 5-year reference period, for the given country or state.
The thus obtained fitted line is used (by extrapolation) to predict the expected all-cause mortality. The one-standard-deviation (1σ) uncertainty in the expected all-cause mortality is estimated as sqrt(π/2) times the average magnitude of the 5 deviations in the 2015-2019 reference period, for each particular week of interest. This simple relation is exact in the limit of a large sampling number, for a normally distributed uncertainty.
Finally, the one-standard-deviation uncertainty of the actual excess all-cause mortality is the combined error that includes the 1σ uncertainty in the expected value and the independent statistical (1σ) error in the all-cause mortality (sqrt(N)).
4. Results
The calculation of counterfactual mortality by time (Equation 7) and the corresponding total all-cause mortality by time that is advanced as the postulated reality in the hypothetical absence of vaccination is illustrated for the USA, 2018-2024, for the counterfactual results (3.2 million lives saved) of Fitzpatrick et al. (2022), in Figure 1.
For the rest of this article please go to source link below.