# COVID vaccine risks and disease risk comparison

Research addendum checked 6 October 2026. This is an editorial evidence memo, not individualized medical advice. It complements the vaccine dossier and prioritizes the user's requested baseline–vaccination–infection context.

## Main conclusion

A defensible page can say: **COVID vaccines can cause serious harm, including rare deaths. Claims that they killed millions are not supported by reliable causal evidence.** Explain the evidence rather than asking readers to trust an institution. “Millions harmed” is too undefined for a yes/no verdict: billions of doses can cause very large numbers of temporary fevers and sore arms, while permanent disability and death are different outcomes requiring different evidence.

Do not use “no mortality signal” to mean zero vaccine-caused deaths. Population all-cause mortality measures net outcomes: prevented disease deaths can coexist with vaccine-caused deaths, and observational bias can obscure effects. Triangulate clinical investigation, active surveillance, product-specific safety studies, and controlled mortality evidence.

## The numeric myocarditis comparison to publish now

### Same endpoint and 42-day window, two matched comparisons

[Barda et al., NEJM 2021, original Pfizer BNT162b2 primary series in Israel](https://pmc.ncbi.nlm.nih.gov/articles/PMC8427535/), DOI 10.1056/NEJMoa2110475:

- Vaccination versus matched unvaccinated controls: **27 additional myocarditis cases per million people** within 42 days; 95% CI **10–46**.
- Documented infection versus separately matched uninfected controls: **110 additional cases per million** within 42 days; CI **56–158**.
- Ages ≥16; no documented previous infection; vaccination December 2020–May 2021. The vaccination comparison includes the first and second doses. These are historical estimates, not current booster estimates.
- Table 2 directly reports 21 myocarditis events in vaccinated people versus 6 in 938,812 matched controls per group. These counts must not be divided by starting sample size and labeled 42-day risk: censoring occurred. Kaplan–Meier estimation produced the reported excess risks.
- Full main text inspected, particularly Methods, Table 2, and Discussion. Supplement Figure S1 contains absolute-risk curves, but its file could not be retrieved. **A directly tabulated common absolute baseline is not verified.** Do not publish reconstructed rounded-RR baselines as measured values. The two comparisons have different population composition; they are not a randomized three-arm trial.

### The important young-male exception

[Patone et al., Circulation 2022](https://pmc.ncbi.nlm.nih.gov/articles/PMC9439633/), DOI 10.1161/CIRCULATIONAHA.122.059970. In males under 40 in England, 2020–21, **additional hospital admissions/deaths with myocarditis per million exposed during days 1–28** were:

- Pfizer BNT162b2 second dose: **11 (95% CI 9–13)**
- Moderna mRNA-1273 second dose: **97 (91–99)**
- Positive SARS-CoV-2 test before vaccination: **16 (12–18)**

These are excess events above modelled baseline, not total risks. This refutes “infection always causes more myocarditis than vaccination.” It does not establish a complete net-benefit comparison. Self-controlled case series controls stable individual characteristics; undetected infections, diagnostic coding, time-varying factors, and inclusion only of people eventually vaccinated remain limitations. Sparse adolescent events prevented a separate estimate. Indexed full methods/results/limitations inspected in the preceding dossier. Locate the paragraph beginning “In men younger than 40 years” and Table 4.

### How to show the missing baseline honestly

Use a label “Additional cases above the study's background rate,” not a bar labeled “baseline = 0.” Offer two Barda paired-comparison panels and a separate young-men panel. State: “We haven't verified one common baseline suitable for all three groups.” Background myocarditis exists and differs by sex, age, case definition and period. Do not compare an annual general-population rate with a seven-day post-dose rate or confuse myocarditis alone with myocarditis/pericarditis. Baseline verification remains a specific source-access gap, not a reason to withhold the verified excess comparisons.

Do not use the FDA's 2023–24 formula estimates (8 per million overall aged 6 months–64; 27 per million males 12–24, days 1–7) in this historical chart: those are **unadjusted observed myocarditis/pericarditis incidence**, with different product era, window and endpoint. Source and limitations are in the earlier dossier.

## What prominent claims actually say

### Seventeen million deaths

[Rancourt et al. September 2023 report landing page](https://correlation-canada.org/covid-19-vaccine-associated-mortality-in-the-Southern-Hemisphere/) and [author's contemporaneous summary](https://denisrancourt.substack.com/p/covid-19-vaccine-associated-mortality) extrapolate a 17-country ecological estimate to roughly 17 million deaths worldwide from 13.5 billion injections. The original PDF download failed; the summary was inspected. Do not pretend this is a full independent numerical replication.

In his [7 December 2024 explanation](https://denisrancourt.ca/entries.php?id=146&name=2024_12_07_did_the_c19_vaccine_kill_17_million_yes_but_how_not_what_you_think), Rancourt acknowledges that toxicity estimates are far smaller than the claimed total, then attributes deaths to accompanying rollout measures and medical practices. Full page inspected; passages under “The paradox,” “A solution to the paradox,” and “Conclusion.” That is materially different from 17 million deaths caused by injections themselves.

Editorial analysis: coincident national mortality and rollout peaks cannot identify individual causation, distinguish infection waves and care disruption, or justify transferring one pooled fatality rate to heterogeneous countries/products/ages. An uncertainty interval around an ecological model does not include all uncertainty in its causal assumptions. Critique that inferential leap, not the author's personality.

### Twenty-six million injuries and disability claims

[Phinance Technologies, Vaccine Damage Project—Human Cost](https://phinancetechnologies.com/HumanityProjects/The%20VDamage%20Project%20-%20Human%20%20Cost.htm), March 2023, claims 26.6 million injured employed Americans and connects this estimate to absence and lost-worktime changes. It discusses disability and excess deaths too. Full page inspected, particularly “Group 2” and “Death—Extreme outcome.” The page acknowledges difficulty separating vaccination from other causes of excess deaths.

Editorial analysis: work absence is not a diagnosis of vaccine injury. These aggregate figures do not establish which people were vaccinated, when an illness began, its cause, or how infection and other causes contributed. Do not describe the claimed 26.6 million as 26.6 million verified permanent injuries; the source's injury category includes temporary effects. Do not use mortality studies alone to dismiss all nonfatal illness.

### VAERS totals multiplied by an underreporting factor

The [official VAERS interpretation guide](https://vaers.hhs.gov/data/dataguide.html) explains that reports include both coincidental and vaccine-caused events. The [FDA surveillance explanation](https://www.fda.gov/vaccines-blood-biologics/safety-availability-biologics/covid-19-vaccine-safety-surveillance) explains why numerator-only signal detection cannot itself supply adverse-event incidence. These reports matter: they can detect unusual patterns and trigger investigation. They are neither a list of proven injuries nor meaningless anecdotes.

Important current data change: since May 2025, public VAERS files include secondary reports about previously reported patient/vaccine/dose events. Additional records need not mean additional patients/events. Deduplicate appropriately before any trend analysis.

The often-cited [2011 AHRQ ESP:VAERS grant report](https://digital.ahrq.gov/sites/default/files/docs/publication/r18hs017045-lazarus-final-report-2011.pdf) includes the phrase “fewer than 1% of vaccine adverse events are reported” (Results, printed page 6). Its data concern 2006–09 and possible events after many vaccines. It does not estimate the reporting sensitivity of COVID-vaccine deaths in 2021 onward. Multiplying every COVID VAERS death by 100 incorrectly assumes both that each report is causal and that one reporting factor applies across severity, vaccine, reporting requirements and era. Underreporting is real; it must be measured for the question at hand.

### The one in eight hundred trial claim deserves a direct answer

[Fraiman et al., Vaccine 2022](https://pmc.ncbi.nlm.nih.gov/articles/PMC9428332/), DOI 10.1016/j.vaccine.2022.08.036, reclassified serious adverse events of special interest in the early Pfizer/Moderna randomized trials. Pooled excess: **12.5 events per 10,000 participants (95% CI 2.1–22.9)**. This is a randomized-data safety signal worth discussing, not a VAERS analysis. Full indexed methods/results inspected; direct PMC open met a browser check.

It is not a death estimate or a demonstrated rate of permanently injured unique people. Some people had multiple events; authors adjusted standard errors for clustering but lacked participant-level data. The composite grouped different diagnoses, used an expanded AESI list and short early-trial follow-up, and excluded COVID efficacy endpoints. A complete benefit–risk assessment needs patient-level adjudication, severity, overlapping outcomes and a common time horizon. Criticism should not pretend randomization disappears simply because the analysis was secondary. Further scrutiny is justified; multiplying the pooled point estimate across every current dose/product/population is not.

## Strong population mortality evidence and its limits

### France linked records

[Semenzato et al., JAMA Network Open, December 2025](https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2842305), DOI 10.1001/jamanetworkopen.2025.46822. Linked national records of 28.7 million adults aged 18–59 did not show increased four-year all-cause mortality after mRNA vaccination. Main analysis began six months after index; early mortality was studied separately using adapted self-controlled models. Methods included weighting for 41 comorbidities and socioeconomic variables, plus negative controls. Full main methods/limitations and December 19 author response inspected.

Use the authors' restrained conclusion, not “vaccination reduces every cause of death by 25%.” Healthy-vaccinee and other residual confounding remain; the six-month main-analysis lag must be disclosed. This supports absence of a large net mortality increase in the studied population, not proof that every product, age or death mechanism is harmless. Inspect “Statistical Analysis,” “Limitations,” and the author response addressing confounding and follow-up.

### United States active surveillance

[Xu et al., Mortality risk after COVID-19 vaccination, Vaccine](https://stacks.cdc.gov/view/cdc/158686/cdc_158686_DS1.pdf). Full 16-page author manuscript inspected. Modified self-controlled analyses at eight Vaccine Safety Datalink sites studied deaths December 2020–August 2021 and 14/28-day windows after primary series. No increased overall, non-COVID or cardiac mortality was found. The method accommodated death preventing subsequent vaccination, multiple doses and calendar time. Pages 7–8 explain differences from a Florida analysis and acknowledge residual end-of-life vaccination-selection bias, incomplete cause-of-death data, inadequate power for cardiac mortality in men under 40, and short follow-up. It cannot settle that subgroup or long-term risks alone.

### England young adults

[Nafilyan et al., Nature Communications 2023](https://www.nature.com/articles/s41467-023-36494-0.pdf), DOI 10.1038/s41467-023-36494-0. Full PDF inspected. Self-controlled analysis of ages 12–29 found no significant overall increase in cardiac/all-cause mortality in the 12 weeks after vaccination. It found a cardiac-death association after first non-mRNA doses in women (IRR 3.52, CI 1.71–7.26); this should not be hidden. Positive SARS-CoV-2 tests were associated with increased mortality. Subgroup size, vaccination prioritization of clinically vulnerable young people, exposure and death registration limit interpretation. A null overall result does not erase the subgroup signal or establish exact zero risk. Inspect Results subgroup analyses and Discussion.

### A serious methodological counterargument

[Chemaitelly et al., eLife 2025, Qatar](https://pmc.ncbi.nlm.nih.gov/articles/PMC12148324/), DOI 10.7554/eLife.103690. Indexed full methods/discussion inspected. Despite careful matching, the primary-series cohort's non-COVID mortality hazard ratio was 0.35 in the first six months and 1.52 thereafter; booster comparisons also changed over time. Authors interpret the early pattern as healthy-vaccinee selection, with composition/indication effects complicating later periods, and still report protection from severe COVID. This demonstrates why neither early low non-COVID mortality nor later higher mortality can automatically be assigned to the vaccine. It is a concrete reason to triangulate designs, not dismiss observational research wholesale.

## Known serious harms must be visible

- [CDC December 2021 Janssen review, Table 1](https://www.cdc.gov/mmwr/volumes/71/wr/mm7103a4.htm): 54 TTS cases and 8 attributed deaths among 14,080,087 doses through August 2021, approximately **3.8 cases and 0.6 deaths per million doses**. Women 30–39 had 10.6 cases and 1.9 deaths per million. These are adjudicated syndrome reporting estimates, not a comprehensive global vaccine-death tally and not mRNA risks. They show safety reports can lead to clinically meaningful causal conclusions and policy changes.
- [National Academies 2024 evidence review, Summary Table S-1](https://www.nationalacademies.org/read/27746/chapter/2): causal relationship established for myocarditis with Pfizer/Moderna; evidence favors causation for Janssen TTS and Guillain–Barré syndrome. For numerous other outcomes, evidence was insufficient, which means neither proven harm nor proven absence. This was an expert causal review with structured searches, not a formal systematic review or frequency estimate. Summary and methods inspected.
- Anaphylaxis, product-specific risks and lasting sequelae must be covered using current labels and follow-up studies before a comprehensive harms page ships. Avoid “all cases mild”; the earlier dossier includes FDA follow-up uncertainties for myocarditis.

## A benefit example with an actual absolute denominator

[Nordic nationwide cohort, BMJ 2023](https://www.bmj.com/content/382/bmj-2022-075286), DOI 10.1136/bmj-2022-075286, compared a fourth bivalent dose with remaining at three doses among adults ≥50. At three months, BA.4–5 was associated with **91.9 fewer COVID hospital admissions per 100,000 (CI 31.4–152.4 fewer)** and **34.1 fewer COVID deaths (28.2–40.1 fewer)**. Hospitalization reductions were approximately 29.1 per 100,000 below age 70 versus 210.4 at ≥70. This shows why age matters to absolute benefit. Main indexed text, Table 2 and matching methods inspected; direct full-PDF retrieval failed.

These are observational 2022-era effectiveness estimates; residual confounding, recorded prior infection, calendar timing and selection matter. Do not subtract young men's 2021 second-dose myocarditis from this older cohort's fourth-dose benefit and call the result a valid tradeoff. A harmonized net-benefit estimate requires harms in the same population, product/dose and period.

## Standard comparison framework for every topic page

1. Define population: age, sex, clinical risk, previous infection/vaccination and geography.
2. Specify alternatives: this dose now versus no additional dose during a stated period. “Unvaccinated” and “uninfected forever” are different alternatives.
3. Specify product/formulation, dose, interval and calendar/variant era.
4. For each endpoint show baseline absolute risk, absolute risk with exposure, and difference, with uncertainty and equal follow-up. If only excess risk is available, label it and preserve the gap.
5. Separate death, hospital admission, persistent impairment and transient symptoms. Do not count several diagnoses in one person as several injured people without disclosure.
6. For vaccination decisions include residual disease risk after vaccination plus vaccine harms; disease probability without vaccination depends on future exposure and prior immunity. Do not assume everyone certainly becomes infected during a seven-day window.
7. Explain that all-cause death estimates net mortality, while nonfatal harms and quality of life require separate evaluation.

A teaching example, not a study estimate: if two otherwise comparable groups have hospitalization risks of 200 versus 100 per million over a season, the benefit is 100 fewer per million. If baseline risk is 20 versus 10, the same 50% relative effect yields 10 fewer. Populate a real page only with justified baseline estimates; do not turn this arithmetic into medical advice.

## Concise skeptical-reader page draft

### Did COVID shots kill millions?

COVID vaccines can cause serious harm, and rare vaccine-related deaths have occurred. Those cases deserve investigation and clear acknowledgment. They do not establish that millions died from vaccination. [Janssen review above]

A death reported after a shot is a reason to investigate, not automatically a death caused by the shot. Reporting systems help researchers spot patterns. They do not, by themselves, show how many deaths vaccination caused. Large studies linking vaccination and health records have not found the mass increase in deaths alleged by these claims, although each study has limitations. [VAERS guide; France and US studies above]

The comparison also matters. COVID itself can cause myocarditis and other serious illness, but the balance differs by age, product and dose. One historical English study found higher excess myocarditis risk after Moderna's second dose than after a positive test among men under 40. A page that hides this exception would be misleading. [Patone above]

We'll show the original claim, the comparison group, the numbers and what remains uncertain. “No risk,” “millions killed,” and “every dose benefits every person equally” all require more than an assertion.

## Remaining verification work before publication

Retrieve Barda's original absolute-risk curves/tables; do not replace unavailable data with inferred measured baselines. Obtain the original 17-country report if offering a detailed methods critique beyond the verified summary. A comprehensive current-formulation net-benefit matrix requires contemporary stratified safety and effectiveness data; the historical studies here are labelled illustrations. Recheck corrections/retractions and current labels at publication. No study in this addendum individually determines the exact global count of vaccine-caused deaths.