Six years later: you can take the tinfoil hats off

COVID-19 was not harmless and the vaccines were not risk-free, and the numbers that settle the argument are not the ones either side keeps quoting, because 22.1 million excess deaths, 1,245,272 American death certificates, and 97 extra cases of myocarditis for every million second doses of Moderna given to men under 40 all belong in the same honest sentence

Anyone who knows me knows I run on facts and numbers and have very little patience for anything that is neither, so here are the numbers, and every one of them below comes from a primary source I opened myself before writing a line about it.

In the United States, 1,245,272 death certificates issued between January 1, 2020, and August 8, 2026, name COVID-19 as a cause or a contributing cause of death. That figure comes from the National Center for Health Statistics, it was current as of August 13, 2026, and it is provisional in the specific sense that counts for earlier weeks are continually revised and may move in either direction as records arrive and get recoded. In Germany, the Robert Koch Institute counted 189,710 people who died of or in connection with COVID-19, a number I pulled from the institute’s own public data repository on the morning I wrote this. Worldwide, between 2020 and 2023, the World Health Organization estimates 22.1 million excess deaths against 7.0 million reported COVID-19 deaths.

Those 3 numbers are the entire argument, and I will spend the rest of this piece taking them apart, because taking them apart is the only way to show that they hold.

I am writing this because of something that keeps happening to me. I meet more and more people who have no idea how dangerous this virus actually was, not even a rough one, not even the right order of magnitude. Not deniers, not activists, not people with a theory. Ordinary people who absorbed 6 years of contradictory headlines and came out the other side with the impression that the whole thing was somewhere between a bad flu season and an administrative overreaction. That impression did not come from the data. It came from the way the data was reported to them, and from the fact that almost nobody went back to look once the sirens stopped.

So let us go back and look.

What 2020 did not have and 2026 does

In the spring of 2020 the honest position was ignorance. Nobody knew the true infection fatality ratio, because nobody knew the denominator. Nobody knew how often severe long-term sequelae would occur, because there had not been enough time for anything to become long-term. The first vaccines did not exist outside a freezer on a research campus somewhere. Governments made decisions under genuine uncertainty, epidemiological models were right in some places and wrong in others, and the political class occasionally treated a forecast as though it were a law of nature, which is a category error that predates virology by several centuries.

Six years later we know a completely different amount, and this is the part that gets lost.

We now have complete death records. We have hundreds of millions of linked medical records in national health systems that were built for billing and turned out to be excellent for pharmacovigilance. We have billions of administered doses with large-scale passive adverse-event reporting attached to them. We have cohort studies with actual control groups, excess mortality analyses that do not depend on anyone’s willingness to code a death certificate correctly, autopsy series, and age-stratified mortality across 45 countries. We also have enough distance to look at both the pandemic and our own response to it without the adrenaline.

Which makes it remarkable that 2 claims have survived intact into 2026.

The first: COVID-19 was never really that dangerous. The second: the vaccines harmed or killed more people than the virus did. Both claims are still circulating, both are still repeated with total confidence, and neither has an epidemiological basis that survives 20 minutes with the primary sources.

1,245,272 death certificates do not argue back

Start with the American number, because the American vital statistics system publishes its methodology in full and invites you to check the arithmetic yourself.

The standard objection arrives immediately, and in German it has an unbeatable rhythm that does not survive translation: “died with corona is not died of corona.” The English version is clumsier and the point is the same. If a man tests positive on admission and dies of something else entirely, counting him as a COVID-19 death inflates the total.

That objection is correct in principle. It is also the reason the statistics were built the way they were.

Cause-of-death coding does not treat a death certificate as a list of equals. Part I of the certificate carries the causal sequence, running from the immediate cause backward to the condition that started it, and Part II carries other significant conditions that contributed without being part of that sequence. The International Classification of Diseases provides explicit rules for selecting exactly 1 underlying cause from what is reported in Part I, and the underlying cause is defined as the condition that set the whole chain in motion. These are different fields, they are coded differently, and they are published separately.

Here is where the objection stops working. Across the years for which the National Center for Health Statistics publishes the breakdown, 2020 through 2023, COVID-19 was the underlying cause rather than a contributing mention in 86% of deaths coded to U07.1. Broken out by year: 91% in 2020, 90% in 2021, 76% in 2022, and roughly 66% in 2023. The declining trend is real and it is interesting, and I will come back to why it declines, because the reason is not the one you would guess.

One caveat belongs here rather than in a footnote, because leaving it out is how you lose an argument you should win. The code U07.1 is applied when a certifier reports COVID-19 as probable or presumed, without laboratory confirmation, on the grounds that the circumstances were compelling enough. That inflates the count at the margin, while the underlying-cause share deflates it in the other direction. Anyone who wants to attack the total has to carry both corrections at once, and they do not point the same way.

But 86% is not a rounding error, and it is not a bureaucratic artifact. It is the majority of a very large number.

None of this means every person who died was in perfect health beforehand, and obviously that was not the case. An 84-year-old can have coronary artery disease, type 2 diabetes, and stage 4 chronic kidney disease, and still die of SARS-CoV-2 pneumonia, and the pneumonia is still what killed him. Comorbidities do not make a cause of death meaningless. If they did, we would have to apply the same logic to influenza, to bacterial pneumonia, to myocardial infarction, and to every death certificate that names a heat wave, and the entire discipline of mortality statistics would collapse into a single line item reading “was alive, then was not.”

There is a second number in the same dataset that almost nobody quotes. Over the same period, all-cause deaths in the United States totaled 21,227,421. That puts COVID-19 on the certificate of about 5.9% of all registered American deaths since the beginning of 2020, which is roughly 1 death certificate in 17. Apply the underlying-cause share to that and you land somewhere near 1 in 20 as the stricter figure, though the 2 numbers cover slightly different windows and I will not pretend the combination is precise. Either way it is a single infectious disease, in a country with modern intensive care, over a span of 6 years.

Germany counts more cautiously, and the number still stands

The German phrasing is deliberately careful, and I want to give it the credit it deserves before I use it.

The Robert Koch Institute reports people who died “of or in connection with” COVID-19. That formulation does not claim that SARS-CoV-2 was the sole and undisputed cause in every single one of the 189,710 cases. The formulation says something close to the opposite. It says: this is a surveillance count, not a forensic adjudication, and we are telling you which one it is.

What the formulation does not permit is the reverse inference. From “we did not adjudicate each case individually” you cannot get to “these people were mostly just incidentally positive.” That is not caution, that is a different claim wearing caution’s coat.

Causes of death in medicine are almost never simple, and anyone who has read more than a handful of autopsy reports knows it. A patient in intensive care with COVID-19 may ultimately die of respiratory failure, of a pulmonary embolism, of multi-organ failure, or of a bacterial superinfection that took hold in lungs the virus had already wrecked. Every one of those is a proximate cause, and every one of them is true. Not one of them makes SARS-CoV-2 irrelevant, any more than blood loss makes the bullet irrelevant in a shooting.

The German count has one more feature worth noticing. It starts on March 9, 2020, with 2 deaths in the entire country, and it has now essentially stopped moving. On August 8, 2026, the register showed 189,709. 11 days later it showed 189,710. A single additional death across an entire country in 11 days is what an endemic respiratory pathogen does to a population with layered immunity, and it is also what a surveillance system looks like after it has largely wound down. That is the good news, and it is the reason the argument has become so easy to lose track of. When something stops killing people at scale, the memory of it killing people at scale becomes strangely negotiable.

The number that survives every conspiracy: 22.1 million

Reported COVID-19 deaths depend on testing capacity, on certification practice, on the honesty of health ministries, and on whether a country has a functioning civil registration system at all. If you distrust all of that, and there are countries where distrust is the only sane response, then you need a measure that does not depend on any of it.

Excess mortality is that measure. It asks a different question entirely. The question is not what was written on the certificate. The question is how many more people died than the historical pattern of this population predicted.

The current WHO figures are worth reading slowly. For 2020 through 2023, the organization estimates roughly 22.1 million excess deaths globally, against 7.0 million reported COVID-19 deaths. Excess mortality peaked in 2021 at 10.4 million and fell to 3.3 million by 2023. Global deaths ran 6.2% above expectation in 2020 and 17.9% above expectation at the 2021 peak. The WHO states plainly that the estimate includes both direct and indirect consequences of the pandemic, which means overwhelmed hospitals, canceled surgeries, and untreated heart attacks are in there alongside the viral pneumonias.

Roughly 2 additional deaths occurred for every reported COVID-19 death. That ratio is not evenly distributed across time, and the distribution tells you something. In 2020 there was about 1 additional death per reported death, and by 2023 there were about 9. The early gap was mostly indirect deaths from health systems under strain. The late gap is mostly underreporting, because testing collapsed and nobody was writing U07.1 on certificates anymore.

This is the number I would put in front of anyone who believes the pandemic was a statistical illusion, and I would put it there without any of the surrounding argument. You can believe every health ministry on earth lied, and for the sake of argument I am willing to grant you exactly that. The bodies still had to be buried, and burials are recorded by people with no interest in epidemiology whatsoever.

There is a corollary that cuts against the official narrative just as hard, and I want it on the record because leaving it out would be dishonest. The same WHO report notes that as of the end of 2025, only 18% of countries reported mortality data to WHO within a year, and nearly a third have never reported cause-of-death data at all. Of an estimated 61 million deaths worldwide in 2023, only about a third were reported with any cause-of-death information at all, and only about a fifth carry meaningful coded cause-of-death data. The global death registry is a wreck. The 22.1 million figure is a model laid over that wreckage, and a model laid over wreckage carries error bars wide enough to be embarrassing. It is still the best instrument we have, and it points in one direction only.

7 days in an office I could not leave

I have been treating this as a data problem, which is what it mostly is, but it is not the whole of what I know about it.

I had COVID-19 myself, and I have never in my life been that sick. My oxygen saturation fell, and breathing turned into physical work in a way that healthy people cannot really imagine, because breathing is supposed to be the one thing you never have to think about. I spent 7 days alone in my office. Going home was not an option that existed for me.

My wife was in the middle of a course of chemotherapy for multiple sclerosis, and that treatment does precisely what it is designed to do, which is to dismantle a person’s immune system down to the last working part. Carrying a respiratory virus into that house was not a risk I got to weigh against anything, because there was nothing on the other side of the scale. So I stayed where I was and shut the door.

What probably saved me is that I keep oxygen in the lab, and I hooked myself up to it. I am writing that down with a warning welded to it: it worked out, it was not a sensible thing to do, and nobody should read that sentence as a procedure. Friends left food outside the door and walked back to their cars without knocking. Bandit was with me, and a Malinois needs to go out whether or not the person on the other end of the leash can breathe. Getting down the stairs and back up again took everything I had, and I still do not have a better description of it than that.

At some point I concluded that I was going to die. That was not panic and it was not a feeling, it was an assessment, made by someone who spends his working life making assessments.

So when someone tells me in 2026 that the whole thing was invented, or that it was never worse than a bad flu season, I smile, and my face does something I have long since stopped trying to control.

Bergamo, and why the pictures were real

The videos and photographs from the first wave are the piece of this that people have quietly agreed to file away, and they are also the easiest piece to check.

Take Bergamo as the starting point. An analysis of death records held by the regional health protection agencies put the excess at 5,740 deaths in the Bergamo agency area over the first 4 months of 2020, against 2019 as the reference, which is a 2.55-fold increase in the number of deaths. The same analysis put Brescia at 3,703 excess deaths and a 1.93-fold increase. At the end of March, the daily standardized mortality ratio in Bergamo peaked at 9.4. On those days, roughly 9 times as many people died as the historical pattern predicted, in an ordinary northern Italian province with an ordinary hospital system.

Now look at what was happening inside those hospitals. The Lombardy intensive care cohort covered 1,591 consecutive patients admitted across 72 hospitals between February 20 and March 18, 2020. Of the 1,300 with respiratory support data recorded, 1,150 received mechanical ventilation and 137 received noninvasive ventilation. Median PEEP was 14 cm H2O, the fraction of inspired oxygen exceeded 50% in 89% of patients, and the median ratio of arterial oxygen partial pressure to inspired oxygen fraction was 160. In the larger Lombardy cohort of 3,988 critically ill patients through April 22, invasive mechanical ventilation was required at ICU admission in 2,929 cases, 87.3% of those with the value recorded. In the subgroup followed to completion, 915 of 1,715 patients died in the hospital, which is 53.4%.

Nearly everyone got a tube. Slightly more than half of them died anyway.

Those numbers are the reason the pictures existed. Across the Atlantic, New York City had logged 13,831 laboratory-confirmed and 5,048 probable COVID-19-associated deaths by May 2, 2020, which is 18,879 deaths in a single city in roughly 7 weeks, and the city’s own health department stated in the same report that this count probably understates the true toll. Nobody parks refrigerated trailers behind a hospital as a communications strategy. They did it because the morgues were full, and the morgues were full because the people in them were dead.

Every one of those videos was real, and so were the photographs, and so were the trucks parked at the loading bays.

What the doctors had in March 2020 was a protocol written for a different disease. Acute respiratory distress syndrome was the closest thing anyone had ever seen to what was arriving in the ambulances, so the ARDS playbook was what got used, at scale, under conditions where a single physician might be responsible for far more patients than any staffing model was ever built to allow. Whether that was the right call is a question the profession has argued about ever since, and I want to be careful about how I put the answer, because there is a popular version of it that the data does not support.

The popular version says the early intubations killed people. A meta-analysis of the studies that compared early against late intubation in COVID-19 patients found pooled hospital mortality of 32.1% and no significant difference between the two approaches, with an odds ratio of 0.81 and a confidence interval running from 0.32 to 2.00, which is a range wide enough to drive a truck through. 4 studies and 498 patients are not much to build on, and the patients intubated early were sicker on admission as measured by their organ-failure scores, which muddies the comparison further. What that adds up to is not a vindication and not an indictment. It is an open question that never got a randomized answer, because you cannot randomize people in a catastrophe.

So I will say the part that is actually established and leave the rest alone. Clinicians in Bergamo and Milan and Queens applied the best available standard to a disease nobody had ever treated, they applied it to people who were dying in front of them at 9 times the normal rate, and more than half of the ventilated patients died. That is not a scandal and it is not a conspiracy, it is what medicine looks like when it meets something new at full speed, and the people who did that work watched it happen from an arm’s length away, shift after shift, for months.

They were old anyway

This argument comes up constantly, and what is astonishing about it is not the medicine, it is what it quietly says about the person making it.

Yes, COVID-19 mortality rises steeply with age. The pattern is one of the most consistent findings in the entire pandemic literature. An analysis of age-specific death data from 45 countries combined with 22 seroprevalence studies found the infection fatality ratio lowest among children aged 5 to 9, rising log-linearly with age above 30. Log-linear means each additional decade multiplies the risk rather than adding to it, and it is why a virus with a small average lethality could empty a nursing home in a matter of weeks while barely registering in an elementary school.

Yes, the very old and the severely comorbid carried by far the highest risk. That is true, it was true from the start, and it was communicated badly by nearly everyone.

None of which gets you to the conclusion. Because the same age gradient applies to a large fraction of all lethal diseases, and we do not use it to dismiss a single one of the others. An 85-year-old with hypertension and diabetes does not suddenly acquire a remaining life expectancy of 3 weeks the moment he becomes epidemiologically inconvenient. Old people can live for years. If an infection makes someone die years earlier than they otherwise would have, that is life lost, caused by the infection, and calling it something else is not skepticism but bookkeeping fraud.

We do not have to argue this abstractly, because someone already did the arithmetic, and he did it from the position most favorable to the skeptics.

The 2025 analysis by Ioannidis and colleagues is by a wide margin the most conservative serious estimate of vaccine benefit in the literature, and I will come back to it in that role shortly. For now, take only 2 of its results. In the main analysis, more than 2.5 million deaths were averted worldwide through October 1, 2024, and 14.8 million life-years were saved. Divide the second by the first and you get 5.92 years of life per death averted. The study’s own dose figures say the same thing: 1 death averted per 5,400 doses against 1 life-year per 900 doses, which comes to 6.0 years. That second calculation is a consistency check rather than an independent method, since it is the same ratio approached from the other end, and I am flagging that because presenting it as confirmation would be exactly the kind of arithmetic theater I am complaining about. There is a further caveat worth stating: the paper says more than 2.5 million deaths averted, so 6 years is closer to an upper bound than to a point estimate.

6 years is not 3 weeks. 6 years is a grandchild getting all the way through high school.

And that number comes from the analysis that skeptics quote when they want to argue the vaccines did less than advertised. Even in the version of reality most hostile to the vaccination campaign, the people whose deaths were prevented had, on average, closer to 6 years left in them than to none.

The part nobody counts, because the people did not die

Mortality is the cleanest endpoint we have, but it is also a terrible summary of what a disease does to a population, because it only counts the ones who lost.

An Australian health survey drew a stratified random sample from a state COVID-19 database and did the thing that almost no long COVID discussion bothers to do. It included a control group. Close contacts of infected people, matched by circumstance, unmatched by infection. Among 11,174 people with confirmed infection, at a mean of 12.6 months afterward, 39.1% reported at least 1 persistent new symptom, against 20.8% of the 1,514 controls.

That gap is the whole point of the study design. A fifth of the uninfected population also reports new symptoms after a year, because human bodies produce symptoms whether or not a virus visits. The signal is the difference, and the difference is enormous.

Under a stricter definition, at least 1 persistent new symptom plus less than 80% recovery at 3 months, 14.2% of the infected cohort met criteria for clinical long COVID. Of the whole infected cohort, 3.2% still reported at least moderate problems with ordinary daily activities a full year after infection. Risk was lower for infections acquired while Delta or Omicron dominated than for those acquired under the ancestral strain, which is real and encouraging and does not make 3.2% of a very large number small.

I want to be careful here, because this is the section where an author who wanted a stronger sentence would reach for one. Long COVID prevalence estimates vary wildly across studies, definitions, and countries, self-reported symptoms are self-reported, and I am quoting a single well-designed survey from one Australian state rather than a settled global figure. There is no settled global figure. What there is instead is a controlled comparison showing that infection roughly doubled the rate of persistent new symptoms a year out, and that a small but non-trivial slice of infected people were still functionally impaired 12 months later. That much is enough on its own, and it does not need to be inflated.

Now the vaccines

Here the discussion gets more interesting, because here the claim I have been attacking has a mirror image that is equally false.

The statement “the COVID-19 vaccines were completely safe” is not true either. It was never true, it was said too often in 2021 by people who should have known better, and the damage that sentence did to public trust is difficult to overstate. Real and sometimes severe complications exist, they were found, and they were found by the systems that critics insist were designed to hide them.

Myocarditis and pericarditis occur rarely after mRNA vaccines. The CDC’s current clinical guidance says so without hedging: cases occurred most frequently in adolescent and young adult males within 7 days of a second mRNA dose, though cases were also observed in females and after other doses. The agency advises that recipients, especially males aged 12 to 39, be told about it. Development of myocarditis or pericarditis within 3 weeks of a dose is treated as a precaution against further doses, meaning subsequent doses should generally be avoided. That is not a footnote buried in an appendix. That is a standing clinical instruction on a public page.

Adenoviral-vector vaccines produced a different and nastier problem.

For the Janssen vaccine, the FDA fact sheet carries a boxed warning for thrombosis with thrombocytopenia syndrome, and the language on causality is unambiguous: “currently available evidence supports a causal relationship between TTS and the Janssen COVID-19 Vaccine.” Symptoms began approximately 1 to 2 weeks after vaccination. The highest reporting rate, roughly 8 cases per million doses administered, occurred in women aged 30 to 49. Approximately 15% of TTS cases were fatal. The clinical course shares features with autoimmune heparin-induced thrombocytopenia, which means the reflexive treatment for a clot can make this particular clot worse.

Read that paragraph again if you were told the vaccines had no serious side effects. Then read it again if you were told the side effects were covered up. It is the fact sheet the FDA authorized for distribution with the product, it uses the word causal, and it has been publicly downloadable the entire time.

These are not conspiracy theories. They are documented adverse drug reactions, listed in regulatory documents, with rates and case fatality attached. A regulatory system that suppressed vaccine injuries would not publish the case fatality rate of the injuries it was suppressing.

Young men are where the argument gets honest

The data out of England on myocarditis is the single most useful thing in this entire debate, and it is useful precisely because it refuses to give either side a clean win.

A self-controlled case series covered 42,842,345 people aged 13 and over who received at least 1 dose in England between December 1, 2020, and December 15, 2021. Myocarditis occurred in 2,861 of them, which is 0.007%, with 617 events falling in the 1 to 28 days after a vaccine dose. The design compares each person to himself, which removes the confounding that wrecks most vaccine-safety comparisons.

For the population as a whole, the answer was clear. The incidence rate ratio for myocarditis after a positive SARS-CoV-2 test was 11.14 before vaccination and 5.97 after, against ratios between 1.33 and 1.72 for the ChAdOx1 and Pfizer-BioNTech doses. Infection carried several times the myocarditis risk those injections did, which is the finding that got quoted everywhere.

The exception is the finding that got quoted almost nowhere.

The second dose of the Moderna vaccine produced an incidence rate ratio of 11.76, and the effect was strongest in men under 40. In that group the excess event count was 97 cases per million after a second Moderna dose, against 16 per million after a positive test. Six times higher from the shot than from the virus, in that specific group, at that specific dose, with that specific product. In women under 40 the same comparison was 7 per million against 8 per million, which is to say no meaningful difference at all.

That is what a real risk-benefit landscape looks like. Not a slogan in either direction.

Medicine does not run on the principle that vaccines are good and viruses are bad. It runs on the principle that a given intervention has a different expected value for an 82-year-old man with heart failure than for a healthy 18-year-old, and that the difference is quantitative and can be calculated. In 2021 those 2 people were frequently given the same advice with the same confidence. Anyone who lumps them together epidemiologically is making exactly the same mistake as the person who claims the vaccine was more dangerous than the disease for everyone. It is the same error carrying the opposite sign.

Did the shots save lives, and how would we know?

The famous number is 19.8 million, and I am not going to use it, for reasons I want to be explicit about.

The 2022 modeling study in The Lancet Infectious Diseases estimated that vaccination prevented 14.4 million COVID-19 deaths across 185 countries and territories in the first year of the campaign, between December 8, 2020, and December 8, 2021, when measured against officially reported COVID-19 deaths. When the same team used excess mortality instead as the true extent of the pandemic, the estimate rose to 19.8 million, a 63% reduction in total deaths over that year.

The operative word in both figures is estimate.

Nobody identified 19.8 million specific human beings who would otherwise have died. The result is the output of a transmission model fitted to reported and excess mortality data, and it inherits every assumption in that model, including assumptions about counterfactual transmission dynamics in the absence of any vaccination at all. A counterfactual is not a measurement of anything. Counterfactuals are arguments with equations attached, and the quality of the argument depends entirely on the plausibility of the assumptions, which reasonable epidemiologists dispute.

This is exactly why the 2025 analysis matters so much.

Ioannidis and colleagues used a substantially different approach, extended the window through October 1, 2024, and arrived at more than 2.5 million deaths averted in the main analysis, with sensitivity analyses spanning 1.4 to 4.0 million. 90% of those averted deaths were in people aged 60 and over. Children and adolescents accounted for 0.01% of lives saved, and adults aged 20 to 29 for 0.07%.

Note what that last pair of figures actually says, because it is the strongest evidence-based criticism of the vaccination campaign that exists, and it comes from a peer-reviewed journal rather than a forum post. The benefit was overwhelmingly concentrated in the old. In the young and healthy it was, by this analysis, close to nil. If you argued in 2021 that mass vaccination of healthy 22-year-olds was a poor use of a scarce resource, this study is your evidence, and you were more right than the people shouting at you.

One correction while we are here, because this specific number gets misquoted constantly and I have seen it misquoted in good faith by people making my own argument. The 82% figure in that paper does not refer to people over 60. It refers to the share of averted deaths occurring among those vaccinated before any infection. The over-60 share is 90% of lives and 76% of life-years. If you are going to cite the conservative study, cite it correctly; otherwise the other side gets to dismiss your best source on a technicality.

The gap between 2.5 million and 19.8 million is enormous, and pretending otherwise would be exactly the kind of tidiness I am arguing against. Part of that gap is not disagreement at all, and this is where most people quoting the range go wrong. The 2 endpoints do not measure the same quantity. The Lancet study covers 1 campaign year across 185 countries and takes excess mortality as the true toll, while Ioannidis runs through October 2024 against reported deaths with a different counterfactual underneath. Compared on the same basis, against reported deaths, the Lancet figure is 14.4 million rather than 19.8 million, which narrows the argument without closing it. What remains after that correction is genuine model uncertainty, and it is still large enough that anyone quoting either endpoint as settled fact is doing propaganda rather than epidemiology.

It also shows something else. 2 research groups with sharply different methods, different time windows, and openly different priors about vaccine effectiveness both landed on millions of lives saved. Neither produced evidence that the campaign killed more people than it saved. That is not the same as agreement. It is the far more useful thing, which is convergence from opposite directions.

“Died after the shot” does not establish a causal chain

This is the most common reasoning failure in the whole debate, and it is not a stupid one. It is an intuitive one, which is worse, because intuitive errors survive correction.

When 100 million people are vaccinated within a few months, millions of medical events necessarily follow. People have heart attacks, people have strokes, people receive cancer diagnoses that were years in the making, and people die of the ordinary things people die of. Some of that happens on the same day, some the next day, some a week later, and all of it would have happened at approximately the same rate if the injection had contained sterile saline.

This is background incidence, and it is not a subtle concept, but it is genuinely hard to feel. “My neighbor died 5 days after his shot” is a sentence about temporal proximity and nothing else, and the human brain converts temporal proximity into causation automatically, without asking permission. That conversion is why we have inferential statistics at all.

None of that means such reports should be waved away, because the correct response is the exact opposite.

Every one of them is a safety signal, and the correct response to a safety signal is to test it. You take the observed rate of the event in the vaccinated within a defined window, and you compare it against the rate in a comparable unvaccinated population, or against the expected background rate derived from prior years. If the observed rate exceeds the expected rate by more than chance allows, you have a real association and you go looking for a mechanism.

That is precisely how myocarditis was found. That is precisely how TTS was found. The system caught real adverse reactions, published them, put them in boxed warnings, and changed clinical recommendations because of them. That sits very badly with the theory that every vaccine complication was systematically concealed. A cover-up that publishes the thing it is covering up is a peculiar sort of cover-up.

I will grant the counter-argument its best form, because it deserves one. Passive reporting systems undercount, sometimes badly. Signal detection finds what it is looking for and can miss what it is not. Rare events with long latencies are genuinely hard to detect in this framework. All true, all serious, and none of it supports the leap from “detection is imperfect” to “the deaths are being hidden by the million.” Imperfect instruments are still instruments. A thermometer that reads half a degree low does not mean the room is on fire.

Why excess mortality is not a vaccine-death curve

The strongest-sounding argument I encounter goes like this: excess mortality continued after the vaccination campaigns started, therefore the vaccines caused it.

The premise is true, and excess mortality did in fact continue well past the start of the campaigns.

The inference is where the whole thing falls over, because excess mortality is by construction a measure without attribution. It tells you that more people died than expected. It tells you precisely nothing about why. That is its strength as a control measure and its weakness as an explanation, and it cannot be both a neutral instrument when you like the answer and a causal claim when you do not.

Consider what was happening simultaneously. SARS-CoV-2 continued to circulate in wave after wave. Influenza came back hard after seasons of suppression, into a population with reduced recent exposure. Heat waves killed thousands of elderly Europeans. Populations aged, as they do every year. Diagnoses postponed during the acute phase surfaced later as advanced disease. Health systems in several countries were still operating with degraded capacity and exhausted staff.

The Nordic data makes this concrete in a way that nothing else does, and it is my favorite dataset in this entire debate because it is so uncooperative with everyone.

A cause-specific analysis of Denmark, Finland, Norway, and Sweden for 2020 through 2022 found 32,491 COVID-19 deaths across the 4 countries. It also found 11,610 excess deaths from cardiovascular disease, most pronounced in Finland and Norway in 2022. And it found the opposite of excess as well, a genuine mortality deficit: 9,878 fewer deaths than expected from respiratory diseases other than COVID-19, and 8,721 fewer than expected from dementia, with the dementia deficit especially pronounced in Sweden in 2021 and 2022.

Sit with that last one for a moment. Fewer people than expected died of dementia. Not because dementia was cured. 2 mechanisms can produce that deficit and the data does not cleanly separate them: some deaths that would have been certified as dementia were instead certified as COVID-19, and some people who would have died of dementia later died of something else earlier, because the population most vulnerable to one was the population most vulnerable to the other. Mortality displacement is real, it moves in both directions, and it means the excess mortality curve is several processes laid on top of each other, running on different timescales.

Anyone who reads a single cause into that curve is committing the exact methodological error that vaccine critics correctly accuse health authorities of committing. Post hoc is not automatically propter hoc, and the rule does not become optional when the conclusion is one you like.

And Sweden

Sweden deserves a sober look, and it almost never gets one, because both camps discovered years ago that Sweden can be made to say whatever they need.

The comparison that actually holds up methodologically comes from the Nordic countries themselves, using national registers with a 2010 to 2019 reference period and age and sex standardization. Sweden had excess mortality in 2020, at 75 excess deaths per 100,000 population, with a prediction interval running from 29 to 122. Denmark, Finland, and Norway had theirs in 2022, at 52, 130, and 88 per 100,000, respectively. In those 3 countries mortality began rising in mid-2021 and stayed above expectation through 2022.

The interesting finding is not that one country won. The interesting finding is that the timing differed enormously between neighbors who resemble each other in almost every other respect. Sweden took its losses early, its neighbors took theirs late, and Finland’s 2022 figure of 130 per 100,000 is well above Sweden’s 2020 figure of 75. The study reports annual rather than cumulative national totals, so it does not support a statement about who ended up worse off overall, and I am not going to manufacture one.

That is scientifically fascinating, and it supports almost none of the conclusions people draw from it.

It does not permit “Sweden had fewer restrictions, therefore all restrictions were useless.” Timing differences of this kind have several plausible explanations, including differences in nursing-home structure, in household composition, in the seeding of the first wave, and in the immunity landscape each population carried into each subsequent variant. The study compares national trajectories, it does not test a policy, and its authors make no claim that it does.

Nor does anything in the German figures demonstrate that every German measure was necessary. A death count establishes that people died, and it establishes nothing at all about which interventions were worth their cost.

Pandemic countermeasures, pathogen virulence, and vaccine effectiveness are 3 separate questions with 3 separate evidence bases, and the habit of collapsing them into a single loyalty test is the reason this conversation has been useless for 6 years. You can hold, without any contradiction at all, that SARS-CoV-2 was dangerous, that the vaccines prevented a large number of severe cases and deaths mostly among the old, and that individual political measures were ineffective, disproportionate, poorly justified, or maintained long after their evidence base had evaporated. I hold all 3 of those positions at once, and there is no tension whatsoever between them.

Why the underlying-cause percentage fell, and what it tells you

Earlier I promised to come back to the declining share of American deaths where COVID-19 was the underlying rather than a contributing cause. From 91% in 2020 to roughly 66% in 2023 is a substantial drop, and it is the kind of trend that gets weaponized in both directions within about 20 minutes of anyone noticing it.

The naive reading is that certification got sloppier, or that hospitals kept coding COVID-19 for money after it stopped mattering. The opposite reading is that the virus never was the real cause and the statistics are finally admitting it.

Both readings assume that only 1 thing changed, and by 2023 almost everything had.

By 2023 nearly everyone had immunity from vaccination, from infection, or from both, most circulating variants descended from Omicron rather than from the ancestral strain, and effective antivirals existed. In that setting, a COVID-19 infection that kills a frail 89-year-old is genuinely more likely to kill him by tipping an existing heart failure over the edge than by drowning him in his own alveolar exudate the way the 2020 version did. The proportion shifted because the clinical picture shifted. A virus that has become one stressor among several in an immune population will show up more often as a contributing cause and less often as an underlying one, and that is not a scandal, that is exactly what you would predict.

Several things moved at once here: the virus, the immunity landscape, the availability of treatment, the collapse of routine testing, and the share of admissions where a positive result was genuinely incidental. The published data does not let me apportion the shift between them, and I am not going to pretend otherwise, because that is the same move I criticized in the section on the life-years. What the shifting percentage does not do is change the number of certificates.

What I am not saying, stated plainly so it cannot be misquoted

I have spent several thousand words arguing that a virus was dangerous and that vaccines prevented deaths, and in the current climate that gets a person sorted into a box within seconds. So let me nail the box shut from the inside.

None of this means governments were right. Many governments were spectacularly wrong, repeatedly, and several were wrong in ways that were obvious at the time to anyone reading the same papers they were reading.

None of this means pharmaceutical companies are altruistic organizations. They are corporations with quarterly obligations, they negotiated liability protections that would be unthinkable in any other industry, and the contracts under which several European states purchased these products remain partly redacted to this day. That is a real scandal and it has nothing to do with whether the products worked.

None of this means every dose administered was sensible. The evidence in this piece says the opposite. When 90% of the benefit accrues to people over 60 and 0.01% to children, a policy that treated all groups as equally urgent was not following the science, it was following an institutional need to appear decisive.

None of this means the political and social decisions of 2020 to 2022 should escape a hard, documented, uncomfortable reckoning. School closures, mobility tracking, the treatment of people who asked reasonable questions, the disgraceful handling of care home residents in several countries, the speed with which emergency powers became ordinary powers. All of that deserves an inquiry with subpoena power and a long memory.

Criticism only works if you are willing to let your own claims fail against data. That is the entire deal. If your position cannot be wrong, it is not a position, it is an identity, and identities do not update.

Skepticism that refuses correction has another name

Anyone who was skeptical in 2020 had excellent reasons. The information was contradictory, the authorities contradicted themselves in public, models were presented with a confidence their error bars did not support, and several official statements were later quietly walked back without anyone apologizing. Distrust was, in 2020, a rational response to an untrustworthy information environment. I do not say that grudgingly. I say it as someone who spent decades being paid to distrust confident statements and check the underlying material, which is a habit that does not switch off because a health minister is on television.

But anyone who still claims in 2026 that COVID-19 was essentially harmless, and that the vaccines killed more people than the virus did, no longer has an information problem.

He has an evidence problem.

The data has been public for years. The primary sources are free, the death records are downloadable, the FDA writes the word causal in its own boxed warnings, and the most conservative published estimate of vaccine benefit still lands at millions of lives and roughly 6 years of remaining life apiece. Every one of those things can be checked by anyone with a browser and an afternoon. The barrier is no longer access. The barrier is the cost of having been publicly certain.

That brings me to the tinfoil hat, and I want to be careful with it, because the joke is cheap and the point underneath is not.

The hat was never about physics. Nobody who ever wore one, literally or otherwise, believed aluminum foil blocked anything. The hat is a stance. It says: I am the one who is not being fooled, and everyone around me is. That stance felt like intelligence in March 2020, when the official story was changing weekly and being the person who noticed was genuinely valuable. For a while it was the right way to stand.

The trouble with a stance of that kind is that it has no exit condition. Being unfooled cannot be falsified. If the data supports you, you were right. If the data contradicts you, the data was made by the people fooling everyone else. There is no observation that could ever require you to take the hat off, which means the hat stopped being an instrument of skepticism the moment it became permanent.

So take the thing off your head. Not because the authorities have earned your trust, because they have not, and not because pharmaceutical companies deserve the benefit of the doubt, because they do not. Take it off because the whole point of skepticism was to follow the evidence, and the evidence has been sitting there in public for 6 years, waiting for someone to check the arithmetic.

Skepticism that cannot be corrected by data has stopped being skepticism, whatever it still calls itself.

It is faith, with worse sources.

References

• Ahmad, F. B., & Cisewski, J. A. (2026). Provisional COVID-19 mortality surveillance. National Center for Health Statistics. https://doi.org/10.15620/cdc/20250305007
• Centers for Disease Control and Prevention. (2025). Safety considerations for COVID-19 vaccines. https://www.cdc.gov/covid/hcp/vaccine-considerations/safety-considerations.html
• Centers for Disease Control and Prevention, National Center for Health Statistics. (2026). Provisional COVID-19 death counts by week ending date and state [Data set]. https://data.cdc.gov/dataset/Provisional-COVID-19-Death-Counts-by-Week-Ending-D/r8kw-7aab
• Chong, W. H., Saha, B. K., Murphy, D. J., & Chopra, A. (2022). Comparison of clinical characteristics and outcomes of COVID-19 patients undergoing early versus late intubation from initial hospital admission: A systematic review and meta-analysis. Respiratory Investigation, 60(3), 327–336. https://doi.org/10.1016/j.resinv.2022.02.007
• Conti, S., Ferrara, P., Mazzaglia, G., D’Orso, M. I., Ciampichini, R., Fornari, C., Madotto, F., Magoni, M., Sampietro, G., Silenzi, A., Sileo, C. V., Zucchi, A., Cesana, G., Manzoli, L., & Mantovani, L. G. (2020). Magnitude and time-course of excess mortality during COVID-19 outbreak: Population-based empirical evidence from highly impacted provinces in northern Italy. ERJ Open Research, 6(3), 00458-2020. https://doi.org/10.1183/23120541.00458-2020
• Forthun, I., Madsen, C., Emilsson, L., Nilsson, A., Kepp, K. P., Björk, J., Vollset, S. E., Lallukka, T., & Skrindo Knudsen, A. K. (2024). Excess mortality in Denmark, Finland, Norway and Sweden during the COVID-19 pandemic 2020-2022. European Journal of Public Health, 34(4), 737–743. https://doi.org/10.1093/eurpub/ckae091
• Grasselli, G., Greco, M., Zanella, A., Albano, G., Antonelli, M., Bellani, G., Bonanomi, E., Cabrini, L., Carlesso, E., Castelli, G., Cattaneo, S., Cereda, D., Colombo, S., Coluccello, A., Crescini, G., Forastieri Molinari, A., Foti, G., Fumagalli, R., Iotti, G. A., … Cecconi, M. (2020). Risk factors associated with mortality among patients with COVID-19 in intensive care units in Lombardy, Italy. JAMA Internal Medicine, 180(10), 1345–1355. https://doi.org/10.1001/jamainternmed.2020.3539
• Grasselli, G., Zangrillo, A., Zanella, A., Antonelli, M., Cabrini, L., Castelli, A., Cereda, D., Coluccello, A., Foti, G., Fumagalli, R., Iotti, G., Latronico, N., Lorini, L., Merler, S., Natalini, G., Piatti, A., Ranieri, M. V., Scandroglio, A. M., Storti, E., … Pesenti, A. (2020). Baseline characteristics and outcomes of 1591 patients infected with SARS-CoV-2 admitted to ICUs of the Lombardy region, Italy. JAMA, 323(16), 1574–1581. https://doi.org/10.1001/jama.2020.5394
• Holmes, A., Emerson, L., Irving, L. B., Tippett, E., Pullin, J. M., Young, J., Watters, D. A., & Hamilton, A. (2024). Persistent symptoms after COVID-19: An Australian stratified random health survey on long COVID. The Medical Journal of Australia, 221(Suppl 9), S12–S17. https://doi.org/10.5694/mja2.52473
• Ioannidis, J. P. A., Pezzullo, A. M., Cristiano, A., & Boccia, S. (2025). Global estimates of lives and life-years saved by COVID-19 vaccination during 2020-2024. JAMA Health Forum, 6(7), e252223. https://doi.org/10.1001/jamahealthforum.2025.2223
• New York City Department of Health and Mental Hygiene COVID-19 Response Team. (2020). Preliminary estimate of excess mortality during the COVID-19 outbreak, New York City, March 11-May 2, 2020. MMWR Morbidity and Mortality Weekly Report, 69(19), 603–605. https://doi.org/10.15585/mmwr.mm6919e5
• Nilsson, A., Emilsson, L., Kepp, K. P., Knudsen, A. K. S., Forthun, I., Madsen, C., Björk, J., & Lallukka, T. (2024). Cause-specific excess mortality in Denmark, Finland, Norway, and Sweden during the COVID-19 pandemic 2020-2022: A study using nationwide population data. European Journal of Epidemiology, 39(9), 1037–1050. https://doi.org/10.1007/s10654-024-01154-0
• O’Driscoll, M., Ribeiro Dos Santos, G., Wang, L., Cummings, D. A. T., Azman, A. S., Paireau, J., Fontanet, A., Cauchemez, S., & Salje, H. (2021). Age-specific mortality and immunity patterns of SARS-CoV-2. Nature, 590(7844), 140–145. https://doi.org/10.1038/s41586-020-2918-0
• Patone, M., Mei, X. W., Handunnetthi, L., Dixon, S., Zaccardi, F., Shankar-Hari, M., Watkinson, P., Khunti, K., Harnden, A., Coupland, C. A. C., Channon, K. M., Mills, N. L., Sheikh, A., & Hippisley-Cox, J. (2022). Risk of myocarditis after sequential doses of COVID-19 vaccine and SARS-CoV-2 infection by age and sex. Circulation, 146(10), 743–754. https://doi.org/10.1161/CIRCULATIONAHA.122.059970
• Robert Koch-Institut. (2026). COVID-19-Todesfälle in Deutschland [COVID-19 deaths in Germany] [Data set]. https://github.com/robert-koch-institut/COVID-19-Todesfaelle_in_Deutschland
• U.S. Food and Drug Administration. (n.d.). Fact sheet for healthcare providers administering vaccine: Janssen COVID-19 vaccine. https://www.fda.gov/media/146304/download
• Watson, O. J., Barnsley, G., Toor, J., Hogan, A. B., Winskill, P., & Ghani, A. C. (2022). Global impact of the first year of COVID-19 vaccination: A mathematical modelling study. The Lancet Infectious Diseases, 22(9), 1293–1302. https://doi.org/10.1016/S1473-3099(22)00320-6
• World Health Organization. (2026a, May 13). Global health gains face threat of reversal. https://www.who.int/news/item/13-05-2026-global-health-gains-face-threat-of-reversal
• World Health Organization. (2026b). World health statistics 2026: A visual summary. https://www.who.int/data/stories/world-health-statistics-2026–a-visual-summary

Disclaimer: This article reflects the author’s own research and opinion as of the publication date shown above; later findings or legal changes may have overtaken it, so always check that date. Sources are cited for independent verification, and no liability is accepted for third-party studies. This is general information, not medical, legal, or professional advice: for medical questions see a doctor, for legal questions a lawyer, and in an acute crisis contact emergency services or a crisis helpline.

Read the full disclaimer

About the header image: it is AI-generated. Cheaper than a photo shoot, and I have made my peace with the age of AI. Everything inside the article is real, the diagrams, the skulls, the findings, and every word. The machine gets the opening shot and not one inch past it.