New French study claims vaccines reduced COVID deaths by 75% and were associated with a 25% reduction in non-COVID ACM. Here are the flaws.

Even exact 1:1 matching cannot undo the healthy vaccinee effect (HVE). This study used MUCH cruder matching and claimed a benefit. Nah, that is wishful thinking.

Executive summary

A French COVID vaccine study published in JAMA which could not do 1:1 matching (prohibited by French privacy laws) had to estimate to match the cohorts. They found that after matching, if you were vaccinated you had a 25% lower chance of non-COVID death. Which means their matching was flawed. Which means the comparison is flawed as well.

The tweet claiming victory:

The excess mortality trends in France show the mortality “benefits” were non-existent!

If the shots worked, there would be a knee in the excess mortality trendline. I don’t see a knee. Do you see a knee?

AlterAI exposes the flaws

Here is the AlterAI analysis exposing the flaws. Great read!

This shows the study was flawed.

Precision 1:1 matching cannot adjust for the healthy vaccinee effect

Papers from Qatar and Denmark both show epidemiology cannot adjust for HVE even using the most exact 1:1 matching ever done in history.

The authors of the JAMA study didn’t have anything close to the data in the Qatar and Denmark papers. And even if they did, they showed the current epidemiological methods are too imprecise to make any determinations of harm or benefit.

The French study, which was forbidden by French privacy law from doing 1:1 matching, claims they matched the cohorts and you should believe their results that it reduced COVID deaths 75%, even in the face of rising excess mortality. Are you serious?

Great matching (Denmark and Qatar) —> pointless.

Rough matching (France) —> more than pointless.

It’s even worse than first glance

Nic Hulscher, in a private e-mail, pointed out the following:

The Semenzato study uses several methodological tricks that manufacture an artificial survival advantage for vaccinated individuals: it excludes all vaccine deaths occurring in the first 6 months by starting follow-up only after a mandatory 6-month “grace period,” meaning vaccinated people must survive 6 months to enter the analysis while unvaccinated deaths during that same period count; it assigns fake index dates to unvaccinated people, so if they die soon after their randomly assigned date, it inflates unvaccinated mortality; it censors unvaccinated individuals the moment they get a shot, removing healthier, higher-engagement individuals from the unvaccinated pool and leaving behind a disproportionately sicker group; it relies on inverse probability weighting that cannot correct for massive behavioral and socioeconomic differences (alcohol, tobacco, deprivation, chronic illness—all far higher in the unvaccinated); and it uses biased “negative control outcomes” (trauma, accidents) that differ by behavior and therefore cannot detect confounding. Buried results reveal that the apparent “benefit” rapidly shrinks over time (HR 0.61 early → ~0.79 by 42 months), and the strongest “protective effect” appears illogically in 18–29-year-olds who had virtually no COVID-19 mortality—strong evidence of healthy-vaccinee bias, not biology. The study also lacks cause-of-death data for the final 15 months.

In short, by excluding early deaths, fabricating index dates, censoring selectively, and comparing two fundamentally different populations, the study structurally guarantees the conclusion of “no increased long-term mortality,” regardless of the real effects of mRNA vaccination.

Summary

The matching was flawed. The impossible NCACM benefit proves it. The mortality watch excess mortality graph proves it too. The AlterAI analysis shows the flaws.

But people just want to believe what they want to believe, don’t they?

The better method: KCOR. But nobody wants to talk about it because the results show they were all wrong.