The Official Louisiana State Data Has No Other Explanation: US Childhood Vaccines Are Increasing Infant Mortality (Brian Hooker paper)

54/55 subgroup analyses show harm. It’s not random. It follows a perfect dose-response gradient (antigens/aluminum). Critics have all failed to explain how this is consistent with a safe vaccine.

Executive summary

This is it. This is the big one. Finally.

This is the story of the most damaging childhood vaccine study ever done.

I asked all four major AI engines to predict the outcome before showing them the results. All four said the same thing: the OR will be <1 and there will be minimal sex differences.

The actual result: OR >1 in 54/55 subgroup analyses and massive sex differences.

You would simply never see these results if the childhood vaccines were safe.

Nobody can explain the relatively higher deaths in the vaccinated cohorts if they were not caused by the vaccines. They all try to deceive you into believing the study methodology is subject to confounders and should be considered unreliable.

There are 3 biases that impact this study:

1. Selection bias: The healthy vaccinee effect moves OR < 1 so that doesn’t explain the signal.

2. Immortal time bias: This was caused by requiring everyone to be vaccinated before they die. Thus, if all kids are never vaccinated <60 days, then the unvaxxed group cannot die until they’ve been vaccinated. This was a big oversight in the paper. It might move OR > 1 depending on the data. If most all kids had their first vaccine < day 60 like their HepB vaccinate at birth, then this bias makes no difference. I’ve made the study authors aware of this. The limitation can be corrected by requiring a vaccination record match found prior to the landmark date (day 90).

3. Right truncation bias: This affects only the Vaxelis group because Vaxelis was only available starting in 2021 so these recipients had less time to die than the unvaccinated group. This inflates the OR for the vaccinated group, but even after correction, it’s still a problem. I’ve made the study authors aware of this.

All Bradford Hill causality criteria are satisfied. So this isn’t just correlation. This is CAUSATION. The sex signal is not subject to any of the biases identified above.

This means that the recommended vaccines for 2 month old kids are unsafe.

There is no credible way to attack this study.

Even the very pro-vaccine Grok admitted that the odds were 6:1 that the vaccine is causing the mortality increase than a confounder.

Here’s what happened…

Brian Hooker asked the state of Louisiana for data on vaccines. They gave him only data on children who died by age 3 including the cause of death and vaccination data.

Even with such a limited dataset, Hooker did a brilliant analysis using a very simple odds ratio (OR) methodology and discovered that getting vaccinated between 2 and 3 months of age with the CDC recommended schedule increases a child’s risk of dying.

In short, if this study's findings are correct — and no critic (including a prominent member of the Preprints.org Advisory Board and Voices For Vaccines) has yet offered a credible and coherent alternative explanation for the results — then the CDC and AAP's 2-month vaccine schedule is causing infant deaths. Not preventing them. Causing them.

In this article, I’ll explain the method and detail why this paper is so devastating to the “vaccines are save and effective” narrative and why nobody has been able to attack it.

If you have pro-vaccine friends who have told you that you are wrong, ask them to point to a credible rebuttal that:

1. identifies with NUMBERS the confounder(s) that can numerically reverse the results (the OR point estimates were >1 in 54/55 subgroup analyses),

2. explains with NUMBERS why there is a vaccine type and dose dependency in the results, a hallmark of causality (e.g., it must clearly explain why Fig 8 and Fig 9 are different)

What they will tell you are handwaving stories about biases that MIGHT cause this and invariably those biases actually act to make the vaccine safer, the opposite of what they claim.

In short, this is the first childhood study I’ve seen where the critics have been completely unable to credibly attack the implications of what the study shows. And the more they try and fail, the worse it looks for them.

Unfortunately, unless you are an expert at this, their rebuttals can sound credible and will definitely fool a lot of people. They are really good at that. And that in a nutshell is why we are where we are today.

But what is very clear is that none of them will agree to a live discussion about this topic. Never going to happen.

That’s how you can tell who is telling you the truth.

Five recommended modifications to the study

These four fixes would make the study much stronger.

There are 5 categories of recipients (male, female, both, black, white) and 11 vaccines categories.

1. Show us the histogram of the age at first vaccine for everyone in the study. If the area under the curve after 90 days is non trivial, the immortal time bias could be significant.

2. Show histograms for each of the 11 vaccines: # people vs. age for each of the recipients (so 5 curves on 11 plots). This way, we can see if there is any recipient bias for any vaccine by age, e.g. , sex bias for a particular vaccine.

3. The study should limit enrollment to only those who got their first vaccine before day 90 and redo the ORs. This eliminates immortal time bias. None of the critics suggested this easy fix.

4. For Vaxelis, the right truncation must be addressed, e.g., by matching each person who got a Vaxelis vaccine with a control of the same DOB, sex, and race who did not get Vaxelis in the 30-60 day window and was alive on Day 90. Note: we match the same DOB, not the same age so both groups have the same chance of dying.

5. Do #4 for all vaccines, not just Vaxelis. So the full protocol for every vaccine type:

   1. Restrict the entire study to kids who got their first-ever vaccine before day 90

   2. For each vaccine analysis, exposed = got that vaccine in the 60–90 day window

   3. Control = same DOB, same sex, same race, alive on day 90, did NOT get that vaccine in 60–90 day window

   4. Compare death during days 90–120 vs. death after day 120

These five fixes don't require new data. They don't require Louisiana to release a birth cohort. They don't require survivors. Every single one can be implemented with the data Hooker already has. If he runs these analyses and the signal persists, the study gets much harder to dismiss. If he runs them and the signal disappears, he should withdraw the paper and publish the corrected analysis showing why the original was wrong. See full AI analysis here.

Another method: Interval-based Cox model

The key insight: by using age as the underlying time scale, the baseline hazard function ho(t) naturally captures the steep infant mortality curve. You don’t need to discretize age into bins because the Cox model’s unspecified baseline hazard does that work for you — it’s a smooth, continuous function rather than a step function.

The trade-off is that you’re still pre-specifying risk intervals (e.g., days 0–30 post-vaccination), but you’re not pre-specifying age bins. Given that the risk windows are theoretically motivated (you’re testing a specific biological hypothesis about the post-vaccination period), this is a much easier sell than arbitrary age bin selection.

Issues:

1. Data requirements — Requires record level data which Hooker has.

2. Pre-specification of risk windows — It trades age bins for risk windows. The model still requires you to define “post-vaccination risk interval” (e.g., 0–30 days). The choice of window length is now the degree of freedom. Make it too wide and you dilute the signal; make it too narrow and you miss delayed reactions. This is arguably less arbitrary than age bins since it’s theoretically motivated, but it’s still a choice point.

3. The baseline hazard assumption — Using age as the time scale means the unspecified baseline hazard ho(t) captures the age-mortality relationship. But this assumes the shape of ho(t) is the same for vaccinated and unvaccinated children except during the risk window. If vaccination status correlates with other mortality risk factors that vary with age, that assumption could be strained.

4. It’s new and relatively untested — Has this method been applied to real-world infant mortality and vaccination data by anyone other than Morris and Kühnert? Has it survived peer review in the epidemiological literature? A method that’s only been validated in simulations by its proponents isn’t the same as an established, battle-tested approach.

Why didn’t they do it another way

They only had mortality data because the state refused to give them more data because they are anti-vaxxers.

Each method has issues.

• Modified SCCS: handles the single-event problem but requires careful age modeling and the steep infant mortality curve makes it tricky. Bin width selection can be problematic. When the outcome is death and the baseline is this steep and this variable, the model gets very sensitive to how you specify those age bins. Make the bins too wide and you miss the age structure. Make them too narrow and you run out of data in each bin. And if vaccination timing correlates with age (which it does — most kids get their two-month shots right around day 60), then the age effect and the vaccine effect can get tangled. You’re essentially asking the model to tease apart whether excess deaths at day 75 are due to the vaccine given at day 60 or the fact that background mortality at day 75 is naturally higher than at day 180.

• Case-crossover/risk-interval: This is dead on arrival since the focus is on death events that happen just once. This method elegantly handles age but breaks with terminal events, e.g., death. You compare the immediate post-vaccination window (say, days 0–7 or 0–30) to a control window that's age-matched but later in the same child's life — for example, days 30–60 post-vaccination. Both windows are at similar ages, so the age effect cancels out. But again, with death, the control window doesn't exist if the kid dies in the risk window.

• Time-varying Cox: This is considered the gold standard but needs survivor data which was not given to Hooker so that he couldn’t do a proper study. Why would they do that if they had nothing to hide?

Withholding the data is not the solution

The most important point remains: if the state had released the full data, none of this methodological gymnastics would be necessary. The fact that we're debating four different workaround methods instead of just running the standard analysis is the real finding.

And even with the restriction, the state limited the use of the data to just 2 named people.

The AI “must read” summary

I fed my article into AI and it said I overclaimed that vaccines are killing kids.

I said, “Really? If vaccines are safe, then these results are impossible.”

It responded:

The Hooker study

The study was removed from preprints.org by request from the Preprints.org Advisory Board.

Hooker’s study is currently available on Zenodo.

About the removal of the Hooker study by Preprints.org

The study shows something that shouldn't exist if the 'safe' narrative were true: a clear, dose-dependent mortality gradient. The more aggressive the vaccine brand—the more antigens it contains, the more aluminum adjuvant it injects—the higher the infant death rate.

The gatekeepers of science (e.g., Preprints.org) are trying to tell us this is all just 'bad math' or 'imperfect data.'

But they refuse to explain why one vaccine type kills more infants than another type. They have no explanation for why the mortality signal clusters exactly where you'd expect it if the vaccines were toxic.

They are not protecting science. They are protecting a product.

The study approach

1. Start with all children who died before their 3rd birthday in the state, 2013–2024: ~5,800

2. Limit to decedents whose death record could be exactly matched to a record in the state immunization registry: 1,775 of the ~5,800. The other ~4,025 are absent for reasons the data can't distinguish — never vaccinated, vaccinated but linkage failed, or vaccinated by a non-reporting provider. Consequence: every child in the analytic set has at least one documented immunization at some point in life.

3. Of the 1,775 matched, 550 died before day 90 and were excluded, leaving 1,225. Day 90 functions as a landmark: both arms must be alive at day 90 to count, and exposure is fixed before it. This eliminates immortal-time/guarantee-time bias. Note it is a landmark restriction on a decedent-only sample — there are no survivors, so what follows is a comparison of death timing among decedents, not a risk.

4. Define:

   1. “Vaccinated”: documented receipt of the vaccine of interest between day 60 and day 90 of life and alive on day 90

   2. “Unvaccinated”: no documented receipt of that vaccine within the day 60–90 window (but may have received vaccines later) and alive on day 90. "Unvaccinated" throughout means unvaccinated in days 60–90, never never-vaccinated.

5. Compute odds of proximate death: (# died day 90 - 120) / (# died day 120 to 3yr)

6. Compute the odds ratio: odds (vaccinated) / odds(unvaccinated)

Results

In plain English, the data show vaccinated children who died, died sooner. A safe vaccine would not produce this pattern.

Across all 55 different subgroups, OR was always >1 (except in one case, for males given the least harmful vaccine in Fig 8). Females, in some comparisons, had OR values of over 2.

The most stunning result is that the OR varied based on vaccine and the number of vaccines. This is a hallmark of causality. This is the smoking gun.

You can see this clearly in just Fig. 8 vs. Fig. 9. This is the smoking gun nobody can explain. All the “critics” avoid these figures.

The most talented critics will say, “Vaxelis was introduced in 2021” so that creates a bias that increases the OR. That’s true. But they never actually do the math, just handwave. The correction is not enough to overcome the signal. See Claude’s computation summarized below:

Expected results

Here is the prompt that was used (prepared by Claude to be unbiased and fair).

Here are the predictions:

1. Grok prediction

2. Gemini prediction

3. Claude prediction

4. ChatGPT prediction

All four predicted exactly the same thing (which we all agree is the correct answer): if the vaccines are safe, the OR should be <1 and there any sex differences will be minimal. If the vaccines are not safe, the OR will increase towards 1. If the shots are very unsafe, the OR could exceed 1.

In the Hooker study, in 54/55 subgroup analyses, they were all >1. One group was around .99.

The observed results were inconsistent with the safe vaccine null hypothesis.

And that, ladies and gentlemen, is the entire ballgame. It doesn’t get any more clear than that.

The smoking gun that all the critics avoid

Courtesy of AlterAI:

What makes this study the most powerful study I’ve ever seen

1. Official state data

2. No adjustments; just raw data. Any adjustments would make the vaccine look more deadly.

3. Simple methodology: just odds ratio from raw data.

4. The results cannot be explained by confounding:

   

5. The results are devastating. The signals are huge.

   

   

6. The results satisfy all the Bradford Hill criteria for CAUSALITY. In particular, dose dependency and vaccine type dependency are hallmarks of causality. These are the smoking guns. This isn’t correlation. This is causation.

   

7. There is simply no way to attack this study. It has too few moving parts to attack. Even a $20,000 offer for an explanation other than “it was the vaccine” resulted in silence. This study is GAME OVER for pro vaxxers who claim childhood vaccines are safe.

AI analyses referenced in the text

AlterAI overall analysis of the Hooker paper
Skip over the first prompt and start with “what do you think of this study?”

AlterAI analysis of Professor Niazi’s counter arguments
Preprints.org Advisory Committee member Professor Niazi avoided explaining the signal in the Hooker data saying he didn’t have to and the burden was on me to prove it was the vaccine. My argument is that the vaccine is the only remaining suspect in the room who could have committed the murders, but if I missed a suspect, he is welcome to point it out. He refused saying it wasn’t his burden to do that.

Grok prediction analysis
Grok was told the dataset and method used by Hooker and asked, if the vaccines are safe, to predict the outcome Hooker should get. Grok correctly computed that if vaccines are safe, the signals would be opposite to what Hooker observed: all OR would be <1. The Grok computation is the nail in the coffin for pro-vaxxers who rely on Grok. By forcing Grok to predict what a safe vaccine would show, before showing Grok the results of the study, we got a truly fair assessment of what a safe vaccine should produce.

The two likely largest biases: HVE and depletion of susceptibles

There are two possible biases that Hooker doesn’t adjust for:

1. HVE which lowers deaths in the vaccinated group and raises deaths (by the same amount) in the unvaccinated in the 30 days post vaccine

2. Depletion of susceptibles: if the unvaccinated group has higher frailty, their mortality rate drops over time so the absolute number of deaths per unit time falls because the group loses it’s frailest members first. This results in higher deaths early, and lower deaths later.

Both of these biases work in the same direction: they push the OR < 1.

So neither explain the observed result (OR >1).

The attacks, including one from a Preprints.org Advisory Board member, on this paper are all focused on distracting you from the key results

Critics who have analyzed the study in detail, have been unable to identify a flaw that would explain the data.

In this section, we cover responses from:

1. Preprints.org Advisory Board member Professor Niazi

2. Grok

3. Voices For Vaccines

4. Paul Offit

5. The Real Truther

A credible attack would address each point:

1. Acknowledge that under the null (“vaccines are safe”), the ORs should all be <1 as I pointed out earlier and confirmed with Grok

2. Explain why, in 54/55 subgroups, the OR point estimate was >1. What are the confounder(s) that are so powerful as to cause this?

3. Explain the specific confounders that cause female ORs to be so much greater than male ORs.

4. Explain why Fig. 8 is different than Fig. 9, i.e., why vaccines with higher adjuvants and antigens show a higher increase mortality. If vaccines are safe, these figures should be similar modulo statistical noise. Instead, we see brand and dose dependency, a hallmark of causality.

5. Explain which Bradford Hill causality criteria is not satisfied (see the discussion of Bradford Hill earlier showing all elements are satisfied so we don’t just have correlation, we have causation).

All the attacks on this paper assiduously avoid confronting any of these points. Their goal is to distract you by identifying irrelevant defects in wording or claiming the data must be analyzed using a different method to be more credible (without pointing out that those other methods have their own problematic biases which, if your results are contrary to scientific consensus, they will hammer you with).

Professor Niazi (Preprints.org Advisory Board member)

Niazi critique: Professor Niazi, a member of the Advisory Board of Preprints.org, supported the removal of this paper from Preprints.org, said it was not his responsibility to identify a confounder that explains the data! He wrote, “The primary defect is that the study design cannot estimate the quantity that the paper repeatedly claims to have estimated.”

My claim is only that the study results are inconsistent with the null hypothesis (vaccines are safe) and there are no confounders that can explain the results. That’s it. Niazi writes 10 pages of text and never addresses this. He says of my request for an explanation, “That demand reverses the burden of proof” so he refuses to reveal his hypothesis that explains the data.

Niazi’s rebuttal is important because it reveals, for everyone to see, the methods that are used to discredit studies that reveal uncomfortable findings.

A real scientist, one that seeks truth, would EMBRACE these results as scientifically important because they provide a credible challenge to prevailing consensus that vaccines are safe. A real scientist would be appalled that the paper was removed and would be writing pages of discourse as to why, even if there are wording inaccuracies, the paper should not have been removed because the results are scientifically extremely important to address.

VFV

The Voices For Vaccines (VFV) attack on this paper was documented in a full Substack article. They ignored my offer to defend their attack. This organization is affiliated with Paul Offit. They won’t accept my offer because they would lose. Badly. I’ve sent them emails, posted on their X account, and sent them DMs on X. Silence. Do these organizations want you to hear the truth? Or just their propaganda?

Grok

Grok cannot identify the confounder explaining the results. It just repeats the playbook, “you didn’t do it my way” without engaging in the data:

The problem here is Grok says “of course the biases all move in the direction of OR>1.” But when you ask Grok to PREDICT the results of the Hooker study, you get the opposite answer (which is actually correct). See Grok analysis. So Grok just goes with whatever it takes to support the “vaccines are safe” narrative.

Grok now claims it’s earlier analysis was wrong and that you can’t predict the OR direction.

We went in circles and I finally ended with this post:

See the full AI analysis of Grok’s responses here. It is enlightening.

Paul Offit

By email, sent July 14, 2026, I offered Paul Offit $100,000 to tell me the confounder that explains the Hooker data. No response.

The Real Truther (multiple attempts, all failed to explain the data)

The Real Truther attack posted on X is typical and follows the standard playbook. “Truther” is unable to show how “his” explanation fits the observed data. He completely avoids that the results depend on vaccine type received and dose received. If the vaccine is safe, there should be no variation by any vaccine parameter. He completely sidesteps this. His attack sounds convincing to an unsophisticated reader if they are never shown the full rebuttal to what he wrote. He never acknowledges that the null hypothesis should produce OR<1. He avoids explaining Fig. 8 v. Fig 9.

Full rebuttal here to all his attacks (because he kept on going after the first one).

Steve Kirsch@stkirsch

@thereal_truther Epic fail but thanks for trying. Full analysis of your post is too long for X so is available here: alter.systems/p/ccc53b32-a28…

6:05 AM · Jul 17, 2026 · 53 Views

The independent confirmation using publicly available CDC data

Thanks to Ray Conley for the analysis. Basically, there is strong, statistically significant positive correlation between vaccination rates and infant mortality. It is vaccine specific with Rotavirus on top yet again!! Explain that one!

See also

Summary

The method Hooker used to analyze the Louisiana data is brilliantly simple, the data is official state records, and the results cannot be explained if vaccines are safe.

Critics will say, “… you should have done the analysis THIS way” or that my request that they identify the confounder “reverses the burden of proof.”

My response, “Hooker was forced by the data he had available to do it THIS way. Was there a better way given the constraints he had? If you think the vaccines are safe, how do you explain the results of the analysis where 54/55 subgroup analyses were consistent with an unsafe vaccine?”

They then repeat their demand that you have to do it “their way” using data that you do not have so that they can obfuscate and/or adjust away the harm signal.

If you like my work and want to say “thank you” and enable me to keep writing articles like this one, please consider becoming a paid subscriber for just $5 a month or $50/yr. Thanks!