Designed to Fail: Brand New Hailed Study Exonerating Aluminum Has a Fatal Flaw

A study designed not to find harm will never find harm. It will, however, reliably find publication. Legacy medical media outlets dutifully oblige. STAT News uncritically lauds the study.

“The problem would have been obvious if they had excluded sick children who potential had pre-24 month related chronic illness. Instead, they statistically ‘adjusted for’ being chronically ill before 24 months of age and this hides the risk.”

Headlines were made 2022 a study that found increased risk of asthma following exposure to higher amounts of aluminum via vaccines in the first two years of life. Today, an already widely publicized new study—published in Annals of Internal Medicine under the title “Aluminum-Adsorbed Vaccines and Chronic Diseases in Childhood”—purports to settle one of the most important questions in modern toxicology: Is injected aluminum in childhood vaccines associated with increased risk of chronic illness? The authors tracked over 1.2 million Danish children born between 1997 and 2018 and claimed to find no increased risk of 50 health outcomes—including autism, ADHD, autoimmune diseases, and allergic conditions—based on cumulative aluminum exposure by age two.

This conclusion has already been dutifully parroted in headlines on legacy media (Yahoo News, NBC News) and corporate obedient medial news outlets (Gene Online, Infection Control Today, CIDRAP). Medpage today’s headline blares “Aluminum in Vaccines Not Culprit in Kids' Chronic Diseases, Study Shows”. Even Stat News has accepted the study uncritically, citing experts who “welcome” and “feel a sense of gratitude” for the study:

“I was very encouraged by the data, particularly for asthma as I know vaccine hesitancy groups are concerned by this,” said Anna Durbin, director of the Center for Immunization Research at Johns Hopkins Bloomberg School of Public Health. Matthew Daley and Jason Glanz, two authors of an earlier U.S.-based study that suggested aluminum in childhood vaccines might be tied to an increased risk of developing asthma before age 5, also welcomed the publication. Daley and Glanz, both of whom work at Kaiser Permanente Colorado’s Institute for Health Research, both described the study as well done. On the issue of asthma risk in particular, Daley said this new study should provide reassurance to parents that vaccinating their children according to the recommended schedule should not increase their risk of developing asthma.”

The study’s conclusion is not just premature. It is, at its foundation, scientifically and ethically indefensible. The study is compromised not by a single flaw, but by an ensemble of methodological maneuvers—each subtle, plausible-sounding, and common in pharmacoepidemiology—that, when combined, render the entire exercise a façade of scientific inquiry.

The Core Distortion: Conditioning on Consequences of Exposure

The most egregious violation lies in the study’s statistical modeling: the authors adjusted their hazard ratios for the number of general practitioner (GP) visits before age two. This variable is not a confounder. It is a direct downstream consequence of vaccine exposure itself. Infants receiving multiple aluminum-containing vaccines in their first year may present with fever, seizures, inconsolable crying, rashes, hypotonia, gastrointestinal distress, or developmental regression. These symptoms drive parental concern, which in turn drives GP visits.

To adjust for GP visits in the same window as vaccine exposure is to adjust for an intermediate—an effect of the exposure.

Adjusting for having aluminum-related chronic illness before 24 months is an obvious statistical trick that would make the association between exposure to aluminum-containing vaccines and chronic illness after 24 months disappear. Imagine studying the association between smoking and lung disease adjusting for having nicotine-stained fingers, carrying a lighter, or frequency of coughing in the 24 months before lung cancer diagnosis.

What you are doing, in effect, is mathematically erasing the very signal you are supposed to be detecting. Stat News never critically examined the design of analysis.

From the study:

“Hazard ratios were adjusted for sex, calendar year and season of birth, preterm birth, birthweight, number of visits to the general practitioner before age 2 y, maternal age at delivery, maternal place of birth, maternal smoking during pregnancy, maternal health conditions, maternal prescription drug use during pregnancy, and parental household income.”

Kids under 24 months of age receiving more aluminum-containing vaccine on their way to being sick due to aluminum are effectively removed by this statistical maneuver. The study authors do not provide any justification for adjusting for these particular variables and do not show the results of alternative models. They cherry-picked the model that gave the result they desired.

Aluminum Exposure in Danish Infants Before Age Two: A Quantified Reality

To appreciate the implications of this study, one must understand what Danish infants are actually receiving before age two in terms of aluminum-based vaccine adjuvants. According to the official Danish childhood vaccination schedule published by the Statens Serum Institut, infants are administered two primary aluminum-containing vaccine formulations at 3, 5, and 12 months of age: the DTaP-IPV/Hib combination vaccine and the pneumococcal conjugate vaccine (PCV).

Each dose of the DTaP-IPV/Hib combination contains between 330 and 625 micrograms of aluminum, depending on the formulation. PCV contributes an additional 125 micrograms per dose. With three doses of each given in the first year of life, a typically vaccinated infant receives:

• DTaP-IPV/Hib: 3 doses × ~500 μg = 1,500 micrograms (μg)

• PCV: 3 doses × 125 μg = 375 micrograms (μg)

That amounts to a cumulative aluminum exposure of approximately 1,875 micrograms by 12 months—nearing or exceeding 2,000 micrograms (2 milligrams) by 24 months depending on any additional doses or formulations used.

This is not incidental trace exposure. It is a scheduled, cumulative parenteral administration of neuroactive, immunostimulatory aluminum salts—injected during the most vulnerable period of neurodevelopment, before the blood-brain barrier is fully mature, and long before renal clearance is robust.

Aluminum exposure in the first two years of life is treated by the authors not as a biological event to be measured or modeled with care, but merely as a numerical input into a hazard ratio algorithm that, via improper adjustment, systematically suppresses signal. The actual quantity—whether in milligrams or micrograms—is never meaningfully discussed in the study. There is no acknowledgment of dose accumulation, no exploration of individual variability in retention, no recognition of how this burden compares to dietary or environmental exposure.

The maneuver to adjust based on the number of doctor’s office visits is not neutral. It is not "conservative." It is a structural mechanism of deception. They may as well adjusted for the micrograms of aluminum received in the first two years of life and concluded that the removal of the association with chronic illness after adjusting for aluminum exposure exonerates aluminum.

The Absurdity of This Adjustment

For the GP visit adjustment to be valid, the number of visits in Year 1 and 2 would have to be statistically independent of aluminum exposure. Simple math would say the more often a child visits a doctor, the more often they are likely to be vaccinated. But it’s also true that if aluminum is negatively related to health outcome, they have to go to the doctor’s office more office due to health issues related to aluminum injection.

In order to control for a variable, that variable must not be influenced by the exposure. When that parity fails, adjustment introduces collider bias, a statistical distortion that misrepresents the relationship between exposure and outcome. In plain terms, the study design effectively deletes from analysis the children most likely to show early signs of aluminum-related harm.

This alone would be sufficient to invalidate the study. But the Annals of Internal Medicine appears to have missed the boat on sufficient and detailed review. This one fatal flaw is bad enough, but it is only the beginning.

Assumed Knowledge: A Study that Depends on What the Public Doesn’t Know

This study assumes the reader possesses fluency in biostatistics and causal inference, when in fact its apparent rigor depends on the public not understanding any of it.

Most readers do not realize that registry-based diagnoses are often delayed, incomplete, and highly susceptible to surveillance bias. Many chronic conditions—particularly those of immune or neurodevelopmental origin—manifest subtly and accumulate over time. By defining disease onset based solely on date of diagnosis, the study obscures the fact that many children may have exhibited symptoms long before a code was entered into the registry.

The study also assumes readers will accept that aluminum exposure was accurately quantified. In reality, the authors did not use lot-level data, individual medical records, or aluminum biomarkers. They assigned cumulative dose based on policy, assuming that all children followed the schedule and received vaccines of standard aluminum content. This gross misclassification further biases results toward null and suppresses any possibility of detecting a true effect.

Misleading Certainty, Implausible Results

As a result of these structural manipulations, the study not only fails to find increased risk—it finds decreased risk. Higher aluminum exposure is associated with lower rates of autism and ADHD. These findings are not just counterintuitive—they are biologically absurd. No plausible mechanism exists by which aluminum salts could prevent neurodevelopmental delay. These inverse associations are a statistical artifact of the model’s self-sanitizing logic, symptomatic of model overfit. Yet they are being promoted as further reassurance.

We’ve seen this before. For decades, null or inverse associations have been delivered to the public as proof of safety, often based on studies designed with precision-engineered blind spots. The HPV vaccine “safety” studies excluded adverse events within 30 days. Influenza vaccine pregnancy studies removed preterm births that occurred shortly after vaccination. Time-window exclusion, exposure misclassification, adjustment for mediators, and silent conflicts of interest—this is the language of modern regulatory science.

The Illusion of Independence

Though most authors claimed no relevant conflicts, one of the lead authors, Anders Hviid, disclosed involvement with VAC4EU—a European consortium devoted to vaccine surveillance—and received funding from Novo Nordisk Fonden and Lundbeckfonden, both deeply intertwined with Danish health policy and biomedical investment. While these relationships were declared to be "outside the scope" of the study, their influence is unmistakable. Hviid’s long-standing role in vaccine safety research—always producing results favorable to industry and public health authorities—is a feature, not a bug. Example, see “An Autopsy on Hviid et al. 2019’s MMR/Vaccine Science-Like Activities”.

Of the new study, Hviid was held up by Stat News in his intrepretation. Stat News reported:

“Hviid and his co-authors said that while there wasn’t a statistically significant increase seen for any of the conditions they studied, they could not rule out the possibility of a small increased risk for some very rare conditions. ‘So that tells us that clearly that there’s not an epidemic of chronic diseases in childhood’ associated with aluminum in vaccines, he said.”

Translational Significance of the Study as Published vs. As Interpreted

Translational research requires knowing what part, if any, of a study generalizes to the full population—and how. A direct read of the study for translational implications tells us that for the result to hold—for aluminum-containing vaccines to be deemed unrelated to chronic illness—infants with a higher need for doctors’ visits in the first 24 months of life should not receive any more aluminum-containing vaccines after the age of two. That is the only way the adjustment for GP visits could validly isolate the effects of aluminum without introducing bias. But the Danish vaccination schedule includes additional aluminum-adjuvanted vaccines beyond age two, and children who interact more frequently with the medical system are precisely those more likely to remain on the full schedule.

Given the well-established mechanisms of disease-inducing effects of aluminum exposure on health, and the indicator of the effects of exposure on children, for the result to be translated for clinical care properly, children who see the doctor more often would need to be less likely to receive vaccines after 24 months—or at least, less aluminum. The pea-and-shell game hides the effect of aluminum on the subset that would be the exposure significant in the full, unadjusted analysis.

The authors instead claim that their study provides reassurance about the safety of aluminum in early-life vaccination broadly, extrapolating findings from a flawed, internally inconsistent model to the entire pediatric population. They make no effort to explore or account for the fact that the model conditions on a downstream variable, nor do they clarify that any translational application of the findings must assume a clinical scenario that does not exist—one in which early health concerns lead to fewer vaccinations later in life. Their statistical adjustment leads to an irrelevant result because the model does not apply in real life.

The problem would have been obvious if they had excluded sick children who potential had pre-24 month related chronic illness. Instead, they statistically “adjusted for” being chronically ill before 24 months of age. This hides the risk.

The result is a dangerous fiction: a statistically constructed claim of general safety that depends, for its validity, on a clinical pattern diametrically opposed to actual practice. The study, as published, cannot be translated into public health guidance without reinforcing a model that erases the most vulnerable children from visibility and analysis. It is not merely unhelpful to translational science. It is antithetical to it.

If translation of the study into clinical practice were to be conducted in parity to the study’s design of analysis, infants who go to the doctors’ office frequency before 24 months born of a certain gender, of a certain birthweight, to mothers with poor health who smoked during pregnancy, used prescription drugs during pregnancy and who lived in households of certain income should not receive aluminum-containing vaccines after the age of 24 months, either. These are the adjustments made to the data; to be relevant to real-life, clinicians in Denmark should be developing intake forms to find these risk factors and then refuse to vaccinate such children.

Then and only then aluminum-containing vaccines would become less risky at the population level. Then and only then would the study be relevant to the population.

What the Public Deserves to Know

Aluminum adjuvants are not trace contaminants. They are purposeful immunostimulants, injected in multiple doses into neonates and infants. They persist in tissues, migrate across membranes, accumulate in brain and lymph, and can bypass first-pass hepatic metabolism. Their biological impact cannot be evaluated by population-averaged, registry-tethered, schedule-derived epidemiology that assumes linearity and adjusts away symptoms.

The public deserves real science: prospective studies, with biomarker validation, mechanistic anchoring, and full exposure characterization. We need modeling that respects the complexity of biology, not spreadsheets engineered to erase it.

Additional Methodological Failures: A Masterclass in How Not to Study Toxicology

Beyond the core flaw—adjusting for general practitioner visits, which eliminates signal by conditioning on downstream consequences—this study suffers from a constellation of other critical failures. Each would be disqualifying on its own in a properly refereed toxicological analysis. Together, they form a template for how to engineer a study that appears robust but is epistemically sterile.

To begin, the study entirely lacks a truly unexposed control group. With over 98% of Danish children vaccinated in line with the national schedule, all participants in the study were exposed to aluminum—some more, some slightly less. This design precludes any statement about absolute risk. A dose–response gradient measured entirely among the exposed is not a study of toxicity. It is a study of program compliance. You cannot measure the effects of aluminum in a population where every child has received it. This violates the first rule of causal inference: you need contrast.

The reliance on birth year as a proxy for exposure invites ecological confounding on a grand scale. Over the 24-year period covered by the study, diagnostic criteria, healthcare access, registry completeness, and even the social acceptability of seeking help for developmental delay all changed dramatically. Autism, in particular, underwent diagnostic broadening in the early 2000s. The study assumes that children born in different decades can be compared without adjusting for these changes. That assumption is false. And it distorts the results.

The issue of conflicts of interest, while nominally addressed in disclosure forms, also deserves scrutiny. The senior author, Anders Hviid, discloses ties to VAC4EU and funding from organizations that are deeply embedded in the biomedical establishment—Novo Nordisk Fonden, Lundbeckfonden, and others. While none of these ties are presented as influencing the current manuscript, they reflect a broader institutional orientation: one that defends the vaccine program’s integrity first and investigates its unintended consequences last—if at all.

The study includes no biomarker validation. No attempt was made to measure aluminum in serum, hair, urine, or tissue. No subgroup analyses were performed for children with known genetic susceptibility—MTHFR polymorphisms, glutathione pathway defects, or mitochondrial vulnerabilities. The results are averages, smoothed across a heterogeneous population, with no attention paid to those most at risk.

The study models aluminum’s effects as if they occur in isolation, despite well-documented synergistic interactions with other toxicants. Thimerosal, acetaminophen, antibiotics, environmental pollutants, and early-life infections can all modulate neuroimmune status. Yet none are included in the analysis. Toxicology without cofactor modeling is fiction. It imagines that children are chemical monads, not ecological beings immersed in overlapping exposures.

And then, with all of these flaws—design, data, model, omission—the authors deliver their conclusion in press-ready language: aluminum is safe. No association found. Narrative complete.

Final Word

This study did not fail to find harm. It was constructed to not find it. Its adjustments obscure, its assumptions distort, and its conclusions mislead.

If public health officials or regulatory bodies cite this paper as evidence that aluminum adjuvants are safe, they do so not in ignorance, but in allegiance—to a methodology and a legacy of scientific obfuscation that has protected products at the expense of people.

When a model removes the sickest from analysis and declares the vaccine suitable for everyone, it has not found safety. It has fabricated it.

And that is what this study represents. Not science. Not safety. Fabrication. Any ethical physician or policymaker must look at this study, and the organizations representing it as good science, with deeply cautious skepticism.

James Lyons-Weiler, PhD hold the positions of Strategic Director of Policy Integration and Research Realignment, mahainstitute.us, Founder of ipak-edu.org, and Editor-in-Chief of Science, Public Health Policy & the Law.