RFK Jr. Aluminium Claims about a Danish Vaccine Safety are fallacous: The Evidence Explained

By /

FACT: While real-world observational studies are not as controlled as randomised trials, they can capture much larger, more diverse populations over longer periods, and their results—when combined with other high-quality studies—form an essential part of the total body of evidence on safety and effectiveness.

In July, the largest-ever study that specifically looked at the amount of aluminium-containing vaccine exposure and child health was published. It was rigorous, transparent, and reassuring. As with any study of this kind, it has limitations (I dive into these). But for RFK Jr., it became yet another conspiracy theory to dismantle—with arguments that don’t hold up to scrutiny.

The irony, of course, is that Kennedy’s critique is a masterclass in distortion. His piece doesn’t engage with the actual methods or findings of the study in good faith. Instead, it misrepresents key design choices, misunderstands statistical principles, and promotes long-debunked aluminium scare narratives. He has also continued his attack on social media. Let’s walk through his central claims—and why they don’t hold up under scrutiny. 

FACT: Exclusion of incomplete or invalid data is standard in epidemiology.

Large-scale studies routinely exclude children with missing or implausible data to avoid distorted results and protect the validity of findings.

Fact check: Excluding children with incomplete follow-up is standard epidemiological practice to prevent biased exposure-outcome associations. Including children who died early or lacked consistent data could inflate or distort associations due to incomplete information. This is not “rigging” the study—it’s preserving its internal validity.

RFK Jr. claims 34,547 children with “high exposure” were removed. In reality, the authors excluded records with clearly erroneous data—e.g., exposure levels that couldn’t be reached under Denmark’s national vaccine schedule. This is transparent, documented, and reasonable.

Excluding children with implausible or missing data is akin to throwing out broken crash dummies in a car safety test—necessary for valid results, not deliberate rigging. This is standard protocol described in basic epidemiology textbooks (see Epidemiology 101 primers on loss to follow‑up and exclusion criteria). Intro to Epidemiology 101

Bottom line: Exclusion for data integrity ≠ data manipulation. It’s Epidemiology 101.

FACT: Denmark’s universal healthcare system allows for near-complete data coverage.

With over 90% vaccine uptake and population-wide registries, Danish cohort studies avoid the kind of “healthy user bias” common in less comprehensive systems.

Fact check: “Healthy user bias” refers to systematic differences in healthcare behaviours between vaccinated and unvaccinated individuals—often relevant in adult observational studies. Here, we’re looking at a nationwide birth cohort with >90% vaccine coverage and universal healthcare, where such bias is minimal.

Moreover, this wasn’t a voluntary cohort. It was a population-wide dataset, and exposures were objectively measured via vaccine registry—not self-reported or biased by health behaviours.

FACT: Adjusting for healthcare utilisation helps ensure diagnoses aren’t biased by access.

Epidemiologists often adjust for how often people use healthcare services to account for differences in diagnosis rates—not to hide associations.

Fact check: Collider bias occurs when you condition on a variable influenced by both the exposure and outcome. RFK Jr. incorrectly assumes that GP visits are necessarily downstream of aluminium exposure and early illness. In fact, adjusting for healthcare utilisation is a well-established method to control for diagnostic bias.

Imagine you’re trying to figure out whether eating ice cream causes sunburn. You decide to adjust for “beach attendance” in your analysis—because people who go to the beach are more likely to eat ice cream and get sunburned. That’s controlling for a confounder—a good idea.

Now imagine someone says: “Wait! You can’t do that! You’re hiding the fact that ice cream causes sunburn!” That’s RFK Jr.’s level of reasoning.

In the Andersson study, general practitioner visits were used to adjust for health-seeking behaviour and early diagnosis rates, not as a proxy for exposure or outcome. It’s not like adjusting for coughing when studying smoking and lung cancer (his flawed analogy); it’s more like adjusting for how often people visit doctors to ensure you’re comparing groups with similar chances of being diagnosed

The authors transparently justified this adjustment to reduce confounding—not to hide associations. His smoking analogy (adjusting for coughing) is flawed and shows a misunderstanding of how covariate adjustment works in longitudinal studies.

That said, this theoretical concern can be addressed by simply removing the variable and seeing what happens. Maybe they will.

Tönnies T, Kahl S, Kuss O. Collider Bias in Observational Studies. Dtsch Arztebl Int. 2022 Feb 18;119(7):107-122. doi: 10.3238/arztebl.m2022.0076. PMID: 34939918; PMCID: PMC9131185.

FACT: Most outcomes in the study showed no association with aluminium exposure.

Across 50 chronic conditions, the study found no consistent or biologically plausible links between aluminium exposure and adverse outcomes.

Fact check: The authors did not conclude that aluminium was “protective.” They found no consistent or biologically plausible associations between aluminium exposure and any of the 50 chronic disorders studied. Some hazard ratios were <1, some were >1, none were statistically compelling, and most were compatible with no effect. At no time, anywhere, do the authors suggest a protective effect. 

Here’s what they actually do:

  • Primary analyses: They report no statistically significant association between aluminium exposure and autism spectrum disorders (ASD) overall, or with most other chronic conditions.
  • Hazard ratios (HRs): Some HRs for certain outcomes are below 1, meaning the point estimate leans toward a lower risk — but the authors treat these as statistical variation, not as evidence of a real protective effect. They explicitly caution that these results should be interpreted in the context of multiple testing and possible residual confounding.
  • Wording: They never state, suggest, or imply that aluminium exposure protects against autism. Instead, they repeatedly use neutral language like “no association” or “findings consistent with no effect.”

Bottom line: The “protective effect” claim comes from misinterpretation or deliberate spin — not from the study authors.

“Cumulative aluminum exposure from vaccination during the first 2 years of life was not associated with increased rates of any of the 50 disorders assessed… For most outcomes, the findings were incompatible with moderate to large relative increases in risk, although small relative effects, particularly for some rarer disorders, could not be statistically excluded.” acpjournals.org

RFK Jr. cherry-picks a supplementary analysis (Asperger’s in a subgroup), ignoring multiple comparisons and lack of replication. If you test dozens of outcomes, a few will be “statistically significant” by chance alone. That’s basic statistical literacy.

FACT: Very few Danish children remain unvaccinated, limiting comparisons.

In high-coverage populations, unvaccinated children are rare and systematically different in ways that make them unsuitable as a control group in observational research.

🚫 Why “Vaccinated vs. Unvaccinated” Studies Are Not Ethical or Feasible

  • Ethical obligation to protect health: Once a vaccine is shown to be highly safe and effective, withholding it from a control group would expose them to preventable illness, disability, or death—violating medical ethics and the principle of do no harm.
  • Loss of equipoise: Randomised controlled trials (RCTs) require genuine uncertainty about the benefit–risk balance. For established vaccines, the benefit is already clear, so randomising children to go without is unethical.
  • Self-selection bias in observational comparisons: In real-world settings, those who remain unvaccinated often differ in health behaviours, healthcare access, and socioeconomics, introducing major confounding that is hard to eliminate.
  • Feasibility limitations: In countries with high vaccine uptake, the unvaccinated population is too small and unrepresentative to yield meaningful, generalisable results.

🧠 How We Study Vaccine Safety and Effectiveness Without RCTs

  • Cohort studies: Following large vaccinated and partially vaccinated groups over time using national health records, adjusting for confounding factors.- like this danish one
  • Case–control and test-negative designs: Comparing vaccine status in people with a disease versus those without, controlling for healthcare-seeking behaviour.
  • Dose–response analyses: Looking for consistent patterns between number/timing of doses and outcomes (e.g., Andersson et al.’s aluminium study).
  • Self-controlled case series (SCCS): Comparing periods shortly after vaccination with other time periods in the same person, eliminating between-person confounding.
  • Global post-licensure surveillance systems: Pooling registry and safety data from multiple countries to detect rare events and monitor trends.

Fact check: In a country with high vaccine uptake, true unvaccinated children are exceedingly rare, often with data limitations. The authors included them in the lowest exposure category but explicitly acknowledged this limitation.

Comparing vaccinated and unvaccinated groups in non-randomised settings introduces huge confounding. Children who are unvaccinated differ in many ways (socioeconomic, health-seeking behaviour, parental beliefs) that can skew results. That’s why dose–response studies—like this one—are more reliable.

FACT: Population-based registry studies are not designed to test hypothetical toxin interactions.

Epidemiological studies using health registries are built to detect real-world health patterns, not to explore speculative gene–toxin combinations unsupported by population data.

Fact check: The study used real-world, registry-based data—it wasn’t a mechanistic or toxicological study. Kennedy wants them to simultaneously model gene-environment interactions, synergistic toxin effects, and hypothetical predispositions—none of which are supported by reliable human data at scale.

Imagine you’re using a high-powered telescope to study patterns of planetary motion. It’s great for identifying orbits, predicting eclipses, and understanding celestial mechanics. Then someone shows up and says, “But your telescope is useless unless it can also diagnose cancer from a skin mole on Mars.”

That’s exactly what RFK Jr. is doing.

The Danish study is designed to detect population-level associations using large health registries. It’s not a genetic study. It’s not a lab experiment. It’s not supposed to model every possible interaction between genes, mercury, aluminium, mitochondrial disease, and the kitchen sink. To expect that is not just unreasonable—it’s a fundamental category error.

Adding poorly defined, unmeasured variables would introduce noise, not insight. There are other approaches for exploring these kinds of things, and such studies, when well designed,  should be supported in my opinion.

FACT: Danish health registries are internationally recognised and widely used.

The registries used in this study have high accuracy and are trusted worldwide for autism research and other epidemiological studies. In fact, Denmark’s Psychiatric Central Research Register captures both inpatient and outpatient ASD diagnoses at a national scale. https://www.nature.com/articles/s41591-024-03479-5

https://pubmed.ncbi.nlm.nih.gov/19728067

Fact check: The Danish National Patient Registry and Danish Psychiatric Central Research Register are internationally respected data sources with high positive predictive value for autism diagnoses. Are they perfect? No. But they’re good enough for robust population-level research, and their limitations are well known and discussed in the study.

Moreover, RFK Jr. is wrong to assume that most cases are missed. Moderate-to-severe developmental conditions like ASD are routinely captured in Denmark’s comprehensive health system.

FACT: A few significant findings in a large dataset are expected by chance.

In any large study with many outcomes, a small number of statistically significant results will arise by chance and must be interpreted with caution.

When a study examines a large number of outcomes—like the 50+ chronic conditions in the Andersson et al. analysis—some results will appear “statistically significant” purely by chance, even when no real association exists. This is called the multiple comparisons problem (or multiplicity). For example, if you test 100 independent hypotheses at a 5% significance level, you’d expect about 5 “significant” results just by random variation.

Epidemiologists use several strategies to manage this:

  • Pre-specifying primary outcomes before analysis to limit the role of chance findings.
  • Adjusting for multiple testing (e.g., Bonferroni or false discovery rate corrections) in exploratory analyses.
  • Looking for consistency across related outcomes and subgroups—true associations should appear repeatedly, not just once.
  • Assessing biological plausibility and replication in other studies before drawing conclusions. – 

FACT: The FDA’s Mitkus model remains the best available method for assessing aluminium pharmacokinetics.

The Mitkus et al. (2011) study is a U.S. FDA pharmacokinetic modelling paper that estimates aluminium body burden in infants following vaccination. It’s one of the key references regulators use when evaluating aluminium adjuvant safety in early life.

Mitkus RJ, King DB, Hess MA, Forshee RA, Walderhaug MO. Updated aluminum pharmacokinetics following infant exposures through diet and vaccination. Vaccine. 2011;29(51):9538-9543. https://doi.org/10.1016/j.vaccine.2011.09.124

Fact check: The Mitkus et al. model, while imperfect, is the best available approach to estimating aluminium pharmacokinetics in infants. It used conservative assumptions and showed that aluminium exposure from vaccines remains well below theoretical toxicity thresholds. The idea that this renders it “irrelevant” is misleading. No model is perfect—but Kennedy offers none of his own.

Also: The aluminium in vaccines is not freely circulating—it’s in the form of insoluble adjuvants designed for gradual immune activation, not neurotoxicity.

[Click Here for a more detailed commentary on this model]

FACT: Public institutions fund a wide range of health research, not just vaccine studies.

Statens Serum Institut and the Novo Nordisk Foundation support independent public health research under strict conflict-of-interest policies.

Fact check: Kennedy targets the authors’ affiliations with Statens Serum Institut (which is under the auspices of the Danish Ministry of Health for disease control and research and not a pharmaceutical company) and other funders, including the Novo Nordisk Foundation ( an independent, Danish enterprise foundation that supports scientific, humanitarian and social causes). These organisations fund a vast range of independent public health research, not just vaccine promotion.

In contrast, Kennedy runs Children’s Health Defense, an anti-vaccine lobbying group that profits directly from sowing mistrust in vaccines. Accusing researchers of bias (when in this case it looks pretty benign) while failing to disclose your own (does not appear benign) is the epitome of hypocrisy.

FACT: Danish law protects personal health data and applies to all researchers equally.

Denmark’s privacy laws, (like many countries) prohibit releasing raw health data—not to conceal information, but to safeguard citizens’ personal records, in line with EU GDPR protections.

Fact check: Danish data privacy law prohibits individual-level data sharing outside secure environments. This applies to all researchers equally. The same laws protect Danish citizens’ health records. This isn’t secrecy—it’s ethics.

The study is reproducible within the Danish data environment, where access is granted based on scientific merit and ethics approval.

Some Additional strengths and limitations of the study

The Andersson et al. study examined chronic health outcomes, many of which — such as autism spectrum disorders, asthma, ADHD, and other neurodevelopmental or allergic conditions — are rarely diagnosed before age two and require a period of observation for accurate detection.

Additionally:

  • Exposure definition: The main exposure variable was cumulative aluminium dose from vaccines during the first two years of life. Including children before they reached age two would give incomplete exposure data.
  • Follow-up completeness: Children who died, emigrated, or otherwise lacked full registry follow-up before age two would contribute incomplete or biased data to analyses of long-term outcomes.

Pros of this approach

  • Avoids misclassification of exposure: Ensures that each child’s aluminium exposure is fully measured over the intended 0–24 month window.
  • Avoids outcome misclassification: Minimises counting early, unstable, or incorrect diagnoses.
  • Improves comparability: All children included have the same observation period for exposure and the same minimum time before outcome assessment.

Cons of this approach

  • Loss of early-life events: Any acute outcomes occurring before age two (e.g., severe allergic reactions, very early-onset conditions) would not be captured in the main analysis.
  • Excludes the sickest infants: Children who died before age two are excluded, which can remove some of the most vulnerable cases — this is a potential form of selection bias (“survivor bias”).
  • Limits generalisability: Results primarily apply to outcomes diagnosed after age two.

What the authors did to address this limitation

  • Defined exposure period uniformly: All children were followed for aluminium exposure during the first two years to ensure consistency.
  • Used national registries with complete follow-up: For those who survived past age two, registry data provided long-term, standardised follow-up for outcomes.
  • Addressed limitations openly: In the discussion, they acknowledged that excluding deaths before age two means very early adverse effects would not be captured in this design, and that their results apply to post-infancy health outcomes.

In Denmark, the childhood immunisation schedule is highly standardised — nearly all children receive the same vaccines at the same times.

  • Why this matters: With little variation in schedule or formulation, it’s harder to detect subtle differences in risk between exposure groups.
  • How the authors addressed it: They measured cumulative aluminium dose per child and analysed it as both a continuous and a categorical variable, making the most of the limited variation that does exist.
  • Implication: The results are strong for the Danish schedule but may not capture potential effects from schedules with substantially different timing or vaccine composition, like in the U.S.

Denmark has a relatively homogeneous population in terms of genetics, socioeconomic factors, and healthcare access.

  • Why this matters: Findings may not generalise to more diverse populations where genetic susceptibility, environmental exposures, and healthcare access vary more widely.
  • How the authors addressed it: They acknowledged this in the discussion and framed their findings as one part of the broader global evidence base, not the sole definitive study.
  • Implication: Results are internally valid for Denmark, but replication in more diverse populations would strengthen generalisability. I would like to see such work funded.

It’s like testing a car on the same stretch of smooth, well-paved road in perfect weather. You can be confident how it performs under those conditions, but you still want to see how it handles in rain, snow, or heavy traffic to know how it will perform everywhere.

  • The study relies on registry-coded diagnoses, which are highly accurate for many conditions (including autism) but may miss mild or subclinical cases that never come to specialist attention.
  • Diagnostic practices and awareness may have changed over the study period, particularly for neurodevelopmental disorders.

That said, as noted earlier, Denmark does very well in capturing cases. Other countries may not be as reliable.

  • Although the authors adjusted for many covariates (birth year, sex, maternal education, GP visits, etc.), unmeasured factors — such as detailed genetic risk, parental health behaviours, or environmental exposures — could still influence outcomes.
  • This is a limitation of all observational studies, not just this one.
  • Some of the 50 conditions are rare. Even in a large cohort, small numbers of cases limit the ability to detect small increases in risk with precision.
  • This is why hazard ratios for rare outcomes have wide confidence intervals.

Final Thoughts

RFK Jr.’s attack on this study is not a scientific critique—it’s a politically motivated screed riddled with logical fallacies, cherry-picked data, and fundamental misunderstandings of epidemiology. He demands “transparency” while spreading fear, “independence” while running a litigation-driven anti-vaccine business, and “rigour” while rejecting robust science in favour of low-quality animal studies and fringe speculation.

The Danish study is not perfect—no observational study is. But it’s large, transparent, methodologically sound, and consistent with decades of research showing no evidence that aluminium adjuvants in vaccines are harming children.

It’s time we stopped letting anti-vaccine ideologues define the terms of public health debates.

Discover more from

Subscribe to get the latest posts sent to your email.

Related Posts

Scroll to Top

Discover more from

Subscribe now to keep reading and get access to the full archive.

Continue reading