Don't Be Foiled by RFK Jr.'s Rebuttal of the Danish Aluminum Study

Scientists respond with data, not rhetoric

When the Referee Becomes the Player

We find ourselves at a critical juncture in public health history. The person charged with overseeing America's health agencies, including those responsible for vaccine safety monitoring, is actively working to undermine confidence in one of the most comprehensive vaccine safety studies ever conducted. Health Secretary RFK Jr. has launched a blistering attack on a Danish study published in the Annals of Internal Medicine that found no link between aluminum in vaccines and 50 different health conditions, including autism. He’s even gone so far as to call for the publication’s retraction–a step that the well-respected medical journal has made clear it will not take.

His critique, which he claims reveals a "devastating indictment" of aluminum-containing vaccines, fundamentally misrepresents both the study's methods and its findings. Let's walk through what the science shows, and why Kennedy's interpretation doesn't hold water.

Important note: The authors of this study have addressed all of these critiques, publicly explaining their rationale and even conducting additional analyses using the methods critics requested. They reported these secondary results openly, and the findings still support their original conclusions. This is an impressive show of responsiveness on the authors` part to be open to criticism and address concerns. These responses and new data—some released just today—are available here and in the article’s comments section if you scroll to the bottom of the published article.

One more thing: This article is more technical than our typical content, but we felt the detail was essential given the serious nature of these critiques. We've included plain-language explanations where possible, but some complexity is unavoidable when defending good science against misleading attacks.

The Study: What It Did

Denmark has a national Medical Birth Registry that enables comprehensive data collection. This Danish study by Andersson et al. included nearly 1.2 million children born between 1997 and 2018, and examined whether aluminum exposure from vaccines received in the first two years of life was associated with 50 different health conditions:

• 36 autoimmune diseases

• 9 allergic or atopic conditions

• 5 neurodevelopmental disorders (including autism spectrum disorder and ADHD)

The researchers leveraged changes in Denmark’s childhood vaccination schedule over this period, allowing them to examine natural variation in aluminum exposure (ranging from 0 to 4.5 mg by age 2). This "natural experiment" design is particularly powerful because it examines the effects of cumulative aluminum exposure in a real-world context due to system-level rather than individual-level changes.

The results were clear: Each 1-mg increase in aluminum exposure from vaccines did not lead to an increased risk of developing nearly every condition studied. The hazard ratios were:

• Autoimmune conditions: 0.98 (95% CI: 0.94-1.02)

• Atopic and allergy conditions: 0.99 (95% CI: 0.98-1.01)

• Neurodevelopmental conditions: 0.93 (95% CI: 0.90-0.97)

Graphic created by Unbiased Science

Kennedy's Core Claims

Claim 1: "The Study Excluded High-Risk Children to Hide Harm"

Kennedy's Assertion: The researchers deliberately excluded children who died before age 2, those with early illnesses, and those with implausibly high levels of aluminum exposures (2.8% of children) to hide vaccine injuries.

The Reality: These are standard epidemiological practices, not manipulation:

• Excluding early deaths and kids diagnosed before 2 years of age: The researchers were specifically interested in new diagnoses of outcomes after 2 years of age, so they could avoid any shorter-term health differences in children under the age of 2. Children who already had one of the outcomes of interest or died before age 2, by definition, cannot receive a new diagnosis of the outcome after age 2. Starting follow-up at age 2 ensured the total level of exposure to aluminum by age 2 was set and allowed for the expected delay between symptom onset and diagnosis of these outcomes. In a secondary analysis, the authors assessed follow-up starting at 14 months and found similar results.

Graphic created by Unbiased Science

• Excluding severe congenital conditions: Children with conditions like severe immunodeficiency or organ failure have fundamentally different care-seeking behaviors and health trajectories. The authors sought to ensure the children under the age of two were similar at baseline so that their conclusions could be more generalizable to an “average” childhood health trajectory following the receipt of aluminum-adjuvant vaccines. Including children with severe congenital conditions would muddy the waters and make the study findings less generalizable.

• Removing implausible values of aluminum exposure: These were likely data entry errors due to duplicate records or coding mistakes. As Dr. Anders Hviid explained in his response, keeping obvious errors in the dataset would distort the accuracy of the results more than removing the data points entirely.

Late-breaking! Following this critique, the authors performed a new sensitivity analysis, posted today in the comments section of the original article (scroll to the bottom), including children with an implausibly high number of vaccinations in restricted to those fully vaccinated. Even with this change, there was no link between aluminum exposure from vaccines and autoimmune disorders (HR=1.00; 95% CI: 0.92-1.08), atopic or allergic disorders (HR=1.03; 95% CI: 1.00-1.07), or neurodevelopmental disorders (HR=0.96; 95% CI: 0.88-1.04).

Claim 2: "Adjusting for Doctor Visits Creates 'Collider Bias'"

Kennedy's Assertion: By adjusting for general practitioner (GP) visits before age 2, the study inappropriately controlled for a consequence of vaccine injury, masking the true harm.

Late-breaking! Before getting into the scientific rationale for adjusting for GP visits before age 2, the authors responded to this critique with yet another sensitivity analysis posted today in the comments section of the original article (scroll to the bottom). Without adjusting for GP visits before age 2, the findings were the same for autoimmune (HR 1.01 [95% confidence interval, 0.97 to 1.05]) or neurodevelopmental disorders (HR 0.97 [95% confidence interval, 0.93 to 1.00]). While this suggests the adjustment did not meaningfully bias the results, explaining the reasoning is still important, as similar methodological debates are likely to arise in future large observational studies about vaccine safety.

The Reality: Children who see doctors more often are more likely to receive a diagnosis simply because there are more opportunities for evaluation. This can create "surveillance bias," where conditions can appear more common simply because they're looked for more often. At the same time, children who see their doctor more often before age 2 are also more likely to receive more vaccines.

Consider two children: one sees the doctor 10 times before age 2, the other only twice. The first child is much more likely to continue seeing their doctor and get diagnosed with something—not necessarily because they're sicker, but because they're being examined more often. This is especially true for neurodevelopmental concerns, which are regularly screened for at well visits, but might not prompt an urgent trip to the doctor the way something like asthma could. Additionally, the child who only saw the doctor twice may have limited healthcare access or be in a family that doesn’t seek regular care at a doctor’s office, and they will be less likely to be vaccinated on time.

If a study doesn’t adjust for healthcare utilization before age 2, it may find an association between aluminum in vaccines and a condition like autism simply because children with more visits are both more likely to get vaccinations and receive an autism diagnosis, not because a true causal association exists. In other words, the apparent link could be driven by differences in healthcare use, not by aluminum in vaccines causing autism.

To explain this, look at the directed acyclic graph (DAG) below. DAGs are like a map that scientists draw to show how different things might be connected and influence each other. Each item is a “node,” and arrows show the direction of influence—always moving forward, never looping back. Scientists use DAGs to think carefully about cause and effect, and to decide what factors they need to account for so their results aren’t misleading.

Graphic created by Unbiased Science

It is not clear how “collider bias” would have occurred. Collider bias is real and can certainly threaten study validity if present and not accounted for, but that’s not what’s happening here. If the number of GP visits before age 2 were a collider, it would have to be a common result of both aluminum exposure from vaccines and outcome diagnoses that occur after age 2, which just doesn’t make sense unless we have a time machine that can use information from the future to predict the past.

RFK Jr.'s argument starts from the assumption that aluminum in vaccines causes enough illness in children before age 2 to result in significantly more doctor’s visits, and by adjusting for this, you will mask those children. Every statistical model has some assumptions; scientists try to make those assumptions based on the best available evidence. To make this assumption a priori would be speaking into existence an assumption sans scientific backing.

At the end of the day, no study is perfect; methodological choices always involve trade-offs between different types of bias. What the authors did in an attempt to minimize bias is standard epidemiological practice, not manipulation. And in a show of incredible transparency and responsiveness, the authors ran an additional analysis without adjusting for GP visits before 2 years of age and found no difference in their results.

Claim 3: "The Supplementary Data Shows a 67% Increase in Asperger's Risk"

Kennedy's Assertion: Hidden supplementary data reveals a statistically significant 67% increased risk of Asperger's syndrome, contradicting the study's conclusions.

The Reality: This is a textbook example of cherry-picking. Kennedy focuses on a single subgroup finding while ignoring crucial context:

• This finding was based on just 51 cases in children born between 2007-2018.

• The confidence interval (1.01-2.77) barely crossed statistical significance. This means it’s compatible with an increased risk anywhere from 1% to 177%, where a 1% increase in risk, though technically statistically significant, usually does not meet the threshold for clinical significance.

• The pattern didn't appear in the earlier birth cohort.

• It disappeared when follow-up was extended to age 8.

• It wasn't present for autism spectrum disorders overall.

• The above points are important because if this were a real association and not statistical noise, you would expect to see it in these other groups of analysis.

When you run dozens of statistical tests (as any comprehensive study must), you expect some results to cross the significance threshold by random chance alone. This is a well-known occurrence in statistics known as the "multiple comparisons problem". A single weakly significant finding in one subgroup, which is inconsistent across other groups, is noise, not a real signal. In simpler terms, if you flip enough coins, some will land on heads five times in a row because of luck, not because the coin is biased towards heads.

In fact, the authors were transparent about this—these results were not “hidden”. In their main analysis, they found that a 1-mg increase in aluminum exposure was associated with a 7% decrease in risk of diagnosed neurodevelopmental conditions after 2 years of age (HR=0.93; 95% CI: 0.90-0.97). Again, though technically statistically significant, the authors chalked this finding up to a multiple comparisons problem and residual confounding rather than proof that aluminum exposure protects against neurodevelopmental conditions.

Claim 4. “Kids with the Highest Level of Aluminum Exposure Had More Cases of Neurodevelopmental Conditions.”

Kennedy’s Assertion: In their conclusions, the authors ignore supplemental data showing children in the highest aluminum cohort had more cases of neurodevelopmental disorders, autistic disorders, and autism spectrum disorders compared to those with moderate aluminum exposure.

The Reality: While the authors focused their primary analysis on estimating hazard ratios for each 1 mg increase in aluminum “exposure”(described above), they also performed secondary analyses to validate that their primary analytical approach was sound. To do so, they classified aluminum exposure into categories: 0-1.5 mg–low; 1.5-3.0 mg–moderate, and 3.0-4.5 mg–high. However, kids born between 1997 and 2001 could never be included in the high-exposure group due to vaccine formulations and schedule recommendations at that time. The fact that kids from each birth cohort could not be represented in each exposure category is a violation of positivity–an assumption needed to make valid conclusions. While not as egregious as dividing by 0, this violation means that even though they were able to calculate absolute numbers of cases, those cases (and their confidence intervals) need to be interpreted with extreme caution. The purpose of these secondary analyses was not to test the main hypothesis again; it was to ensure their primary analysis methods were solid.

Late-breaking! In response to this critique, the authors re-ran this analysis with only children who were born between 2002-2018 who could hypothetically fall into any of the three exposure groups. The table below summarizes the authors’ new analysis results in the comments section of the original article (scroll to the bottom). All estimates contained 0–the null value for a risk difference (different than hazard ratio)–indicating there is no increased risk of autoimmune, atopic, allergic, or neurodevelopmental disorders in the low or moderate exposure groups compared to the high exposure group.

Risk differences (RD) and 95% confidence intervals, per 10,000 children, both compared to children who received 3-4.5 mg of aluminum through vaccines by age 2. The original Supplemental Figure 11 presents all children born between 1997-2018, whereas the revised analysis excludes children born before 2001 who could not fall into the higher exposure category.

Table created by Unbiased Science; adapted from Andersson et al.’s new analysis results in the comments section of the original article (scroll to the bottom).

Claim 5: "They Should Have Done a Vaccinated vs. Unvaccinated Comparison"

Kennedy's Assertion: The study failed by not directly comparing vaccinated and unvaccinated children.

The Reality: The study did include children who received zero aluminum from vaccines, but used a more sophisticated dose-response analysis across the full range of exposures (0 to 4.5 mg). This approach is more powerful for detecting effects than a simple binary comparison between unvaccinated and vaccinated groups because fully unvaccinated children are very rare in this population. Also, families who completely refuse vaccines differ from vaccinating families in numerous ways. For example, they will differ in healthcare-seeking behavior, lifestyle choices, and socioeconomic factors, among others. These differences are extremely hard to account for. These problems make it impossible to accurately compare vaccinated vs. unvaccinated in this type of registry and be able to trust the result as valid. The dose-response approach in this study allows researchers to detect any harmful effects of aluminum across the full spectrum of exposure.

In the supplementary figures, the authors ran a secondary analysis in which they excluded the 15,237 children who received no aluminum from vaccines to determine whether this group influenced their results. The results did not change, indicating these children did not bias their conclusions.

The Alleged Conflicts of Interest

Kennedy points to author affiliations with Statens Serum Institut (SSI) as evidence of bias. What he doesn't mention is:

• SSI is Denmark's national public health institute, not a "vaccine company."

• It sold its manufacturing capacity in 2017.

• The study was initiated to replicate a U.S. finding of increased asthma risk—hardly the action of researchers trying to hide harm.

• The Novo Nordisk Foundation that Kennedy criticizes also funds researchers he praises, like Professor Peter Aaby.

What About Kennedy's New "Gold Standard" Study?

Recent reports suggest Kennedy is considering reviewing the safety of vaccines that contain aluminum and planning a government-funded study to find "the cause" of autism, with aluminum as a possible target. This raises serious concerns:

1. Prejudging outcomes: Good science doesn't start with a conclusion and work backwards. Good science is driven by questions and hypotheses.

2. Ignoring existing evidence: We already have extensive research on autism's complex, multifactorial origins.

3. Resource allocation: Millions spent chasing disproven hypotheses means less funding for actual autism support and research.

We are concerned that the very tactics used to critique the Danish study could be misapplied to other large databases in the U.S. in attempts to “find” the cause of autism and link it to aluminum. It is particularly troubling that someone planning such a study is publicly criticizing this work with critiques that are not scientifically sound and misapply epidemiologic and statistical tools. These tools are powerful when used correctly, capable of shedding light on complex questions. But when misapplied, they can produce results that appear convincing yet are misleading—or entirely false.

The Bigger Picture: Decades of Research, Same Answer

This isn't vaccine critics' first attempt to find a culprit for autism in vaccines—it's just the latest chapter in a decades-long pattern of moving the goalposts:

The Thimerosal Era: For years, mercury-containing thimerosal was branded as the cause of autism. When it was removed from childhood vaccines in 2001 (despite no evidence of harm), autism rates continued to rise, and critics needed a new target.

The Fake MMR Scare: Andrew Wakefield's fraudulent 1998 study blamed the MMR vaccine. After his research was exposed as fabricated and his medical license revoked, that hypothesis collapsed.

Now, Aluminum: With previous boogeymen debunked, aluminum has become the new focus—despite this Danish study and others finding no link.

What We Know About Autism

The scientific consensus, based on extensive research, tells us that autism risk is largely determined before birth:

• Genetics: Hundreds of genes contribute to autism risk, with heritability estimated at 50-90%.

• Prenatal factors: Advanced parental age, certain infections during pregnancy, and prenatal medication exposures may all contribute to autism.

• Early brain development: Brain differences are detectable in utero, long before any vaccines are given.

The Danish study adds to a mountain of evidence on vaccine safety and autism specifically:

• A meta-analysis of 1.2 million children found no link between vaccines and autism.

• Studies of 81,933 children found no autism risk from maternal Tdap vaccination.

• Research on patients receiving aluminum-based allergy shots found lower autoimmune disease risk.

• Multiple studies show no correlation between vaccine history and blood aluminum levels.

The amount of aluminum in vaccines is minuscule compared to other sources. Infants consume more aluminum through breast milk and formula than they receive from their entire vaccine schedule. Some antacids contain more aluminum in a single dose than the entire childhood vaccine series combined.

Moving Forward: What Real Transparency Looks Like

Kennedy calls for transparency. True scientific transparency and integrity mean:

• Acknowledging when large, well-designed studies contradict your beliefs and being willing to reconsider your conclusions.

• Presenting all findings in context, not cherry-picking outliers.

• Questioning and testing the assumptions through secondary and sensitivity analyses, as done in this study.

• Understanding standard epidemiological methods and decision points before claiming bias.

• Recognizing that Danish and EU privacy laws legitimately protect patient data. We hope that Mr. Kennedy would likewise protect our patient data if another country demanded it.

Dr. Hviid has offered to help qualified researchers access the data through proper channels to replicate the findings and has run additional analyses—that's real transparency, not the release of raw patient data that would violate privacy laws.

The Stakes Couldn't Be Higher

We're watching something unprecedented unfold: America's top health official actively undermining confidence in vaccines using misleading interpretations of solid science. Kennedy's critique doesn't reveal legitimate flaws in the Danish study; it reveals either a fundamental misunderstanding of epidemiological methods or a willingness to misrepresent them.

The Danish study isn't perfect—no study is. And the United States is not Denmark, so we must acknowledge differences between our countries that could not be tested for. But it represents exactly the kind of large-scale, population-based research we need to ensure vaccine safety. With over 1.2 million children followed for up to 24 years, it provides compelling evidence that aluminum adjuvants, at the doses received through vaccines, are not driving an epidemic of chronic disease.

What's truly concerning isn't aluminum in vaccines, it's having someone in charge of public health who seems determined to find harm no matter what the evidence shows. As parents, healthcare providers, and citizens, we deserve health leadership that follows scientific evidence, not ideology.

And our children, who depend on us to make informed decisions about their health, deserve better.

Stay Curious,

Unbiased Science

Want to support our work? The best way is to subscribe to our Substack and share our content. While all our articles are always completely free to read, paid subscriptions help sustain our in-depth reporting on public health and science topics. Thank you for considering it!