Stats Don’t Lie, But Cherry-Picking Does
Debunking RFK Jr.’s Take on the Danish Aluminum Vaccine Study
IYKYK: Tuesdays are speed work days for runners. But with temps pushing 100°F here in the Northeast, I’m saving my track session for dawn tomorrow. Tonight, I’m swapping intervals for something equally pace-driven: a paper on the foundations of trust (inspired by The Human Margin by Katherine Meese and Quint Studer) that’s due Wednesday night.
I’m easily distracted by the news, a leftover reflex from my days chasing breaking stories. So when I saw today’s headlines about the latest RFK Jr. saga, I thought: this is the perfect springboard for reflecting on how trust is built (or eroded), along with a quick refresher on statistics that I hope you’ll find useful.
The Saga in Three Acts
July 15, 2025 – The Annals of Internal Medicine publishes Aluminum-Adsorbed Vaccines and Chronic Diseases in Childhood: A Nationwide Cohort Study, led by Niklas Worm Andersson and colleagues at Denmark’s Statens Serum Institut
Aug 1, 2025 – RFK Jr. writes on TS News, Flawed Science, Bought Conclusions: The Aluminum Vaccine Study the Media Won’t Question, asks the Annals to “immediately retract this badly flawed study”
Aug 11, 2025 – Annals tells Reuters they’re standing by the study and rejecting RFK’s retraction request
What the Danish Study Found
The researchers followed 1.2 million children born between 1997 and 2018 to see whether aluminum exposure from vaccines in the first two years of life was linked to 50 different health outcomes: 36 autoimmune diseases, 9 allergic or atopic conditions, and 5 neurodevelopmental disorders, including autism spectrum disorder and ADHD.
The result? No statistically significant link between higher aluminum exposure and increased risk for any of the 50 health outcomes.
Autoimmune conditions: 0.98 (95% CI: 0.94–1.02)
Atopic/allergy conditions: 0.99 (95% CI: 0.98–1.01)
Neurodevelopmental conditions: 0.93 (95% CI: 0.90–0.97)
I won’t assume all readers are familiar with these numbers and notations. The bolded figures are called the hazard ratio (HR), and the numbers in parentheses are the confidence interval (CI). Here’s a geeky way of representing them, as shown in the Andersson et al paper:
Hazard Ratios: Two Runners, One Track
Apologies for the runner analogy but this is my way of hitting two birds in one stone. Think of a hazard ratio as comparing how quickly something happens in one group vs. another over time.
In runner terms:
HR = 1 → A tie
HR > 1 → Runner A is faster (event happens sooner/more often)
HR < 1 → Runner A is slower (event happens later/less often)
Confidence Intervals: The “Race Replay”
We don’t just race once, we run multiple heats.
Sometimes Runner A wins by 5%, sometimes 15%, sometimes it’s a tie
The confidence interval (CI) is the range of these possible differences
Example:
HR = 1.10, CI: 1.05–1.15 → Narrow range, confident A is faster
HR = 1.10, CI: 0.95–1.25 → Range includes 1.00, so it could be just noise
If the CI crosses 1, you can’t claim a statistically significant effect.
So, back to the study results, here’s how you can read the data:
Autoimmune conditions: 0.98 (95% CI: 0.94–1.02)
Think of two runners competing in the “autoimmune conditions race.” A hazard ratio of 0.98 means the higher-aluminum runner finished almost exactly at the same time as the lower-aluminum runner, maybe a hair slower (2% behind), but close enough that it could just be chance. The CI range (0.94–1.02) includes 1.00, which means we can’t say for sure one was faster.Atopic/allergy conditions: 0.99 (95% CI: 0.98–1.01)
In this race, the runners crossed the finish line side-by-side. HR = 0.99 means there was virtually no difference in pace. The CI is tight (0.98–1.01), so we can be very confident neither runner had a real advantage.Neurodevelopmental conditions: 0.93 (95% CI: 0.90–0.97)
Here, the higher-aluminum runner finished 7% slower than the lower-aluminum runner, meaning fewer cases occurred in that group over time. The CI (0.90–0.97) is entirely below 1.00, so this difference is statistically significant, but still small enough that it could be due to other factors, not just aluminum exposure.
“67% Increased Risk” Claim and Why This is Misleading?
RFK points to a single subgroup in the supplementary data. He wrote (highlights are mine):
“Despite all the deceptive devices the authors used to conceal the signals of harm, Andersson et al.’s own supplementary data are a devastating indictment of aluminum-containing vaccines. These data, which they were forced to publish because of public criticism of their analysis, directly contradict the study’s conclusions. The data show a statistically significant 67% increased risk of Asperger’s syndrome per 1 mg increase in aluminum exposure among children born between 2007 and 2018.”
HR = 1.67 (CI: 1.01–2.77) → “67% faster” in our analogy (higher risk)
One tiny race out of 50. The study held 50 “races” (health outcomes). In most, both runners tied.
Tiny race size. Only 51 children in the higher-aluminum group had Asperger’s. In running terms, one stumble changes the average a lot.
The main race showed no difference. Overall Asperger’s HR = 0.99 (CI: 0.76–1.29)—a tie!
Classic cherry-picking. Highlighting one lucky win while ignoring the full scoreboard isn’t science.
If you run enough races, random variation guarantees a few “big wins” even when the runners are evenly matched. That’s why scientists look at patterns, not one-off heats.
The Trust Angle
The real story here isn’t that a small subgroup had a different result—it’s that the entire body of evidence points to no link between aluminum in vaccines and the conditions studied.
The Danish team earned trust by sharing methods, handling data transparently, and publishing the full scoreboard. On the other hand, lifting one shaky result out of context is the opposite. Trust is built when you tell the whole story, not just the part that fits your narrative.
GO’Manio