By Yaakov Ophir at Brownstone dot org.

Yesterday, U.S. Secretary of Health and Human Services Robert F. Kennedy, Jr. published a sharp and comprehensive critique of the recent study by Andersson et al., which was published in the Annals of Internal Medicine. The study made headlines for claiming that aluminum-adjuvanted vaccines administered in early childhood are not associated with increased risks of autoimmune, allergic, or neurodevelopmental disorders.

Kennedy did not mince words. He described the study as "so deeply flawed it functions not as science but as a deceitful propaganda stunt by the pharmaceutical industry." Among the many questionable features he identified, one stood out to me in particular. Kennedy wrote:

"These sleights of hand magnify the potential for allowing the authors to reach their absurd suggestion that higher aluminum exposure is somehow protective against asthma, allergies, and neurodevelopmental disorders, including autism."

This sentence stopped me in my tracks because I had noticed the exact same thing. While Kennedy voiced this concern from the standpoint of public health advocacy, I approached the same issue from an academic and data-driven perspective. What I found not only aligns with his observation but adds further empirical grounding to it. In fact, this very point was at the heart of a formal comment I submitted to the Annals of Internal Medicine.

The authors of the study responded - but, in my view, did not adequately address the core contradiction. In this short article, I lay out the full story, supported by the data, to show why this implausible pattern of protective effects cannot be ignored.

A Pattern Too Good to Be True

In their central figure (see the screenshot below), Andersson et al. report hazard ratios for 34 different health outcomes, comparing children with different levels of aluminum exposure through vaccination. At first glance, the figure seems balanced and comprehensive. But a closer look reveals a startling trend: 25 of the 34 estimates (73.5%) leaned in the same direction - suggesting that more aluminum exposure was associated with lower risk.

And not merely by random chance: over half of these "protective" associations were statistically significant (with 95% confidence intervals excluding 1.0). Strikingly, children who received higher doses of aluminum appeared to have lower risks of developing conditions such as food allergies, autism, and ADHD.

This pattern is implausible on its face unless aluminum is a miracle drug no one told us about. Even under the null hypothesis that aluminum has no effect, such a one-sided distribution of significant inverse findings would be highly unlikely. But beyond statistical expectations, the results also contradict epidemiological reality.

Ignoring the Direction of History

Although Andersson et al. briefly acknowledge the possibility of residual confounding (i.e., hidden variables not accounted for in the analysis), they fail to confront this strange and highly asymmetric pattern of their results. Instead, they suggest that rising diagnosis rates over time may have inflated risk estimates for children born later, who also happened to receive more aluminum due to changes in the vaccine schedule.

But this explanation only deepens the puzzle! If diagnoses have indeed increased over time, as they rightly note, then we would expect later birth cohorts (with higher aluminum exposure) to show higher observed rates of neurodevelopmental and allergic conditions, even if aluminum had no effect at all. The direction of bias should have skewed the results toward risk, not protection.

Take ADHD, a diagnosis I've studied and questioned extensively, as a telling case. According to the Diagnostic and Statistical Manual of Mental Disorders (DSM), ADHD prevalence was estimated at around 3% in 1980 and rose to 7.2% by 2022 (in the latest edition of the manual). In fact, these figures are considered conservative; many studies have reported strik...