A recent study published in JAMA Pediatrics claims: “Association Between Maternal Fluoride Exposure During Pregnancy and IQ Scores in Offspring in Canada“. This is getting a lot of press, but also a lot of pushback. The study is at odds with a large body of evidence showing no association with normal fluoride exposure and IQ. There is still an open question about high levels of exposure – higher than is added to some public water systems – but there is no association at <2ppm (fluoridated water typically has 1ppm ).
The new study is a prospective study correlating IQ scores in children with the fluoride exposure of their mothers. Exposure was estimated based on the fluoride levels in the community and self-report of water intake, but was backed up by urine fluoride levels. The authors also measured lead and other heavy metals, which is an important control.
They found that:
A 1-mg/L increase in MUFSG was associated with a 4.49-point lower IQ score (95% CI, −8.38 to −0.60) in boys, but there was no statistically significant association with IQ scores in girls (B = 2.40; 95% CI, −2.53 to 7.33). A 1-mg higher daily intake of fluoride among pregnant women was associated with a 3.66 lower IQ score (95% CI, −7.16 to −0.14) in boys and girls.
While many aspects of the methods used by the researchers were reasonable, the results are curiously at odds with prior research, and have come under significant criticism, at least in their interpretation. Perhaps the most significant curious aspect of the results is the stark difference between boys and girls in the study. It’s hard to believe that a drop of 4.49 IQ points in boys was missed by prior research, and why is that effect so large while there is no effect in girls (just a non-statistical trend toward increased IQ).
When combined the statistical significance almost disappears. There was also a significant disconnect between performance IQ (-13.79* (-18.82, -7.28)) and verbal IQ (3.37 (-1.50, 8.24)). Why would verbal IQ increase while performance IQ decrease? Also, look at the wide variance in results.
I am also concerned that the unadjusted IQ differences were not significant. There was only a significant difference when they adjusted the scores, which they describe:
adjusted for HOME score, maternal education, race, child sex, and prenatal second-hand smoke exposure; because city was strongly multi-collinear with water fluoride concentration (VIF >20), it was excluded from the model
It is perfectly legitimate to make adjustments for these kinds of variables, but the need to do so exacerbates the major limitation of this study – it was entirely observational. At best the authors can conclude there is a correlation (but only for boys). They can try to control for known confounders, but doing so is difficult, and they cannot control for unknown confounders.
Further, when you have to do subgroup analysis in order to find significant results that is a red flag for noisy data. There are other red flags as well, like the huge variance in results, and the disconnect between performance and verbal IQ. So we have noisy correlational data at odds with prior research. Perhaps even more damning is that the negative results were driven largely by a small number of boys who had extreme levels of exposure. Take out the outliers, and the overall effect becomes non-significant.
There are also some specific technical criticisms. Alastair Hay, Professor (Emeritus) of Environmental Toxicology, University of Leeds, said:
Firstly, as the authors acknowledge maternal intake of fluoride has not been validated. I see this as a crucial failure. For a substance with a short half-life, such as fluoride, urine concentrations vary hugely and are really only representative of the last drink. Validation of intake is something you must do before looking at associations.
He also points out that while the authors controlled for maternal lead exposure, they did not control for post-natal lead exposure. Lead is known to have a profound effect on IQ, so this was a critical failure.
Rick Cooper, Professor of Cognitive Science, Birkbeck, University of London, pointed out:
The male sample includes a couple of extremely low IQ scores (below 70 would be special needs, but they have two boys with IQ in the 50s). The results would be more convincing if it were shown that they did not depend on these children.
There is no consideration of whether an effect, if present, might be non-linear. For example, it appears that there is no effect at low concentrations, with the effect only arising at extreme levels. The chosen statistical approach does not allow consideration of this possibility.
There are other criticisms along these same themes. There are a lot of anomalies in the statistics, meaning that you could choose to look at the data in different ways and get different results.
There was complete agreement among those experts commenting that these results at best amount to an interesting hypothesis that needs further study. We cannot, however, generalize from these findings or make any public recommendations. Overall the results are unconvincing and are most consistent with noisy data with the effects largely due to a few outliers.
The results remind me of the last study to allegedly find a correlation between lower IQ and fluoride exposure – the Harvard review in 2012. There it was very clear that there was only an effect when the highest levels of exposure were considered – much higher than water fluoridation. So again the best current conclusion is that there may be neurotoxicity from fluoride at toxic levels of exposure, many times what is considered safe, but no evidence upon which to conclude that the levels of exposure from fluoridated water have any neurotoxicity.