New Zealand opponents of community water fluoridation (CWF) are at it again. Their only response to the recently upgraded fluoridation review is to call it “propaganda” and to completely misrepresent it. But it’s interesting to look at their misrepresentation because it does highlight a basic flaw in the studies the anti-fluoride campaign has been promoting.
Fluoride Free NZ (FFNZ) claim in their recent press release (Chief Science Advisor Appears To Deliberately Mislead On Fluoride Science):
“among the many mistakes and reliance on out-of-date science, the most glaring issue is that she refers to two of the best studies ever carried out on fluoride and IQ (Mexico and Canada) as “having high prenatal exposure”. This is probably the most egregious misrepresentation in the review and hard to believe it was not done to purposely misrepresent.”
But this is completely false. In discussing the Canadian study the review actually says it:
“found that the mother’s exposure to fluoride during pregnancy was associated with lower IQ scores  in boys (but not girls), even at optimally fluoridated water levels (i.e. between 0.7-1.2 mg/L). If this finding were replicated in robust studies, it would potentially be concerning as Aotearoa New Zealand recommends fluoridation of water between 0.7 and 1.0 mg/L. There was significant and valid criticism of aspects of the study by many subject-matter experts when it was released (see for example, ‘expert reaction to study looking at maternal exposure to fluoride and IQ in children’). The study used sub-group analysis to find an association that is not explained in the paper (i.e. why were only boys affected  and why verbal IQ was not impacted), the effect appeared to be driven by the minority of participants that had much higher fluoride exposures (i.e. higher than those in Aotearoa New Zealand).” [My emphasis]
So the review does refer to the Canadian study being conducted at “optimally fluoridated water levels (i.e. between 0.7-1.2 mg/L)” – not at the elevated levels leading to “high prenatal exposure” that FFNZ falsely (and “egregiously”) asserts. But the key assertion by the NZ fluoridation review is that “the effect appeared to be driven by the minority of participants that had much higher fluoride exposures.”
Outliers lead to false conclusions
It’s quite simple really. Even within a group exposed to levels of fluoride expected with CWF there can be some individuals who receive higher exposes (f0r instance through consumption of fluoridated toothpaste or industrial pollution).
Looking at the data in the Canadian study in the image below taken from Green et al (2019) we can see that while most data points are clustered together at urinary F concentrations less than 1 mg/L there are a few data points at high urinary F concentrations and these do appear to drive the relationship they report – particularly for boys.
For the more statistically inclined reader, the table below summarises the relationships obtained by linear regression analysis. While the authors reported a statistically significant relationship for all the urinary fluoride concentrations up to 2.5 mg/L when the four high-end outliers (> 2.0 mg/L) are removed there is no significant relationship.
So I think the suggestion of the updated NZ fluoridation review is quite correct. The effect reported by Green et al (2019) is driven by just a few outliers and there is no statistically significant relationship when those four outliers are removed. That gives a false impression of the effect of CWF and in fact, their data shows absolutely no difference between IQ in fluoridated areas and unfluoridated areas.
Note 1: There is a discrepancy in the first table between the relationship reported by Green et al (2019) and that based on digitally extracted data points. Unfortunately, only 82% of the claimed data points could be extracted which is strange as usually close to 100% of data points can be extracted. Other commenters have reported the same problem. So it appears the authors have not included all their data in the figures and they have so far refused to make their data available for independent statistical analysis.
Although the Green et al (2019) paper did not cite R-squared values in her Master thesis did cite an R-squared value of 0.049 for boys. The low R-squared values (meaning the inclusion of the coefficient explains at most only a few per cent of the variation) and relatively high regression standard errors suggest that the reported coefficients are meaningless (they can be ignored in any model) – even if statistically significant.
Note 2: In case anyone suggests I have neglected the FFNZ reference to the Mexican study. That study took place in an area of endemic fluorosis and the authors have no record of the water fluoride levels mothers were exposed to. Bashesh et al (2017) reported:
“By virtue of living in Mexico, individuals participating in the study have been exposed to fluoridated salt (at 250 ppm) and to varying degrees of naturally occurring fluoride in drinking water. Previous reports, based on samples taken from different urban and rural areas, indicate that natural water fluoride levels in Mexico City may range from 0.15 to 1:38 mg/L. Mean fluoride content for Mexico City’s water supply is not available because fluoride is not reported as part of water quality control programs in Mexico.
Despite this, the Bashash study is often unjustly included with studies from areas of CWF by coauthors of Bashash and Green (see for example Farmus et al 2001). Anti-fluoride activists almost always make this mistake. Sure, they may attempt to justify their treatment of Bashash et al (2017) as relevant to CWF based on urinary fluoride values. But these a subject to so much variation and usually involve different collection and correction methods making comparison unjustified.