For part 1 of this series see Anti-fluoridation propaganda now relies on only four studies. 1: Bashash et al (2018).
Paul Connett, director of the Fluoride Action Network (FAN), now claims “You only have to read four studies…” to come to the conclusion that community water fluoridation (CWF) is bad for your health. As I said in the first article in this series that is simply bad science. One should not ignore all the other relevant studies – and anyway, these four studies do not say what Connett claims.
In this article, I discuss the second study Connett recommends. It’s citation is:
Green, R., Lanphear, B., Hornung, R., Flora, D., Martinez-Mier, E. A., Neufeld, R., … Till, C. (2019). Association Between Maternal Fluoride Exposure During Pregnancy and IQ Scores in Offspring in Canada. JAMA Pediatrics, 1–9.
Green at al (2019)
According to Connett:
“The second* came in 2019 when a study published in JAMA Pediatrics essentially replicated the Mexico City finding in Canadian communities.”
Table 2 summarises the results obtained by Green et al (2019). Let’s compare them with the results found by Bashash et al (2018) as presented in the first article in this series – Anti-fluoridation propaganda now relies on only four studies. 1: Bashash et al (2018).
Note: See Table 1 for an explanation of symbols and bars
No, Green et al (2019) did not “replicate” the findings of Bashash et al (2017) – although Connett may be using the word “essentially” to cover up his over-confidence in the claim.
Unlike Bashash et al (2017), Green et al (2019) did not find a statistically significant relationship of child IQ (FSIQ) with MUF. However, when they separated the children by sex the relationship was significantly negative for males (positive, but not statistically significant, for females). They also reported statistically significant relationships of IQ with maternal F intake estimated using an unvalidated subjective method, but not with drinking water F.
When the IQ (FSIQ) was separated into subsets no significant relationship was found for verbal IQ (VIQ) but there was a significant relationship with MUF of performance (PIQ) for boys. There was also a significant relationship of PIQ with drinking water F. So quite a mixed bag – and perhaps indicating that separating the data into different groups based on sex and using different cognitive measures can tweak out significant relationships. But it reminds me of the old saying that one can get the answer one wants if the data is tortured enough.
As for the statistically significant relationships reported by Green et al (2019) – none of them are at all “strong” as can be seen in these figures taken from the paper:
Antifluoride campaigners use of this paper in their propaganda relies on a very weak relationship which, according to Green (2018) explains only 4.7% of the variance in IQ, and required separation of children by sex to get statistical significance.
Polishing data and ignoring non-significant relationships
A concern I have about this study is the differences in the findings reported in the original MA thesis of Green (2018) and the final paper of Green et al (2019). For example, the thesis reported an adjusted association of FSIQ with fluoride intake (B = -3.82, 95% CI: -7.65 to 0.02, p = .05) which she describes as having “just missed significance” while the final paper reports this association as significant (B = -3.66, 95% CI: -7.16 to -0.15, p = 0.04). What was done to move this association into statistical significance?
I am also concerned that not all the non-significant associations recorded in the thesis are reported in the final paper. For example, Green (2018) reported that neither VIQ or PIQ was significantly associated with fluoride intake and VIQ was not significantly associated with water fluoride concentration and these facts were not reported in Green at al (2019). On the other hand, the significant relationship of PIQ with water fluoride concentration was reported in the final paper.
Selectively reporting results of statistical analyses like this gives a false impression of how important the results may be in practice. Sure, I can understand why authors will bias their presentation in this way but a good peer review should identify this bias and insist on the presentation of complete results.
Oh, and the overall comparison of child IQ from areas of residence of their mothers during pregnancy did not show any statistically significant difference due to fluoridation in Green et al (2019) – see Table 3.
Connett is wrong to claim that the Green et al (2019) study “replicated” the Bashash et al (2017). It didn’t by a long shot. The study itself is also very weak, it has several faults and has been widely criticised in the scientific community.
Tomorrow I will discuss the third study Connett now relies on – Riddell et al (2019) – see Anti-fluoridation propaganda now relies on only four studies. 3: Riddell et al (2019).
- Experts complain to funding body about quality of fluoride-IQ research
- Biostatistical problems with the Canadian fluoride/IQ study
- More expert comments on the Canadian fluoride-IQ paper
- Anti-fluoridationists put faith in new “strong” studies to provide evidence missing in draft NTP review
- Fluoridation science and political advocacy – who is fooling who?
- Scientific integrity requires critical investigation – not blind acceptance
- Fluoridation – A new fight against scientific misinformation
- Politics of science – making a silk purse out of a sow’s ear
- If at first you don’t succeed . . . statistical manipulation might help