Professor Christine Till has been given a $300,000 US National Institute of Health grant to test for harmful effects of fluoride.
Malin and Till (2015) published research indicating a relationship between fluoridation and Attention Deficit Hyperactivity Disorder (ADHD). However, that study was flawed because it omitted important confounders. When these are included the relationship disappears.
I analysed that study in my article ADHD linked to elevation not fluoridation where I showed the relationship of ADHD to elevation was much more important than fluoridation. Huber at al., (2015) published work confirming the relationship of ADHD with elevation. So, obviously, elevation is an important confounder and Malin and Till (2015) did not consider it in their study.
My own analysis indicated that there were a number of other confounders which are related to ADHD – with correlations similar to (eg., educational attainment, proportion of the sate’s population older than 65 and Per Capita personal income) or better (mean state elevation, home ownership and % living in poverty ) than that for fluoridation. That rings alarm bells – why consider only one factor (fluoridation) if there are other factors which appear equally or more important? Isn’t that confirmation bias? (I concede that Malin and Till did include a socioeconomic measure in their statistical analysis – but this was clearly not enough).
I tested the relative importance of the different facts using multiple regression and – sure enough – found that once a few important confounders were included water fluoridation could not explain any of the variance in ADHD! The statistically significant factors were mean elevation, home ownership, and poverty. The contribution of fluoridation was not statistically significant in this multiple regression.
A model including mean state elevation, home ownership and poverty explains about 45% of the variance in ADHD – much better than fluoridation could (Malin and Till explained 27 -32% for the fluoridation data).
Now, I read that Professor Till has been given research finds to have another go and possible harmful effects of fluoride. (see York professor leads study that could help answer fluoride safety questions). She plans to look at data from a Canadian investigation of pregnant women exposed to contaminants. She says:
“Our study employs a prospective design that includes biomarkers of exposure to fluoride, detailed assessment of potential confounders, a comparison group, and the use of sensitive cognitive and behavioural measures that have been collected in one of the world’s most comprehensively characterized national pregnancy cohorts (MIREC).”
Now, I am pleased she aspires to a “detailed assessment of potential confounders” but wonder how detailed this will be after the problems with the Malin and Till (2015) study.
I have not yet seen any published response to the Malin and Till paper – maybe the cost of publication (US$2020) that journal charges is discouraging critics. It certainly discouraged me (I do not have institutional support for publication costs). Nevertheless, I hope professor Till has been acquainted with some of the criticism of that paper so that she can pay more attention to important confounders in the coming work
We can draw a few lessons from this.
Be careful of published statistical relationships
These days it is so easy to hunt down data and do this sort of exploratory statistical searching for significant relationships. But a statistically significant relationship is not evidence of a real cause. For example, there is a strong relationship between the sales of organic produce and prevalence of autism – but I have yet to hear anyone seriously suggest the relationship is at all causal.
But the scientific literature is still full of such studies – and I guess the motivated author can easily find arguments and other data in the literature that they, at least, feel convincing enough to justify publication.
Refereeing of scientific papers is, on the whole, abysmal
All authors have a pretty good idea of which journals, and reviewers, will be friendlier to their work – and which would be antagonistic. It is only natural tosubmitt to the friendlier journal.
Unfortunately, the Malin and Till paper was submitted to a journal with editors known to be friendly to a chemical toxicity model of cognitive deficits. Further, it turns out that the reviewers chosen for the paper were also supportive of such an approach.
While one reviewer did suggest including lead as a possible confounder (again showing a chemical toxicity bias) none of them suggested consideration of other confounders more likely to be connected with ADHD.
I discussed the editorial and reviewer problems of the Malin and Till paper in Poor peer-review – a case study. (The journal, Environmental Health, has a transparent peer-review process which provides access to the names and reports of the reviewers.)
Again – another example of readers beware – even readers of scientific papers in credible journals.