Fluoridation and ADHD: A new round of statistical straw clutching

“To clutch at straws – the act of reaching for a solution no matter how irrational or inconsequential.” Source: Advanced Vocabulary for English Language Learners

Anti-fluoridation activists are promoting a number of new scientific papers they argue support their campaigns. But one has only to critically read these papers to see they are clutching at straws. Their promotion relies on an unsophisticated understanding of statistics and confirmation bias.

I will look at one paper here – that of Bashash et al., (2018) which reports an association between maternal prenatal urinary fluoride and prevalence of child ADHD.

The paper is:

Bashash, M., Marchand, M., Hu, H., Till, C., Martinez-Mier, E. A., Sanchez, B. N., … Téllez-Rojo, M. M. (2018). Prenatal fluoride exposure and attention deficit hyperactivity disorder (ADHD) symptoms in children at 6–12 years of age in Mexico City. Environment International, 121(August), 658–666.

I discussed an earlier paper  by these authors – Bashash et al., (2016) which reported an association between maternal neonatal IQ fluoride and child IQ – (also heavily promoted by anti-fluoride activists) in a number of articles:

• Fluoride, pregnancy and the IQ of offspring,
• Maternal urinary fluoride/IQ study – an update,
• Anti-fluoridation campaigners often use statistical significance to confirm bias,
• Paul Connett “updates” NZ MPs about fluoride?,
• Paul Connett’s misrepresentation of maternal F exposure study debunked,
• Mary Byrne’s criticism is misplaced and avoids the real issues,
• A conference paper on the maternal neonatal urinary fluoride/child IQ study has problems, and
• A more convincing take on prenatal maternal dietary effects on child IQ.

Promotion of the new paper by anti-fluoride activists suffers from the same problems I pointed out for their promotion of the earlier paper. In particular it ignores the fact that the reported relationships (between maternal neonatal urinary fluoride and cognitive measure for children in Bashash et al., 2016, and prevalence of child  attention deficit hyperactivity disorder – ADHD – in Bashash et al., 2018) were very weak and explain only a very small amount of the variation. This raises the possibility that the reported weak relationships would disappear if significant risk-modifying factors were included in the statistical analyses.

Bashash, et al., (2018)

Whereas the earlier paper considered measures of cognitive deficits in the children the current paper considers various measurements related to ADHD prevalence among the children. These include parent rating scales (CRS-R). Three were ADHD-related scales from the Diagnostic and Statistical Manual of Mental Disorders (DSM) (Inattention Index, Hyperactivity-Impulsive Index and Total Index [inattentive and hyperactivity-impulse behaviours combined]). They also include several other indexes related to ADHD.

A number of computer-assisted indexes (CPT-II) were also determined.

Most indices were not significantly associated with maternal prenatal urinary fluoride. However, the authors reported statistically significant (p<0.05) relationships for indices of Cognitive Problems + Inattention, ADHD Index, DSM Inattention and DSM ADHD Total.

The data and the relationships were provided in graphical form – see figure below – taken from their Figure 2:

There is obviously a wide scatter of data points indicating that the observed relationships, although statistically significant, explain only a small part of the variation in the indices.

So, just how good are the relationships reported by Bashash et al., (2018) in explaining the variation in these ADHD-related indices? I checked this out by digitally extracting the data from the figures and using linear regression analysis.

Index	% Variance explained
Cognitive problems + Inattention	2.9%
ADHD Index	3.1%
DSM Inattention	3.6%
DSM ADHD Total	3.2%

In fact, these relationships are extremely weak – explaining only a few per cent of the observed variation in the ADHD related indices. This repeats the situation for the cognition-related indices reported on the Bashash et al., (2016) paper (see Maternal urinary fluoride/IQ study – an update).

The fact these relationships were so weak has two consequences:

1. Drawing any conclusions that maternal neonatal fluoride intake influences child ADHD prevalence is not justified. There are obviously much more important factors involved that have not been considered in the statistical analysis.
2. Inclusion of relevant risk-modifying factors in the statistical analysis will possibly remove any statistical significance of the relationship with maternal urinary fluoride.

Credible risk-modifying factors not considered

Bashash et al., (2108) do list a number of possible confounding factors they considered. These did not markedly influence their results. however, other important factors were not included.

Nutrition is an important factor. Malin et al., (2108) reported a signficant effect of nutrition on cognitive indices for a subsample of the mother-child pairs in this study (see A more convincing take on prenatal maternal dietary effects on child IQ).

Their statistical analyses show that nutrition could explain over 11% of the variation in child cognitive indices indicating that nutrition should have been included as a possible risk-modifying factor in the statistical analyses of Bashash et al., (2016) and Bashash et al., (2018). I can appreciate that nutrition data was not available for all the mother-child pairs considered in the Bashash et al., papers. However, I look forward to a new statistical analysis of the subset used by Malin et al., (2108) which includes prenatal maternal urinary fluoride as a risk-modifying factor and tests for relationships with child ADHD prevalence.

Could the reported weak relationship disappear?

Possibly. After all, it is very weak.

The problem is that urinary fluoride data could simply be a proxy for a more important risk-modifying factor. That is, urinary fluoride could be related to other risk modifying factors (eg. nutrition) so that the relationship with urinary fluoride could disappear when these other factors are included.

I illustrated this for a earlier reported relationship of child ADHD prevalence with extent of fluoridation in US states (see Perrott 2017 – Fluoridation and attention deficit hyperactivity disorder – a critique of Malin and Till (2015)). In  that case the relationship was much better than those reported by Bashash et al., (2016) and Bashash et al., (2018) – explaining 24%, 22% and 31% of the variance in ADHD prevalence for the years 2003, 2007 and 2011 respectively. The relationships are illustrated in their figure:

Relationships between water fluoridation (%) and child ADHD prevalence for 20013 (red triangles), 2007 (blue diamonds) and 2011 (purple circles). Malin & Till (2105)

Yet, when other risk-modifying factors (particularly mean state elevation) not considered by Malin & Till (2015) were included in the regression analyses there was no statistically significant influence from fluoridation prevalence. In this case fluoridation prevalence was related to altitude and was simply acting as a proxy for altitude in the Malin & Till (2015) regression.

Conclusion

As the authors admit, this study:

“was not initially designed to study fluoride exposure and so we are missing some aspects of fluoride exposure assessments (e.g., detailed assessments of diet, water, etc.).”

However, they do say these “are now underway” so I look forward with interest to the publication of a more complete statistical analysis in the future.

There are other problems with the data (for example the paucity and nature of the urinary fluoride measurements) and these are the sort of issues inevitably confronting researchers wishing to explore existing data rather than design experimental protocols at the beginning.

Readers should therefore always be hesitant in their interpretations of the results and the credibility or faith that they put on the conclusions of such studies. The attitude should be: “that is interesting – now let’s design an experiment to test these hypothetical conclusions.”

The problem is confirmation bias – the willingness to give more credibility to the findings than is warranted. Scientists are only human and easily succumb to such biases in interpreting their own work. But this is even more true of political activists.

The reported relationships are weak. Important risk-modifying factors were probably not included in the statistical analyses. The observed relationships may simply mean that urinary fluoride is acting as a proxy for a more important risk-modifying factor (like nutrition) and the weak relationship may disappear when these are considered.

So scientific assessment of this study will be extremely hesitant – interpreting it, at best, as indicating need for more work and better designed research protocols.

But, of course, political activists will lap it up. It confirms their biases. Political activist organisations like the Fluoride Action Network are heavily promoting this paper – as they did with the earlier Bashash et al., (2016) paper.

But they are simply clutching at straws – as they often are when using science (or more correctly  misrepresenting and distorting the science) to support their political demands.

Similar articles