Comment by newtonsmethod
10 days ago
I agree, the effects of fluoride probably aren't in the top 5 things to be concerned about (although perhaps they are from a political perspective, with it becoming such a strong topic of debate for a variety of reasons). But do you assume that getting to n=10,000 is going to show little or no effect (e.g. having a level you define as little effect)? I'm not convinced the NTP data is extremely high quality and can't make much conclusion from it on the effects.
Also, for other commenters: the 2832 children number I believe comes from the supplemental content from the supplemental material for the NTP Fluoride Monograph: https://cdn.jamanetwork.com/ama/content_public/journal/peds/... (this url is very long because of some hashing measure, sorry: if it is no longer accessible, it is the supplemental content for doi:10.1001/jamapediatrics.2024.5542), on page 51 of the PDF. I have a small summary table of data I view relevant here:
The columns are:
* Studies used; Fluoride Exposure; Number of Studies / Number of Observations (number of Children)
* Estimate for slope in linear Model, given as increase in IQ points per mg/L increase (95% CI) (p value)
All studies; < 2mg/L; 8 / 10 (N = 3682); -0.18 (-0.40, 0.03) (p = 0.096)
All studies; < 1.5mg/L; 7 / 7 (N = 2832); 0.05 (-0.36, 0.45) (p = 0.816)
Low risk of bias studies; < 2mg/L; 4 / 5 (N = 1632); -0.33 (-0.53, -0.13) (p = 0.001)
Low risk of bias studies; < 1.5mg/L; 3 / 3 (N = 879); -0.32 (-0.91, 0.26) (p = 0.276)
> But do you assume that getting to n=10,000 is going to show little or no effect (e.g. having a level you define as little effect)?
I don't necessarily assume this, but even if you assume linearity, 0.7 * -0.32 is pretty dang small at baseline. I think if you generate n=10000 showing MLE=-.32 or +.05, few people are going to change their minds. I might, but I don't think it does much to shape the debate.