25 Is water fluoridation harmful?
This is the most consequential chapter in the book to get right, and the easiest to get wrong in either direction.
If I understate the evidence on harms, I am doing exactly what this book accuses guidelines of doing. If I overstate it, I hand ammunition to a campaign that has spent seventy years making claims the evidence does not support, and the foreseeable result is children with more tooth decay.
So I am going to be slower and more explicit here than anywhere else, and I am going to separate three questions that are routinely mashed together:
- Does fluoride at any dose harm neurodevelopment?
- Does fluoride at the concentrations used in water fluoridation harm neurodevelopment?
- What about fluorosis, which is the one harm nobody disputes?
The answers are different, and the difference is the whole chapter.
Fluorosis: the undisputed harm
Start with the one everyone agrees on, because it is a useful calibration.
The Cochrane review reports, carrying figures forward from its 2015 version (1):
| At a fluoride level of 0.7 ppm | Proportion affected | 95% CI | Studies |
|---|---|---|---|
| Fluorosis of aesthetic concern | ~12% | 8% to 17% | 40 |
| Fluorosis of any level | ~40% | 35% to 44% | 90 |
Roughly one in eight people examined in a population at 0.7 ppm has fluorosis that someone judged to be of aesthetic concern.
That is a prevalence, not an attributable harm, and an earlier draft of this chapter set it against the caries benefit as though the two were a balance sheet. They are not commensurable, for four separate reasons, and the error is worth spelling out because the comparison is made constantly in public argument.
A prevalence at an exposure is not the difference in prevalence between two policies. Some fluorosis occurs without fluoridated water, from toothpaste and other sources, so the excess caused by fluoridation is smaller than 12% by an amount this table does not report. The two numbers also describe different teeth in different people at different times: fluorosis is a permanent-dentition outcome recorded once enamel has formed, while the 0.24 figure in Chapter 24 is a baby-tooth caries increment. And they are different kinds of quantity, a proportion of people against a count of teeth, so placing them side by side and calling one bigger is arithmetic without meaning.
Two further qualifications, in both directions. “Of aesthetic concern” is a judgment made by examiners, not by the people whose teeth they are, and mild fluorosis is frequently invisible to its owner; nobody has asked the population how it weighs a faint white fleck against a filling. These figures are also low certainty, they were not updated in the 2024 review, which did not search for new fluorosis evidence, and the same review was “unsure” about skeletal fluorosis, bone fractures and skeletal maturity, at very low certainty from five studies.
So what can be said is this: a common, mostly cosmetic, mostly mild finding is more prevalent where water is fluoridated, and the caries benefit in contemporary populations is small and uncertain. Setting those two beside each other is a prompt for a decision analysis nobody has published, not a conclusion. I do not have the numbers to say which way it comes out, and neither does anyone who quotes 12% against 0.24.
Neurodevelopment: what the evidence says
Two documents dominate this argument. I have read both.
The US National Toxicology Program monograph (2) concluded, after an unusually contested review process, that higher fluoride exposure is associated with lower IQ in children. Its confidence statement is deliberately bounded, and the boundary is the point: the association is drawn from studies of exposures above the concentrations used in water fluoridation, principally areas of naturally high fluoride, and the monograph does not conclude that fluoridation at Western levels lowers IQ.
The JAMA Pediatrics meta-analysis (3) pooled 74 cross-sectional and prospective cohort studies. Fifty-two were rated at high risk of bias and 22 at low risk.
First the size of the thing, because a chapter that discusses whether an association exists without ever saying how big it is has broken this book’s own rule (Chapter 4). Across 59 studies with group-level exposure measures, the pooled effect was SMD −0.45 (95% CI −0.57 to −0.33). In the 13 studies with individual-level measures, IQ fell by 1.63 points (95% CI −2.33 to −0.93) per 1 mg/L increase in urinary fluoride, and by 1.14 points (−1.68 to −0.61) among the low risk-of-bias studies. Those are not trivial numbers at a population level, and they are drawn overwhelmingly from exposures well above anything a Western scheme delivers.
Then the review did the thing that matters, which is to restrict by exposure level. Here is its Key Points summary, and I am quoting the whole sentence because in an earlier draft I stopped it one clause early, which changed its meaning in my favor:
For fluoride measured in water, associations remained inverse when exposed groups were restricted to less than 4 mg/L or less than 2 mg/L but not when restricted to less than 1.5 mg/L; for fluoride measured in urine, associations remained inverse at less than 4 mg/L, less than 2 mg/L, and less than 1.5 mg/L; and among the subset of low risk-of-bias studies, there were inverse associations when exposed groups were restricted to less than 4 mg/L, less than 2 mg/L, and less than 1.5 mg/L for analyses of fluoride measured both in water and in urine.
— Taylor and colleagues, 2025 (3)
UK water fluoridation schemes target 1 mg/L. England’s is 1 mg/L; the World Health Organization’s guideline value is 1.5 mg/L. So the rows below 1.5 mg/L are the ones that bear on British policy, and there are two of them that disagree.
| Exposure, water fluoride | Studies | β (95% CI) | P |
|---|---|---|---|
| All studies, < 2 mg/L | 8 | −0.18 (−0.40 to 0.03) | .10 |
| All studies, < 1.5 mg/L | 7 | 0.05 (−0.36 to 0.45) | .82 |
| Low risk-of-bias studies, < 2 mg/L | 4 | −0.33 (−0.53 to −0.13) | .001 |
| Low risk-of-bias studies, < 1.5 mg/L | 3 | −0.32 (−0.91 to 0.26) | .28 |
Read the bottom two rows against the two above them. In the full, mixed-quality set, the estimate at under 1.5 mg/L sits essentially on zero. Restricted to the studies at low risk of bias, the point estimate at the same threshold is −0.32, almost identical to the −0.33 at under 2 mg/L that is statistically significant; it fails to reach significance on three studies and 879 children, with an interval running from −0.91 to 0.26.
That is not the same as the association disappearing, and I said it was. A null result on three studies is an absence of evidence, and telling you otherwise would be the exact error this book was written to object to, committed on the side I happen to find reassuring.
The review’s own conclusion is the careful version: there were “limited data and uncertainty in the dose-response association between fluoride exposure and children’s IQ when fluoride exposure was estimated by drinking water alone at concentrations less than 1.5 mg/L.”
For fluoride measured in urine, the inverse association persists below 1.5 mg/L and survives restriction to low risk-of-bias studies. Urinary fluoride reflects total intake from all sources: water, tea, toothpaste swallowed, food, air. It is a better measure of what actually got into the child, and a worse measure of what the water supply contributed.
So the honest statement of the crux is this. At fluoridation-relevant concentrations the water-based analyses are imprecise and inconsistent between the full and the low-bias subsets, and the urine-based analyses are inverse. It could mean water concentration is too crude a proxy and the urinary finding is the real one. It could mean the urinary studies are confounded by whatever else drives total fluoride intake. Nobody has settled it, and anyone telling you it is settled, in either direction, is not describing this literature.
Three limits on what I can say, stated rather than concealed.
These are observational studies. 74 of them, cross-sectional and cohort. The same confounding problem that runs through Chapter 5 applies here with more force: areas with naturally high fluoride differ from other areas in nutrition, iodine status, lead exposure, socioeconomic position and much else, all of which affect childhood IQ. That inverse associations persisted among the subset of low risk-of-bias studies is a meaningful robustness check, and it is not the same as randomization.
IQ is a difficult outcome. Measured with different instruments across countries and cultures, pooled as standardized mean differences, in a literature where publication of positive findings is more likely.
This is outside my competence. I am a dentist and a methodologist, not a neurotoxicologist. I can appraise a meta-analysis’s structure and read its restriction analyses. I cannot independently evaluate the toxicology, the biological plausibility at low doses, or the adequacy of the exposure assessment. Where this book’s other chapters give you my judgment, this one gives you the literature’s own stratification and stops.
What I will not do is convert a non-significant water-based estimate below 1.5 mg/L into “fluoridation is proven safe”. That is absence of evidence again, and it would be the same error in the opposite direction from the one I usually catch. It would also require ignoring that the low risk-of-bias subset puts the point estimate at that threshold in the same place as the significant one just above it. The honest statement is that at fluoridation concentrations the water-based evidence is too thin and too inconsistent to support a conclusion either way, the urine-based evidence is inverse, and the reason for the discrepancy is unknown.
What follows for policy, and what does not
Here is where I think the reasoning has to land.
The benefit is small and the fluorosis finding is real but mostly cosmetic, and the two are not directly comparable quantities. Post-1975, 0.24 dmft with an interval crossing zero, alongside roughly 12% fluorosis of aesthetic concern. Reasonable people can weigh those differently. That is a values question and I am not going to pretend the evidence resolves it.
The neurodevelopmental question does not currently justify stopping fluoridation, and does justify studying it properly. At fluoridation-relevant concentrations the water-based analyses are imprecise and point different ways depending on which studies are included; the urinary analyses are consistently inverse. Neither supports a policy change on its own, and both deserve serious, well-funded, prospective work in fluoridated Western populations rather than dismissal. The National Toxicology Program was explicit that its monograph “does not address whether the sole exposure to fluoride added to drinking water in some countries (that is, fluoridation, at 0.7 mg/L in the United States and Canada) is associated with a measurable effect on IQ” (2). That question has not been answered by anyone.
Nothing here should change what you do at home. Everything in this chapter is about fluoride in water, at a population level, swallowed. The recommendation with the strongest evidence in this book, Chapter 8, is about fluoride in toothpaste, applied to the tooth surface and spat out. The whole point of Chapter 7 is that you are not meant to swallow it. Someone who reads this chapter and stops using fluoride toothpaste has read it backwards.
PubMed, 21 August 2026. I have the full text of the NTP monograph and of the JAMA Pediatrics meta-analysis with its supplementary material, supplied during writing. Fluorosis figures are from the Cochrane review, which did not update its fluorosis searches for the 2024 version.
This is the least independent chapter in the book. I have not appraised the 74 included studies, I have not assessed exposure measurement, and I have relied on the reviews’ own stratified analyses rather than reconstructing them. That is a deliberate limit, stated here rather than implied.
Full record in appraisals/searches/ch25-fluoridation-harm.md.
Verdict
- Certainty of evidence
- Fluorosis: low, and not updated since 2015. Approximately 12% of aesthetic concern at 0.7 ppm, which is a prevalence under an exposure and not the excess caused by adding fluoride to water. Neurodevelopment above fluoridation levels: an association reported across 74 observational studies, robust to restricting to low risk-of-bias studies. Neurodevelopment at fluoridation levels: very low and genuinely unresolved. Water-measured fluoride below 1.5 mg/L gives β 0.05 (−0.36 to 0.45) across all studies but β −0.32 (−0.91 to 0.26) in the low risk-of-bias subset, on three studies. Those coefficients are on the review’s standardized scale, not IQ points, and both intervals include zero, so neither subset establishes an effect in either direction. Urine-measured fluoride below 1.5 mg/L remains inverse, and a urinary concentration is a measure of total intake from all sources rather than a restatement of the concentration in the water. The NTP monograph declines to address 0.7 mg/L fluoridation and IQ at all. Skeletal harms: very low, five studies, uncertain.
- Directness to the advice as worded
- Poor, and this is the central difficulty. Most of the neurodevelopmental evidence comes from regions of naturally high fluoride, at concentrations above those any Western scheme uses. Applying it to 1 mg/L schemes is exactly the indirectness this book objects to elsewhere, and I decline to do it.
- Is the strength label defensible?
- Not applicable: this is a policy question rather than a DBOH recommendation. My position is that the current evidence does not support stopping fluoridation on neurodevelopmental grounds, does support taking the question seriously enough to fund proper prospective research in fluoridated populations, and that the benefit-versus-fluorosis trade-off is closer than public debate usually allows.
- What would change my mind
- A prospective cohort in a fluoridated Western population, with individual urinary fluoride measured repeatedly through pregnancy and childhood, adjusting for lead, iodine and socioeconomic position, with a standardized cognitive outcome. That study would resolve the water-versus-urine discrepancy, and it is the study this field needs. Conversely, an explanation of that discrepancy showing the urinary finding to be confounded would move me the other way.
What this means for you
Fluoride in your toothpaste is the best-supported recommendation in this book and nothing in this chapter touches it. Use it, spit it out, do not swallow it.
Fluoride in your water, if you live somewhere fluoridated, is contributing a small and uncertain benefit and carries a real chance of mild cosmetic marking of the enamel.
I am not going to tell you what to drink. An earlier draft of this chapter suggested that a worried parent might reasonably switch to filtered or bottled water, and I have removed it. Two reasons, and both are the chapter’s own argument turned on its author. The evidence at fluoridation concentrations is unresolved, so a recommendation either way would be resting on something I have just spent nine pages saying nobody knows. And the suggestion was not even accurate on its own terms: bottled waters vary widely in fluoride and some carry more than a fluoridated supply, while most domestic filters do not remove fluoride at all. If this worries you, the person to ask is your dentist or GP, who knows your water supply and your children.
What I would not do is stop using fluoride toothpaste, and what I would not tell you is that this question is settled. It is not. The people who tell you it is, on either side, are describing something other than the evidence.