24 Does water fluoridation work?
Adding fluoride to public water is the most politically charged thing in this book. It is also, unusually, a question where the evidence changed recently and in a direction that neither side of the argument expected.
I am going to give the answer plainly first, because the rest of this chapter is detail and the detail gets quoted selectively.
Water fluoridation probably still works, and it works considerably less well than it used to. Not because the chemistry changed, but because fluoride toothpaste became universal in the mid-1970s, and you cannot get much extra benefit from putting fluoride in the water of people who are already putting it on their teeth twice a day.
That claim is not mine. It is the conclusion of the 2024 update of the Cochrane review (1), and it is the reason this chapter exists.
What the review found
157 studies, all non-randomized, because you cannot randomize a water supply. The review’s key methodological decision, and the one that changed the picture, was to stratify the results by whether data were collected before or after 1975, the point at which fluoride toothpaste became widespread.
Contemporary evidence, meaning after 1975:
| Outcome | Effect | 95% CI | Studies | Certainty |
|---|---|---|---|---|
| Change in dmft (baby teeth) | MD 0.24 | −0.03 to 0.52 | 2 | low |
| Proportion caries-free, primary teeth | MD −0.04 | −0.09 to 0.01 | 2 | low |
| Proportion caries-free, permanent teeth | MD −0.03 | −0.07 to 0.01 | 2 | low |
| Change in DMFT (adult teeth) | very uncertain | — | 4 | very low |
| Effect of cessation | could not determine | — | 1 | very low |
A difference of 0.24 dmft is about a quarter of one decayed, missing or filled baby tooth per child. And the confidence interval runs from −0.03 to 0.52, which includes no benefit at all. The caries-free proportions work out at about 4 and 3 percentage points respectively, again with intervals crossing zero.
The review’s own summary of this: initiation “may lead to a slightly greater reduction in dmft” and “may lead to a slightly greater increase in the proportion of caries-free children, but with smaller effect sizes than pre-1975 studies.”
For the older evidence, from 1975 or earlier, the review was “unsure of the size of effects” on every outcome, at very low certainty. So the famous large benefits everyone cites come from a body of evidence the reviewers could not rely on either.
Reading this honestly, in both directions
This is the chapter where selective quotation does the most damage, so let me set out what each side gets right and wrong.
What the critics get right. The modern effect is small, the confidence interval includes zero, the evidence is low certainty and non-randomized, and there are only two contemporary studies for the primary outcome. Anyone who tells you contemporary evidence shows a large benefit is not describing this review. And the review includes no eligible studies reporting caries outcomes in adults, which is remarkable for an intervention delivered to entire adult populations.
That is a statement about the review’s eligibility criteria, not about the literature, and an earlier draft of this chapter let it slide into the second. The review asks what happens when fluoridation is started or stopped, so a study of people living under continuous exposure cannot qualify however large it is.
One such study exists and it is very large. LOTUS matched 6,370,280 NHS dental patients aged 12 and over in England on propensity scores and compared those living with water at 0.7 mg F/l or above against those without, over 2010 to 2020 (2). It found a real but very small difference: 3% fewer invasive treatments per person, incidence rate ratio 0.969 (95% CI 0.967 to 0.971), which is 0.173 fewer fillings and extractions per person across ten years. Decayed, missing and filled teeth were 2% lower, a difference of 0.212 teeth (95% CI 0.229 to 0.194 lower). The number of missing teeth per person did not differ (0.006, 95% CI −0.008 to 0.021).
The authors are unusually direct about what that means, and the sentence deserves quoting because it is the kind of thing effect estimates rarely get:
This magnitude of effect is smaller than what most stakeholders we engaged with (50/54) considered meaningful.
— Moore and colleagues, LOTUS, 2024 (2)
They also found “no compelling evidence that water fluoridation reduced social inequalities” in treatments or missing teeth.
This is observational and the authors say so: fluoridation status and outcomes can both be misclassified, and residual confounding cannot be excluded from a natural experiment on routine claims data. A sample of six million does nothing to fix confounding; it only makes the confidence interval narrow, which is why the interval here is absurdly tight around a number nobody thinks matters. But it answers a question the Cochrane review was not asking, in adults, in England, in the era of fluoride toothpaste, and the book has no business pretending it does not exist.
What the critics get wrong. Every estimate points the same direction, toward benefit. “The confidence interval includes zero” is not “the effect is zero”; it is also compatible with half a tooth per child, which across a population is not nothing. The evidence for stopping fluoridation is very low certainty from a single study, so anyone claiming that cessation is demonstrably safe is doing the thing this book objects to, in reverse.
What defenders get wrong. Continuing to cite pre-1975 effect sizes, or the often-repeated “25% reduction”, as though the 2024 update had not happened. It did happen, it was conducted by the same Cochrane group whose fluoride toothpaste review underpins Chapter 8, and its finding is that the benefit has shrunk.
Every study here is non-randomized, so under GRADE the body starts at low certainty, and the review downgraded further for confounding and lack of blinded outcome assessment. It also downgraded older evidence for indirectness, on the grounds that pre-1975 societies are not applicable to contemporary ones. That is exactly the reasoning Chapter 11 applies to sealant trials from the 1970s, and it is the right move.
Could randomization ever happen? Not realistically. You would need to allocate water supplies to communities and follow them for years, with people moving in and out. The review notes migration as a specific complication.
So this is a permanent state of affairs: a population intervention that cannot be tested to the standard we would demand of a toothpaste. That is a reason for humility on all sides, not a trump card for either.
The equity question is unresolved. Fluoridation’s strongest argument has always been that it reaches people who do not brush regularly, narrowing the gap between rich and poor. The review found only one post-1975 study reporting disparities by socioeconomic status, “with no evidence that deprivation influenced the relationship”. One study is not enough to confirm or refute the central policy claim for the intervention. That gap should trouble supporters more than it troubles critics.
PubMed, 21 August 2026, for updates to CD010856. The 2024 version (.pub3) remains current. Its own search ran to 16 August 2023 across CENTRAL, MEDLINE, Embase and four other databases, with two trial registers and backward citation searching.
I have the full text of this review, supplied during writing, including the characteristics of the included studies.
Note the review did not search for new evidence on fluorosis for this update; those figures are carried forward from the 2015 version. That matters for Chapter 25.
Full record in appraisals/searches/ch24-fluoridation.md.
Verdict
- Certainty of evidence
- Low that initiating community water fluoridation slightly reduces caries in children’s baby teeth in contemporary populations. Very low for permanent teeth in contemporary populations, for the effect of cessation, and for all pre-1975 estimates. No evidence in adults for starting or stopping fluoridation, which is the question the review asks. For continuous exposure in adults there is LOTUS, a very large matched observational study whose central estimate is 0.173 fewer invasive treatments per person over a decade, an effect its own authors report as smaller than most of their stakeholders considered meaningful (2).
- Directness to the advice as worded
- Good on the intervention, poor on population coverage. These are studies of communities receiving fluoridated water, which is the intervention. But no study reports adult outcomes, and only one addresses the socioeconomic disparity question that is fluoridation’s main policy justification.
- Is the strength label defensible?
-
Not applicable, and that is itself worth noting. Water fluoridation is not among the 91 recommendations in Delivering Better Oral Health’s summary tables; it is policy set elsewhere, and the guideline discusses it in its fluoride chapter rather than recommending it to clinicians.
So this chapter audits a policy rather than a recommendation. My conclusion is that the contemporary evidence supports a small benefit at low certainty, which is a considerably weaker basis than the public debate on either side assumes.
- What would change my mind
- More contemporary studies, particularly any reporting adults, and any reporting outcomes by socioeconomic position. Two studies is a thin basis for a national policy in either direction. A well-conducted study of a cessation event, of which several have occurred, would be more informative than anything else available and would be relatively cheap to do.
What this means for you
If you live in a fluoridated area, the water is contributing something to your teeth, and on the best current evidence it is contributing less than it did to your grandparents, because you have fluoride toothpaste and they did not.
If you live in an unfluoridated area, the thing that protects your teeth is in Chapter 8, and it is on your bathroom shelf.
If you are arguing about fluoridation policy, the honest position is that the contemporary benefit is small and uncertain, that the evidence for stopping is weaker still, and that the equity claim which motivates the whole policy rests on a single study. Whether a small, uncertain, population-wide benefit justifies the intervention is a question about values and costs, not a question the evidence settles. Chapter 25 deals with the other side of the ledger.