Evidenced Health

Fluoride and Bone Health

8 min read

Orbiter of the fluoride cluster (nucleus: Fluoride and Health). This page holds the skeletal arm of the fluoride benefit/harm menu — the arm the nucleus named as a gap and cashed here from one gold dose-response meta-analysis. The decision object is chronic drinking-water fluoride -> fracture risk, with bone mineral density (BMD) kept deliberately separate because the two dissociate.

The landed finding: fracture risk rises above ~1.5 mg/L water fluoride

The first dose-response meta-analysis of this arm (37 studies 1945-2024; one-stage restricted cubic spline) finds a non-linear positive overall relation — fracture risk flat at low exposure, rising above a ~1.5 mg/L threshold (Mazzoli et al., 2025):

  • «At low levels of fluoride, specifically up to around 1.4 mg/L, there was no change in fracture risk (RR 1.00 at 1.2 mg/L; 95% CI 0.80–1.24). Above 1.4 mg/L … RR of 1.06 (95% CI 0.80–1.41) at 2.0 mg/L … 1.19 (95% CI 0.87–1.64) at 3.0 mg/L … 1.35 (95% CI 0.94–1.95) at 4.0 mg/L» (Mazzoli et al., 2025).
  • Studied range: ~0-4 mg/L drinking water (the pooled water-metric studies). Above 4 mg/L there is no pooled estimate — the curve’s apparent top is the sampling edge, not a located feature.
  • The threshold is not a curve knee first — it is where the data change: «Our data indicated little relation between fluoride exposure and overall fracture risk below 1.5 mg/L … with fracture risk starting to increase gradually above that threshold. The threshold level corresponds to around 3-3.5 mg/day» (Mazzoli et al., 2025). It coincides with the WHO 1.5 mg/L water guideline and the NTP «higher»-exposure referent for the IQ arm — the same exposure line recurs across two fluoride arms (see Fluoride and Health).

Every overall-fracture point estimate has a CI crossing 1.0. This is a positive trend, not a per-dose demonstrated harm; the certainty is low (below).

The one CI that excludes 1.0: postmenopausal (females >50) at ~1.0 mg/L

The decision-relevant sharpening is a stratum, not the pooled curve. In the pre-specified females-over-50 analysis the association starts earlier and reaches significance at a fluoridation-adjacent dose (Mazzoli et al., 2025):

  • «the RR was 0.98 (95% CI 0.83–1.14) at 0.25 mg/L, 1.00 (95% 0.80–1.24) at 0.45 mg/L, and 1.26 (95% CI 1.10–1.46) at 1.0 mg/L» (Mazzoli et al., 2025) — the only reported pooled DR estimate whose CI excludes 1.0 (every overall-fracture, both-sex fragility, male, and both-sex-female fragility point the paper tabulates has a CI crossing 1.0). The curve above 1.0 mg/L continues its «steep upward trend» but is not tabulated past that point.
  • Mechanism candidate (directional,-grade in the source’s own discussion): estrogen is bone-protective; menopause accelerates resorption and may increase «release of fluoride from bone stores», so the postmenopausal skeleton is a plausibly more susceptible support-factor profile (Mazzoli et al., 2025). This is a route-(a)/(c) stratification hypothesis (higher baseline bone-loss risk; possibly a stratum-specific harm), not a demonstrated effect-modification interaction — it rests on a single lab’s pooled observational data. Links to the sex-hormone bone lever on Menopause and the Shifting Levers.

BMD and fracture DISSOCIATE — a BMD rise is not a bone benefit here

The classic surrogate trap fires cleanly. Fluoride can raise bone mineral density at some sites while fracture risk also rises — so reading the BMD signal as benefit inverts the decision. The source states the mechanism of the dissociation itself:

  • «BMD only captures bone quantity (density), not bone quality (e.g., architecture, morphology, micro­damage, material properties). The latter aspects of bone health are crucial determinants of fracture risk» (Mazzoli et al., 2025).
  • The BMD dose-response is not even uniformly upward — it is site- and sex-divergent: «While hip BMD increase in both sexes, though linearly in males and non-linearly (starting above 1.2 mg/L) in females, for spine BMD there was an indication of an inverted U-shaped association both in males and in females, the highest value being around 0.6 mg/L» (Mazzoli et al., 2025). Hip-BMD mean differences in the better-RoB female stratum have CIs crossing 0 (+0.04 g/cm², 95% CI −0.03 to 0.12).
  • Decision consequence: BMD is a bad surrogate for the fracture outcome for this exposure — a rise does not transmit to lower fracture risk, and here co-occurs with higher fracture risk. Steer on the fracture endpoint, not the density marker -> Surrogate Outcomes.

The U/J shape: the UPPER arm is believed, the protective LOWER arm is NOT

The sex-specific and site-specific fragility curves are U/J-shaped (an apparent protective dip at low-moderate fluoride, then a rise): «a clear evidence of a U-shaped curve only in females, with the lowest risk around 0.4 mg/L and a monotonic increase above 0.9 mg/L … RR was 0.84 (95% CI 0.67–1.07) at 0.34 mg/L, 1.00 (95% 0.79–1.27) at 0.9 mg/L, and rose to 1.13 (95% CI 0.93–1.37) at 1.1 mg/L» (Mazzoli et al., 2025). Applying the artifact discipline (a protective/plateau arm must survive a strong check before belief):

  • Upper arm (BELIEVED, decision-relevant): fracture risk rising above ~1.5 mg/L overall, and from ~0.5-1.0 mg/L in postmenopausal females (RR 1.26 at 1.0 mg/L, CI excludes 1.0). This is the arm the recommendation rests on — and it survives because it is where the data are densest and the one significant stratum sits.
  • Lower / protective arm (NOT believed): the apparent RR<1 at low fluoride (the female «lowest risk around 0.4 mg/L»; the male «generally indicative of a decreased risk with increasing fluoride»). No strong check (referent-correction, Mendelian randomization, intervention) supports a true protective effect, and the source itself flags it as a likely artifact: «some types of fractures showed a U-shaped pattern … Such a pattern might be an artifact due to the more limited number of studies on which such [estimates rest]» (Mazzoli et al., 2025). The lowest-risk CIs all cross 1.0.
  • Mechanism candidates for the artifactual dip: unequal between-group precision (far fewer studies at low fluoride -> a spuriously displaced nadir — the gate-6 mechanism), sparse-data instability, and the general fragility of a spline nadir read off wide CIs. Never quote «lowest risk around 0.4 mg/L» as an optimum — it is a sampling artifact of a wide, sparse curve, not a derivable target -> The Underivable Optimum.

Certainty and limits — why this is low, single-lab, and internally weak

  • Internal weakness: every overall-fracture CI crosses 1.0; only the females>50 point at 1.0 mg/L is significant. Risk-of-bias (ROBINS-E) is mostly poor — for fractures, 18 of 27 estimates high-RoB, only 2 low; 16 of 37 studies are ecological designs; the dominant bias drivers are «unaccounted sources of confounding (n = 11) and bias in exposure measurement (n = 8)» (Mazzoli et al., 2025).
  • The signal weakens on RoB-restriction: excluding high-RoB studies drops the fragility-fracture RR to «0.93 (95% CI 0.72–1.21) at fluoride exposure in the 0.8–1.5 mg/L range» and «0.95 (95% CI 0.70–1.29) at 1.8 mg/L» (Mazzoli et al., 2025) — the main all-fracture analysis could not even be RoB-restricted (too few low-RoB studies). A signal that attenuates when the worst studies are removed is the direction that argues against a robust effect.
  • Measurement error: water-fluoride as an exposure proxy omits total intake (tea, dental products, diet) — the same attenuation-toward-null constraint that governs the IQ arm and every dietary curve -> Measurement Error in Dietary Assessment. A flattened low-dose arm is expected here regardless of a true effect.
  • Halo-across-a-lab (single backing): Mazzoli is the same Modena / CREAGEN / Vinceti group as the fluoride-IQ source (Veneri 2023). Different outcome, so this is not laundered independence — but one lab’s productivity is not independent evidential breadth. Independent replication of the skeletal arm by a different group is a named gap; confidence stays low until it lands.
  • Aggregation flavor: the pooled per-dose RR is a magnitude the wiki did not compute — it is [EXTRACTED] from a single MA, not an emergent aggregation across held sources (G (needs aggregation) for any cross-source pooling).

Why this is a decision-change, not a restatement

  • For someone on naturally high-fluoride water (>1.5 mg/L) — the ~0.59% of US residents on community water at/above that, and many private-well and endemic-fluorosis populations — the skeletal-harm arm is now live alongside the IQ-harm arm at the same exposure line: two patient-important outcomes (fracture, child IQ) both turn upward above ~1.5 mg/L, strengthening the case for mitigation (defluoridation / alternative source) at that stratum.
  • For a postmenopausal woman, the susceptibility appears to begin earlier (~0.5-1.0 mg/L) — a stratum-specific reason to weigh water-fluoride source, though on low-certainty single-lab data.
  • Do not read a BMD rise as bone benefit — the density marker and the fracture outcome dissociate here; the decision follows the fracture endpoint.
  • At US community-fluoridation exposure (0.7 mg/L), the overall curve shows no fracture signal (RR ~1.00, CI crossing 1.0) — insufficient-evidence / no-signal, not a demonstrated harm. The exposure number, not the word «fluoride», carries the decision.

References

Mazzoli, R., Filippini, T., Iamandii, I., De Pasquale, L., Veneri, F., Birnbaum, L. S., Rothman, K. J., & Vinceti, M. (2025). The association of fluoride exposure with bone density and fracture risk: a dose-response meta-analysis. Environmental Health, 24(1). https://doi.org/10.1186/s12940-025-01226-y

Graph View

Backlinks