The association of fluoride exposure with bone density and fracture risk: a dose-response meta-analysis.

Bone health is important to both quality of life and independence. Risk of fracture increases with age, along with low bone mineral density, osteopenia and osteoporosis [1]. One of the many environmental factors influencing bone health is fluoride (F), a widespread halogen commonly found in drinking water, but also in diet and dietary supplements, toothpaste, mouthwash, and ambient air [2]. Soil and groundwater fluoride content may be considerably affected by anthropogenic activities [3].

Drinking water is the most abundant source of fluoride for many individuals and populations [2, 4]. Fluoride is added to municipal water supplies in many areas, at concentrations ranging from 0.7 to 1.2 mg/L. The guideline in the USA was lowered to 0.7 mg/L in 2015 [5]. Fluoride from most dietary sources and drinking water is rapidly absorbed into the bloodstream, and deposited in calcified tissues, predominantly in bone and dental enamel [6]. Topical application of fluoride, through fluoride-containing dental products such as toothpaste and mouthrinse, and by direct application of a concentrated solution, is an alternative method of delivering fluoride to tooth enamel, and can be effective for dental caries prevention [2, 7].

Although water fluoridation has long been advocated for oral health promotion, evidence regarding caries prevention is still debated [2, 7, 8], with a recent review finding little benefit of community water fluoridation in caries prevention after the introduction of fluoridated toothpaste [9]. A number of adverse health issues due to excess fluoride intake, particularly in areas with high levels of fluoride in drinking water, have been documented or suspected [2, 4, 10,11,12,13,14]. They include altered cognitive neurodevelopment [11, 14, 15], dental and skeletal fluorosis [13], impaired thyroid function [10], altered bone mineral density and increased risk of fractures [16]. However, the occurrence of such effects and the levels of exposure involved have been debated [17], particularly for the neurodevelopmental effects [2, 12, 18,19,20,21,22,23]. Given the absence of any dose-response analysis on fluoride and bone health, we conducted a meta-analysis to assess the overall relation between fluoride exposure and bone health, focusing on potential critical thresholds.

Study protocol and registration

We registered the protocol of this systematic review in the PROSPERO database (registration no. CRD42022321899) and carried it out according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) 2020 guidelines [24].

Search strategy and study selection

We conducted two independent comprehensive literature searches according to the PECOS criteria (population, exposure, comparison, outcome, study type) [25], driven by the questions: “What is the effect of fluoride exposure on bone strength and density? and “What is the effect of fluoride exposure on risk of fractures?“. We included all studies with: (P) substantially healthy population of any age; (E) assessment of long-term fluoride exposure by determination of drinking water fluoride content or dietary intake and/or using biomarkers of exposure; (C) reporting of health indicators within two or more categories of fluoride exposure; (O) assessment indicators of bone density or strength (e.g., bone mineral density-BMD, bone mineral content-BMC), and/or diagnosis of bone fractures; (S) any study type, including trial, cohort or case-control study. When multiple studies examined the same population, we included only the most complete one (e.g. with the longest follow-up).

To find the eligible studies, we searched the PubMed/MEDLINE, Embase, and Web of Science databases from inception up to August 26, 2025, using MeSH terms and keywords linked to “bone”, “density”, “fracture” and “fluoride” (Supplementary Table S1), deleting duplicate studies. Three authors (RM, II, LDP) independently screened titles, abstracts, and full texts of the retrieved studies to exclude those that did not meet the eligibility criteria. Any disagreements were resolved by the intervention of two other authors (TF and MV), or until a consensus was reached. Additionally, we checked the reference lists of included studies and previous reviews, and employed backward and forward citation retrieval techniques to identify potentially relevant additional articles [26].

We included studies reporting different types of effect and risk measures, such as relative risk (RR), hazard ratio (HR), odds ratio (OR), or mean difference (MD), along with the corresponding 95% confidence interval (95% CI), standard error (SE), standard deviation (SD), or interquartile range (IQR), or data to calculate them as per Cochrane Handbook for Systematic Reviews of Interventions [27]. This included calculating standard errors from reported confidence intervals, converting interquartile ranges to standard deviations using established conversions (e.g., SD?=?IQR/1.35), and estimating standard deviations from p-values using standard normal approximations when applicable. We did not apply statistical transformations to convert between effect estimates, which were pooled only when directly comparable.

Data extraction

Three authors (RM, II, LDP) independently conducted data extraction from the selected articles, with a fourth author (TF) verifying the process. We employed two distinct standardized spreadsheets for collecting data relating to the two outcomes. For each article, we compiled the following details: (1) the first author’s name; (2) study details (title, publication year, study design, country); (3) population characteristics (sex, age, sample size); (4) exposure characteristics (definition, assessment method, categories of exposure); (5) outcome (definition, assessment method, number of cases); (6) effect estimates (RRs, HRs, ORs, MDs) with their 95% CIs, SE, SD from the most adjusted model; (7) potential confounders accounted for in the multivariable analysis.

Risk of bias assessment

Three authors (RM, II, LDP) independently assessed the risk of bias (RoB) of the included studies based on the Risk of Bias for Non-randomized Studies of Exposures (ROBINS-E) criteria [28]. Any discrepancies were resolved with input from two other authors (MV and TF). We categorized the studies into low, moderate, or high risk of bias based on seven domains: (1) confounding bias; (2) participant selection bias; (3) exposure classification bias; (4) departures from intended exposures bias; (5) missing data bias; (6) outcome measurement bias; and (7) bias in the selection of reported results. In (Supplementary Table S2), we detailed the criteria for risk of bias evaluation. We deemed a study to have a moderate or high risk of bias if at least one domain was judged as moderate or high risk, respectively; else, it was classified as having a low risk of bias.

Data analysis

We quantitatively compared fracture risks and BMD values across varying levels of fluoride exposure using a restricted maximum likelihood random-effects model [29]. Additionally, we conducted a dose-response meta-analysis to assess bone fracture risk and BMD with increasing fluoride exposure, based on the one-stage approach [30], a methodology recently applied to pool epidemiologic data and characterize the dose-response relationship across the entire range of exposure [31,32,33,34]. Where mean or median values were unavailable for levels of exposure in open-ended categories, we estimated values 20% higher or lower than the nearest cut-point [35]. To allow for non-linear relationships, we used a restricted cubic spline model with three knots at fixed percentiles (10th, 50th, and 90th) of fluoride exposure. We assessed heterogeneity of included studies using the

statistics evaluating the between-study heterogeneity relative to the variance of the pooled random-effects estimate [36] and through variation of study-specific curves evaluating the variability across individual studies [29, 37]. We conducted funnel plot analyses and Egger’s regression tests wherever sufficient data were available.

In the aggregate BMD analyses, when multiple measurements from different subregions of the hip were reported within a single study (e.g., femoral neck, Ward’s triangle, trochanter), we selected only one to include in the meta-analysis to avoid multiple correlated estimates from the same anatomical region. To ensure consistency and comparability across studies, we applied a predefined hierarchy of hip subregions based on the relative frequency of reporting across studies, prioritizing them in the following order: hip or femur, femoral neck, Ward’s triangle, trochanter, inter-trochanter. When more than one site was reported within a study, we included only the highest-priority site according to this hierarchy.

Whenever possible, we carried out stratified analyses separately for different levels of variables that might modify the effect measure, such as sex and age. We also carried out sensitivity analyses by excluding studies at high risk of bias. Additionally, we examined fragility fractures in females over 50, the average age of menopause onset [38], to assess how menopause and fluoride exposure jointly affect bone health. We used Stata software (v18.0, Stata Corp., College Station, TX, 2023) for all data analyses.

Study selection and characteristics of the included studies

The detailed literature databases search and study selection process are in the PRISMA flowchart (Fig. 1). Regarding bone strength and density, we initially identified a total of 1257 articles. Following duplicate removal, 1049 articles remained, from which 1010 records were excluded after title and abstract assessment as they were considered not relevant or they were not written in English. After these exclusions, there were 39 articles remaining for full-text evaluation. We made additional exclusions after full-text review, excluding an additional total of 22 studies. Details on reasons for exclusion are provided in (Supplementary Table S3).

PRISMA flowchart for study identification and inclusion. Studies included in the qualitative synthesis [16, 39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74]/studies included in the quantitative synthesis [39,40,41, 43, 45, 48,49,50,51,52, 54, 55, 57,58,59, 61, 63,64,65, 68, 69, 72, 73]

Full size image

We applied the same selection process to assess fracture risk, initially identifying a total of 1097 articles before removing duplicates, and 982 articles afterward. Among these, 943 were excluded after screening titles and abstracts, as they were deemed unsuitable publication types, irrelevant to the outcomes, or non-English studies. Out of the 39 articles we assessed in full text, 13 were subsequently excluded. The articles excluded after full-text review, along with the reasons for their exclusion, are listed in (Supplementary Table S3). Furthermore, we retrieved one article regarding bone fractures and one article on whole-body BMD through citation chasing [39].

Ultimately, 37 studies were included in our meta-analysis and deemed suitable to assess the association between fluoride exposure and bone health: 10 pertaining to bone density only [40,41,42,43,44,45,46,47,48]; 19 on bone fractures only [39, 49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66]; and eight on both outcomes [16, 67,68,69,70,71,72,73].

Characteristics of the included studies

The details of the 37 studies included are summarized in Table 1. Regarding the year of publication: three studies were conducted before 1980 [39, 54, 58], three were carried out in the 1980?s [40, 49, 64], the majority, 18, were published between 1990 and 1999 [41, 43, 45,46,47,48, 50,51,52,53, 56, 57, 59, 65, 67,68,69, 72], eight were published between 2000 and 2019 [42, 55, 61, 63, 66, 70, 71, 73], and five were published after 2020 [16, 44, 60, 62, 74].

The majority of studies focused on North American populations, with 16 being from the USA [39, 47, 48, 51, 53, 54, 56,57,58, 62, 67, 70,71,72,73,74] and two from Canada [41, 65]. The second most represented population was Northern Europe, with five studies from Finland [40, 49, 59, 64, 68] and three from Sweden [16, 46, 63]. Additionally, there were three studies from the UK [50, 55, 66], one from Poland [43], one from Italy [52], and one from Germany [75]. From Asia, two studies were included from China [42, 61], one from Taiwan [45], and one from South Korea [60]. Only one study originated from Africa, specifically from Ethiopia [44].

Sixteen studies had an ecological design [39, 43, 46, 49,50,51,52, 54, 56,57,58, 60, 61, 63,64,65], 12 a cohort design [16, 40, 41, 47, 48, 53, 59, 67, 70,71,72, 74], nine cross-sectional [42, 44, 45, 62, 66, 68, 69, 73, 74], with one being interventional [42], one a population-based case-control design [55]. Exposure was predominantly assessed by measuring fluoride drinking water content [39,40,41,42,43,44,45,46,47,48,49,50, 52, 54, 55, 57,58,59,60,61,62,63,64,65,66,67,68,69, 71,72,73,74] reported either in mg/L or ppm, which are numerically equivalent and directly comparable. Urinary fluoride content was expressed either as mg/L [42, 43, 74] or mg/g creatinine [16]. Dietary or daily fluoride intake was expressed as mg/day by two studies [16, 70]. One study assessed fluoride content in serum [73], one in plasma [74], one in hair [43]. and one in toenails [53]. These alternative biomarkers were not integrated into pooled analyses, as exposure-response synthesis was only performed for studies using directly compatible exposure metrics.

Concerning outcomes, bone strength studies assessed BMD [16, 40,41,42,43, 45,46,47,48, 67,68,69,70,71,72,73,74], BMC [47, 70], bone compressive strength [40], bone width [47], and speed of sound [44]. Regarding bone density or strength, the investigated locations were lumbar spine or spine [16, 41, 43, 45, 48, 67,68,69,70,71, 73], hip and femur [16, 40, 43, 67,68,69,70,71,72,73], radius, tibia and forearm [44, 47, 48, 67, 71,72,73, 76], calcaneus [67, 71], tibia and phalanx [44], and whole-body [41, 46, 70, 74].

Risk of fractures was assessed with reference to arm, humerus, elbow and wrist [39, 52, 57, 58, 62, 67, 68, 71, 72], leg, knee and ankle [39, 57, 68, 72], hip and femur [16, 49,50,51,52,53,54,55,56,57,58,59, 61,62,63,64,65,66,67, 69, 71] vertebral [52, 67, 71], and nose [52]. Each study was categorized according to the most specific anatomical location it reported (i.e., if a study included both “arm fractures” and “humerus fractures”, we only considered the latter). Some studies used broader endpoints such as “all fractures” [39, 60, 61, 68, 70, 72], “all non-vertebral fractures” [71], and “osteoporotic” and “non-osteoporotic fractures” [73].

Risk of bias assessment

We provide details about study quality (Supplementary Table S4). Regarding bone density and strength, we deemed nine of the included studies to have overall “high” risk of bias [40, 43, 46, 47, 69,70,71, 76], five at “moderate” risk [41, 48, 67, 68, 72], and four at “low” risk [16, 44, 45, 73].

Considering bone fractures, we judged 18 of the included studies at “high” risk of bias for this outcome [39, 49, 50, 52, 54, 56,57,58, 60, 62, 64,65,66, 68,69,70,71,72], seven at “moderate” risk of bias [51, 53, 55, 59, 61, 63, 67], and two at “low” risk of bias [16, 73].

Regardless of the outcome, the main factors contributing to the high-risk rating included unaccounted sources of confounding (n?=?11) and bias in exposure measurement (n?=?8).

Dose-response meta-analysis on risk of fractures

Four studies [39, 61, 68, 72] assessed the overall risk ratio of fractures, regardless of the site. Figure 2 illustrates the dose-response curve and the threshold effect for all fractures risk in relation to fluoride exposure, in the overall study populations since no sex-specific analysis was feasible. 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, as fluoride concentrations increased, the risk of fracture gradually increased, with a RR of 1.06 (95% CI 0.80–1.41) at 2.0 mg/L, increasing to 1.19 (95% CI 0.87–1.64) at 3.0 mg/L, and to 1.35 (95% CI 0.94–1.95) at 4.0 mg/L. Study specific curves strongly support a consistent positive association between fluoride exposure and fracture risk across studies, with the only exception of one investigation of teen-age males aged 15–17 [39] (Supplementary Figure S1).

Dose-response meta-analysis of fracture risk (all fractures) and exposure to fluoride (F) from water [39, 61, 68, 72] (

Sensitivity analyses on risk of fractures and fragility fractures

Regarding fracture risk in females over 50 years of age (Supplementary Figure S6), the curve exhibited a steep upward trend, with the lowest risk around 0.25 mg/L and a sharp increase in fracture risk starting at around 0.5 mg/L. Specifically, 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. For this analysis, (Supplementary Figure S7) provides the funnel plot and Egger’s test (slope: 1.72, 95% CI 0.19–3.62) indicating a largely symmetrical distribution and Little evidence of publication bias. Concerning BMD, a dose-response analysis in females over 50 was only possible for spine BMD, with an indication of an inverted U-shaped curve with a downward trend starting at a lower F concentrations than in the overall analysis (Supplementary Figure S8). There was some asymmetry in the published data, but not strong evidence of publication bias (Supplementary Figure S9).

Several studies [16, 48, 53, 55, 67, 68, 71, 73] adjusted the results depending on menopause onset and/or hormone replacement therapy use, another important factor in bone health status in females [77]. We had insufficient data, however, to report on this group of studies by themselves.

Concerning the effect of study quality on the pattern of associations, a subgroup analysis excluding the studies at higher risk of bias could be conducted only for a few endpoints, and not for the main analysis concerning risk of all fractures. Regarding overall RR of fragility fractures (Supplementary Figure S10), the RR decreased to 0.93 (95% CI 0.72–1.21) at fluoride exposure in the 0.8–1.5 mg/L range; it was 0.95 (95% CI 0.70–1.29) at 1.8 mg/L, with inconsistencies across the few studies available. In the analysis of hip fracture risk across all sexes after exclusion of studies at high risk of bias (Supplementary Figure S11), there was little change in the RR across the entire range of exposure, again with notable variation between studies: some studies clearly indicated a decrease in risk, while others showed a consistent increase as fluoride levels rise.

Sensitivity analyses on BMD

BMD outcomes for females, specifically for the hip and spine, restricted to studies with low-to-moderate risk of bias, are shown in (Supplementary Figure S12). Regarding hip BMD, the MD curve remained stable up to around 0.9 mg/L, after which a monotonic and almost Linear upward trend emerged. Specifically, MD increased to 0.04 g/cm² (95% CI ?0.03?0.12) up to around 2.3 mg/L, reaching 0.09 g/cm² (95% CI ?0.09; 0.27) at 4.0 mg/L. Study-specific curves consistently mirrored this pattern. For spine BMD, an initial rise leveled off at around 0.5 mg/L, with MD values of 0.02 g/cm² (95% CI ?0.02; 0.07) at 0.20 mg/L and 0.06 g/cm² (95% CI ?0.01; 0.12) from 0.7 to 2.35 mg/L, peaking at 0.07 g/cm² (95% CI 0.11; 0.24) above. Study-specific curves displayed variable shape, in one study peaking between 0.4 and 0.8 mg/L before declining at higher exposure, while others follow the general pattern of the pooled analysis.

(Supplementary Figure S13) illustrates BMD for both sexes, after excluding studies at high risk of bias. For hip BMD no relation with fluoride exposure emerged, with curves of the single studies generally aligning with this pattern. For spine BMD, there was little evidence of a relation with water fluoride concentrations: the curve initially rose to a MD of 0.06 g/cm² (95% CI ?0.02; 0.13) around fluoride levels of 0.60 mg/L. However, as fluoride levels increased beyond 1.27 mg/L, the Line flattened at a 0.05 g/cm² (95% CI 0.01; 0.10) until 2.27 mg/L, followed by a slight decrease to 0.04 g/cm² (95% CI ?0.21; 0.30) after this point. Results of individual studies were very inconsistent, showing discordant trends at increasing fluoride exposure.

By pooling results from previously published studies across all the exposure levels investigated, we could model and characterize the dose-response pattern of fluoride exposure with bone thickness and fracture risk. We found evidence for a positive dose-response relation with fracture risk for doses exceeding a threshold that is close to the current WHO guideline value of 1.5 mg/L of drinking water fluoride. Previous reviews and meta-analyses did not consider the dose-response curve across the full range of fluoride exposure [78,79,80,81,82], concluding that while low and moderate fluoride exposures seem to have limited effects on fracture risk and can increase bone mass in some cases. The larger number of studies available for this review and the pooling of exposure-specific effect estimates allowed us to assess the relation between fluoride and bone quality quantitatively across the range of exposure characterizing the general population.

Our data indicated little relation between fluoride exposure and overall fracture risk below 1.5 mg/L of drinking water fluoride, with fracture risk starting to increase gradually above that threshold. The threshold level corresponds to around 3-3.5 mg/day of exposure from this source (assuming a water consumption of 2 L/day and little contribution of fluoride from other sources) [83, 84]. In several of our pooled analyses, some types of fractures showed a U-shaped pattern, with risk of fracture decreasing with increasing fluoride exposure from zero to moderate levels of exposure, but increasing as exposure increased from moderate to high levels of fluoride exposure. Such a pattern might be an artifact due to the more limited number of studies on which such analyses were based, and a related higher uncertainty and statistical instability, especially in view of the small changes in RR involved. Nonetheless, the hypothesis that exposure to drinking water fluoride around 0.5–0.6 mg/L might have beneficial effects on fracture risk (compared with no exposure), at least on some specific fracture subsets, is consistent with our findings. We also note the inconsistencies between the findings for site-specific fractures (hip, wrist and femur), with different patterns of association, that mirror the inconsistencies found in the BMD analyses for hip and spine. Such inconsistencies may be due to the limited number of studies for these endpoints, or to real differences of the fluoride effects on specific bone districts. Bone composition might also play a role, as wrist is primarily composed of cortical bone while spine is predominantly trabecular bone, which is more metabolically active and responsive to agents that stimulate bone formation [85]. Interestingly, similar dose-response patterns and approximate thresholds have been observed in pooled analyses of fluoride exposure and two key health endpoints, IQ and thyroid function in children [10, 14].

Among females over 50 years, who are at notably higher fracture risk than younger subjects [86, 87], there was a stronger correlation between fluoride exposure and fragility fracture risk, starting at fluoride concentrations around 0.5 mg/L. Furthermore, the decline in spine BMD was slightly stronger in females aged 50 years or over, and the decline began at lower fluoride drinking water levels, compared with all females. Although these findings are based on a limited number of studies, the protective role of estrogen in bone health is well established [88, 89], and it is also widely recognized that postmenopausal females are particularly vulnerable to osteoporosis and fragility fractures [90]. Menopause accelerates bone loss by increasing bone resorption at a rate that surpasses bone formation, both directly [91]. and indirectly [92]. The steep upward trend in fragility fracture risk in females?>?50 years of age may reflect long-term continuous exposure to fluoride in drinking water or increased release of fluoride from bone stores. This possibility underscores the importance of cumulative exposure in a “potentially exposed or susceptible subpopulations,” according to EPA Toxic Substances Control Act [93]. Although our analysis stratified by age (i.e. females over 50) rather than confirmed menopause status, age likely serves as a reasonable proxy for menopausal status in this analysis. The possibility of a negative effect of fluoride exposure, assessed through water and plasma fluoride concentrations, on sex steroid hormones including estradiol is supported by the limited epidemiologic and toxicological evidence available, and post-menopausal fluoride release from bone post menopause could contribute to such adverse effects [10, 94,95,96].

Helte et al. reported an adverse effect of fluoride exposure on all fractures and hip fractures in a cohort study of postmenopausal Swedish females, which we could not include in our analysis because it was based on a urinary biomarker and on overall dietary exposure [16]. In that study, an excess risk for all fractures was apparent among those with urinary fluoride levels in the highest tertile even in an area with relatively low water fluoride concentration (??1 mg/L) [16].

We could conduct this quantitative dose-response pooled analysis only for fluoride exposure in drinking water, and not from other sources, due to the limited number of studies investigating exposure biomarkers or directly assessing dietary intake. Unfortunately, most studies investigating risks associated with water fluoride content did not consider total fluoride exposure and particularly that from two major sources of dietary intake, tea and fluoridated salt, increasing the uncertainty inherent in the overall exposure assessment.Nonetheless, drinking water is generally the major source of fluoride intake even when its fluoride content is low (i.e. ? 1 mg/L) [2, 16, 97,98,99], therefore justifying an exposure assessment based on this indicator [10, 12,13,14]. Furthermore, there is evidence that exposure assessment based on biomarkers, although capturing additional sources of fluoride beyond drinking water, is limited by the short half-life of fluoride (around 6-hours), thus detracting from the utility of biomarkers compared with a long-term stable indicator of exposure such as drinking water fluoride content [100, 101]. Finally, the observational design of the studies relied upon here might be conducive to residual confounding, though the studies that were included generally attempted to control most relevant confounders.

The additional bone health indicator we assessed, BMD, refers to bone quality and is generally considered a key indicator of bone strength and a standard for clinical assessment, being used as the measure to predict fracture risk and assess responses to interventions [102]. The results for the association between fluoride and BMD were not consistent, and they also did not correspond to the pattern observed for fractures, raising the question about the usefulness of this endpoint. BMD only measures bone density, without providing information on bone morphology, architecture, tissue properties, or susceptibility to fracture [103,104,105,106]. Therefore, BMD does not provide a full picture of bone health and does not necessarily predict fracture risk, which is a direct outcome reflecting overall bone strength and health [107]. The hypothesis that a fluoride-induced increase in BMD could coexist with an increased fracture risk due to an inherently lower bone ‘quality’ should not be dismissed [108]. Heterogeneity across imaging methods and populations could also contribute to this discrepancy, but again 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. The association between fluoride exposure and hip BMD were similar in our dose-response meta-analysis and the above mentioned Swedish study [16], whereas spine BMD was not investigated in the latter.

Overall, our findings about the effects of fluoride exposure on fracture risk, assessed through drinking water concentrations, indicate a non-linear, positive dose-dependent relation, with an indication of a U-shaped association with fragility fractures. The association with BMD was less consistent. We also found evidence for an enhanced susceptibility of postmenopausal females to the adverse effects of fluoride on fracture risk. The observed effects that begin near regulatory thresholds and extend into high levels of fluoride exposure may be important to consider in shaping any public health policy that attempts to balance risks against benefits of fluoride exposure.

Data in this systematic review and meta-analysis are extracted from published studies available elsewhere. All processed data are presented in this article and the supplementary material.

BMC:

Bone mineral content

BMD:

Bone mineral density

CI:

Confidence interval

HR:

Hazard ratio

IQR:

Interquartile range

MD:

Mean difference

OR:

Odds ratio

PRISMA:

Preferred reporting items for systematic reviews and meta-analysis

ROBINS-E:

Risk of bias for non-randomized studies of exposures

RR:

Relative risk

SD:

Standard deviation

SE:

Standard error

Not applicable.

None.

Authors and Affiliations

1. Environmental, Genetic and Nutritional Epidemiology Research Center (CREAGEN), Section of Public Health, Department of Biomedical, Metabolic and Neural Sciences, University of Modena and Reggio Emilia, Modena, Italy

   Riccardo Mazzoli, Tommaso Filippini, Inga Iamandii & Marco Vinceti

2. Department of Biomedical, Metabolic and Neural Sciences, University of Modena and Reggio Emilia, Modena, Italy

   Riccardo Mazzoli, Tommaso Filippini, Inga Iamandii, Lisa De Pasquale & Marco Vinceti

3. School of Public Health, University of California Berkeley, Berkeley, CA, USA

   Tommaso Filippini

4. PhD Program in Clinical and Experimental Medicine, Department of Biomedical, Metabolic and Neural Sciences, University of Modena and Reggio Emilia, Modena, Italy

   Federica Veneri

5. Department of Surgery, Medicine, Dentistry and Morphological Sciences with Transplant Surgery, Oncology and Regenerative Medicine Relevance (CHIMOMO), Unit of Dentistry & Oral-Maxillo-Facial Surgery, University of Modena and Reggio Emilia, Modena, Italy

   Federica Veneri

6. Nicholas School of the Environment, Duke University, Durham, NC, USA

   Linda S. Birnbaum

7. Department of Epidemiology, Boston University School of Public Health, Boston, MA, USA

   Kenneth J. Rothman & Marco Vinceti

Contributions

RM, TF, and MV drafted the manuscript. TF, FV, LSB, KJR, and MV contributed to the study design and methodology. RM, II, and LDP conducted the primary literature search, data extraction, and synthesis of results. TF, LSB, and MV assisted in the literature search and data extraction. TF and MV resolved conflicts regarding study inclusion and data extraction. RM and TF conducted the statistical analysis. All authors contributed to critical revision of the manuscript for important intellectual content and gave final approval for the decision to submit for publication.

Corresponding author

Correspondence to Marco Vinceti.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.