Protective versus risky: the complex relationship between prenatal fluoride exposure and birth outcomes in a Chinese large-scale population study

Abstract

Background

Previous studies on fluoride in drinking water and birth outcomes have yielded inconsistent findings. Furthermore, the association between fluoride exposure and gestational age (the duration of pregnancy), along with offspring characteristics at birth, has not been thoroughly explored. Given the widespread occurrence of drinking water fluorosis in China, this large-scale study aims to explore the associations between fluoride and birth outcomes.

Methods

We conducted a database study involving 352,973 participants in China from 2016 to 2022. Generalized linear models and restricted cubic spline curves were employed for the analysis, and the mediating effects on gestational age were systematically evaluated.

Results

The average fluoride concentration was 0.41 ± 0.16 mg/L. Each unit increase in fluoride was associated with a 0.36-week (95% confidence intervals [CI]: 0.34, 0.39) increase in gestational age and a 43.56 g (95% CI: 33.56, 53.57) increase in birth weight. Fluoride exposure exhibited dual effects: it was protective against preterm birth (odds ratio [OR] = 0.74; 95% CI: 0.67, 0.82) and low birth weight (OR = 0.69; 95% CI: 0.61, 0.78), while simultaneously elevating the risks of macrosomia (OR = 2.56; 95% CI: 2.40, 2.74), small-for-gestational age (OR = 1.72; 95% CI: 1.58, 1.88), and large-for-gestational age (OR = 1.27; 95% CI: 1.20, 1.34). Nonlinear exposure-response relationships were observed for all outcomes except low birth weight. Mediation analysis indicated that gestational age served as a significant mediator in the associations between fluoride and birth weight, as well as low birth weight and macrosomia—suggesting that fluoride may increase birth weight and reduce the risk of low birth weight through prolonged pregnancy. Notably, gestational age was not analyzed as a mediator for small-for-gestational age and large-for-gestational age due to its distinct growth deviation nature. However, its paradoxical associations with both fetal growth restriction and overgrowth (macrosomia/large-for-gestational age) warrant clinical attention, although these observational links do not establish definitive causality.

Conclusion

Prenatal fluoride exposure exhibits dual effects in China: prolonging gestational age and reducing preterm birth and low birth weight risks, while increasing risks of abnormal fetal growth. These findings underscore the necessity for region-specific fluoride monitoring and prenatal counseling.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12940-026-01267-x.

Introduction

Elemental fluorine is abundant in the earth’s crust and forms naturally occurring fluorine (F⁻) minerals in soils and aquifer sediments, including fluorite (CaF₂), cryolite (Na₃(AlF₆)), and fluorapatite (Ca₅(PO₄)₃F). These minerals contribute to the accumulation of fluorine in freshwater resources, particularly groundwater [1]. The natural occurrence of high concentrations of fluoride in groundwater presents a global health concern, potentially affecting 200 million individuals across 25 countries [2]. Excessive fluoride intake can damage various bodily systems, including the skeletal [3], endocrine [4], cardiovascular [5], nervous [6], reproductive [7], digestive [8], and immune systems [9]. Evidence suggests that fluoride consumption during early life may have adverse effects on neurodevelopment in children [10–13]. Furthermore, since birth outcomes can also influence neurodevelopment [14], adverse birth outcomes may serve as a pathway linking intrauterine fluoride exposure to potential neurodevelopmental effects [15].

Animal and epidemiological studies conducted in regions with naturally high levels of fluoride in groundwater indicate underlying mechanisms and suggest potential adverse effects of fluoride in water on reproductive health. Evidence suggests that fluoride exposure may disrupt the endocrine system by altering thyroid hormone levels [16, 17]. Numerous other endocrine-disrupting chemicals have been linked to adverse birth outcomes, as hormones play a critical role in regulating normal growth processes during pregnancy and delivery [18]. Furthermore, fluoride exposure can lead to oxidative stress and inflammation [9, 19, 20], which have also been implicated in the pathogenesis of adverse birth outcomes [21].

Emerging evidence regarding fluoride’s impact on birth outcomes reveals significant inconsistencies across global studies. For instance, while Arun et al. demonstrated an inverse association between elevated fluoride levels and reduced birth weight in the United States [22], contrasting findings from Sweden [23], Mexico [24], and Iran [25] reported positive correlations with increased birth weight. Similarly, divergent results exist concerning preterm birth risk, with some studies indicating an elevated incidence and others showing protective effects [26]. Notably, a Canadian cohort study found no significant links between fluoride exposure and either gestational duration or fetal growth metrics [27], further complicating the evidence base. These disparities may arise from methodological variations in exposure assessment, including timing and dosage, population-specific genetic susceptibilities, or environmental confounders. This is highlighted by Goin et al.’s simulation study in California, which projected reduced birth weight and standardized Z-scores under hypothetical fluoride-reduction scenarios after adjusting for critical confounders, including race/ethnicity, health insurance type, fetal sex, and arsenic levels. Such factors may introduce confounding bias by simultaneously influencing fluoride exposure levels—through infrastructure disparities or variability in water sources—and birth outcomes, such as healthcare access and nutritional status [15]. Collectively, the current literature underscores the need for standardized exposure classification and rigorous adjustment for contextual modifiers to reconcile these contradictory observations.

Given the critical role of gestational age in fetal development, we hypothesize that it may serve as a key mediator in the relationship between prenatal fluoride exposure and birth outcomes. Gestational age is a fundamental temporal determinant of fetal development, directly governing the duration of in utero exposure and organ maturation. Previous research indicates that environmental factors, such as chemical pollutants, may indirectly modulate birth weight and fetal growth indices by influencing pregnancy duration [15, 21]. Fluoride exposure could potentially disrupt the timing of parturition by interfering with endocrine homeostasis (e.g., thyroid function) or inducing oxidative stress [9, 16, 19], thereby altering gestational age. While prolonged gestation may increase the window for nutrient transfer, it could also exacerbate placental functional demand, potentially leading to growth imbalances. Consequently, we positioned gestational age as a mediator in this study to elucidate the potential pathway through which fluoride might influence fetal growth by modifying pregnancy length. This approach helps disentangle the direct effects of fluoride on fetal growth from its indirect effects mediated via changes in pregnancy duration, offering a mechanistic explanation for its dual role in birth outcomes.

China faces a significant epidemic of drinking water fluorosis, which impacts 28 provincial-level administrative divisions and affects over 70 million individuals [2]. However, there is a notable scarcity of large-scale studies investigating the effects of fluoride on birth outcomes within the Chinese population. Given the declining birth rate in China over recent decades [28], it is imperative to explore the factors influencing birth outcomes and implement corresponding intervention measures to enhance the quality of the birth population. To address this pressing need, we designed a population-based study that utilizes birth registry data and georeferenced water fluoride measurements in Jinan, a recognized fluorosis-endemic region. By systematically analyzing the association between maternal drinking water fluoride exposure and key birth outcomes, including gestational age, birth weight, and fetal growth indices, this research aims to establish an evidence-based foundation for developing pregnancy-specific fluoride exposure guidelines and preventive health strategies.

Materials and methods

Study participants

Jinan, the capital of Shandong Province (36°02′N − 37°54′N, 116°21′E − 117°93′E), covers an area of 10,244.45 km² and is situated in the central region of Shandong Province in eastern China. There has been no artificial fluoridation of the drinking water supplies in the study area, and fluoride present in both rural and urban water sources is of natural origin. We conducted a database study involving permanent residents of Jinan. All medical institutions in Jinan are required to register prenatal examination and newborn information in the Jinan Maternal and Child Health Information Management System, with all registrations performed by obstetricians. We extracted registration data from this system, covering the period from 2016 to 2022. After excluding participants who were not permanent residents of Jinan, those with a delivery age under 18 years, multiple births, stillbirths, induced labor due to fetal malformation, and cases lacking birth information, we included a total of 352,973 mother-infant pairs in the study. The study received approval from the Ethical Review Committee of Center for Endemic Disease Control, Chinese Center for Disease Control and Prevention, Harbin Medical University (No. hrbmuecdc20250326). The clinical trial number is not applicable.

Fluoride exposure

From 2016 to 2022, annual surveillance of drinking water quality and sanitation conditions was conducted during both the dry and wet seasons. This surveillance encompassed all municipal water supplies in urban areas and 100% of township water supplies in rural areas. A total of 4,326 water samples were collected during this period, with one sample obtained from each drinking water source (e.g., municipal supply station, township treatment facility) per season (dry/wet). All samples were sourced from drinking water outlets (e.g., taps at supply facilities) to reflect end-user exposure. The collection, preservation, examination, and quality control of these samples were performed in accordance with the Standard Test Methods for Drinking Water (GB/T 5750) [29–31], established by the Chinese national administrations. Drinking water samples were collected in 5.0 L polyethylene bottles and stored at 4℃ prior to analysis. The fluoride content in the water was determined using ion chromatography. The home addresses of pregnant women were matched with the fluoride concentrations in the drinking water of their respective supply areas. The mean fluoride concentration in drinking water for both the dry and wet seasons was used to represent the fluoride concentration in the region for that year. The mean fluoride exposure in drinking water during pregnancy was calculated based on the month of conception and the month of delivery. Notably, our assessment of prenatal fluoride exposure was based on area-level average concentrations of drinking water fluoride rather than direct measurements of individual daily tap-water consumption, such as the volume consumed for drinking or cooking. This approach assumes that tap-water use patterns are consistent among participants within the same water supply area, given that individual-level data on water consumption behavior were not available in the study databases.

Birth outcomes

This study examined various birth outcomes, including gestational age in weeks, birth weight in grams, birth weight-for-gestational age Z-score (birth weight Z-score), preterm birth, low birth weight (LBW), macrosomia, small-for-gestational age (SGA), and large-for-gestational age (LGA). The eight birth outcomes evaluated were specifically selected to comprehensively capture gestational duration and fetal growth homeostasis—two core dimensions of perinatal health potentially influenced by prenatal fluoride exposure. Gestational age was determined using both the maternal last menstrual period and ultrasound examinations conducted during the first trimester of pregnancy. When the gestational age calculated from both methods differed by less than one week, the last menstrual period was used as the criterion; conversely, if the difference exceeded one week, the ultrasound examination result was prioritized [32]. A population reference from 11 cities in China was employed to calculate the birth weight Z-score [33]. Preterm birth was defined as delivery occurring before 37 completed weeks of gestation. LBW and macrosomia were defined as birth weights of less than 2,500 g [34] and greater than 4,000 g [35], respectively. The growth standard for newborns by gestational age was utilized to evaluate SGA and LGA [32]. SGA was defined as being below the 10th percentile of sex-specific birth weight-for-gestational age, while LGA was defined as being above the 90th percentile [32].

Statistical analysis

Mean ± standard deviation (SD) was employed to summarize continuous variables, while frequency (percentage, %) was utilized to describe categorical variables. Fluoride concentrations, gestational age, birth weight, and birth weight Z-scores were presented as mean ± SD along with quartiles (P25, P50, and P75). Differential analyses were performed using the Student’s t-test and the Mann–Whitney U test, while categorical variables were compared using the Chi-square test. Generalized linear models (GLM) were utilized to assess the associations between fluoride exposure in drinking water and birth outcomes. The selection of covariates was guided by existing literature on confounders associated with birth outcomes and a directed acyclic graph (DAG) (Figure S1). Three models were analyzed: the crude model, which serves as a baseline to illustrate the unadjusted associations between fluoride exposure and birth outcomes, thus providing a reference for assessing the impact of subsequent covariate adjustments; Model 1, which adjusted for three core demographic covariates—age at delivery, educational level, and sex of the newborn—isolates the effects of these essential demographic confounders. By comparing the results between the crude model and Model 1, we can quantify whether the observed associations between fluoride and birth outcomes are influenced by these basic variables, thereby preventing the misattribution of effects to fluoride when they may instead reflect demographic differences. Model 2 further adjusted for all covariates in Model 1, in addition to three additional physiological and behavioral covariates: Body Mass Index (BMI) in early pregnancy, parity, and smoking status. Subsequently, restricted cubic spline (RCS) curves with four knots were utilized to establish the exposure-response (E-R) relationships between fluoride concentrations in drinking water and birth outcomes, while also examining potential nonlinearity. Mediation effect models were used to assess the mediated proportion (%) of gestational age in the relationships between fluoride and three outcomes: birth weight, LBW, and macrosomia. SGA and LGA were excluded from this analysis, as SGA and LGA are defined by birth weight percentiles relative to gestational age, making gestational age a fundamental reference for outcome classification rather than an independent mediator. Including SGA and LGA would introduce logical circularity and collinearity between the mediator (gestational age) and the outcomes, thereby distorting the estimates of mediation effects.

Stratified analyses were conducted to investigate whether the effects of fluoride exposure in drinking water on birth outcomes vary across different age groups and districts. Various sensitivity analyses were performed to confirm the robustness of the findings. Fluoride concentrations were categorized into quartiles (≤ 25th, Q1; 25th to 50th, Q2; 50th to 75th, Q3; >75th, Q4), and the associations between fluoride and birth outcomes were reexamined using GLM models. Additionally, the associations were reevaluated by excluding participants with high or low BMI during early pregnancy. The E-value was calculated to assess the extent of an unadjusted confounding variable required to diminish the relationship between fluoride and birth outcomes [36]. Statistical analyses were performed using R (version 4.4.1), with the mediation effect model implemented via the “mediation” package. A statistically significant difference (two-tailed) was defined as P < 0.05.

Results

Descriptive statistics

The characteristics of the participants are summarized in Table 1. A total of 352,973 participants were included in the study, with an average age at delivery of 30.71 ± 4.77 years; 20.7% of the mothers were aged 35 years and older. More than half of the mothers (58.3%) were college graduates or above, and 71.6% of the participants resided in urban areas. The early pregnancy BMI averaged 23.15 ± 3.53 kg/m², with 58.1% of participants falling within the normal range. The proportions of participants classified as overweight, obese, and underweight were 25.8%, 10.0% and 6.1%, respectively. Additionally, 0.1% of mothers self-reported as smokers. Among the newborns, 19,726 (5.6%) were diagnosed with preterm birth, 13,069 (3.7%) with LBW, 33,382 (9.5%) with macrosomia, 20,083 (5.7%) with SGA, and 72,388 (20.5%) with LGA.

Table 1.

The basic characteristics of study participants

Variables	Total	Preterm birth			LBW			Macrosomia			SGA			LGA
	(n = 352,973)	No (n = 333,247)	Yes (n = 19,726)	P	No (n = 339,904)	Yes (n = 13,069)	P	No (n = 319,591)	Yes (n = 33,382)	p	No (n = 332,890)	Yes (n=20,083)	P	No (n = 280,585)	Yes (n = 72,388)	P
Age at delivery (years), mean (SD)	30.71 (4.77)	30.64 (4.74)	31.80 (5.07)	< 0.001	30.69 (4.75)	31.29 (5.11)	< 0.001	30.65 (4.76)	31.33 (4.78)	< 0.001	30.78 (4.75)	29.59 (4.90)	< 0.001	30.46 (4.74)	31.68 (4.76)	< 0.001
< 35 years, n (%)	279,963 (79.3)	265,902 (79.8)	14,061 (71.3)	< 0.001	270,210 (79.5)	9,753 (74.6)	< 0.001	254,886 (79.8)	25,077 (75.1)	< 0.001	263,004 (79.0)	16,959 (84.4)	< 0.001	227,085 (80.9)	52,878 (73.0)	< 0.001
≥ 35 years, n (%)	73,010 (20.7)	67,345 (20.2)	5,665 (28.7)		69,694 (20.5)	3,316 (25.4)		64,705 (20.2)	8,305 (24.9)		69,886 (21.0)	3,124 (15.6)		53,500 (19.1)	19,510 (27.0)
Educational level, n (%)				0.077			0.219			0.375			0.120			< 0.001
Junior high school and below	66,378 (18.8)	62,556 (18.8)	3,822 (19.4)		63,862 (18.8)	2,516 (19.3)		60,087 (18.8)	6,291 (18.8)		62,561 (18.8)	3,817 (19.0)		52,579 (18.7)	13,799 (19.1)
High school graduate	80,679 (22.9)	76,156 (22.9)	4,523 (22.9)		77,657 (22.8)	3,022 (23.1)		72,955 (22.8)	7,724 (23.1)		76,207 (22.9)	4,472 (22.3)		63,863 (22.8)	16,816 (23.2)
College graduate or above	205,916 (58.3)	194,535 (58.4)	11,381 (57.7)		198,385 (58.4)	7,531 (57.6)		186,549 (58.4)	19,367 (58.0)		194,122 (58.3)	11,794 (58.7)		164,143 (58.5)	41,773 (57.7)
District, n (%)				0.017			0.025			0.924			< 0.001			< 0.001
Urban	252,781 (71.6)	238,507 (71.6)	14,274 (72.4)		243,535 (71.6)	9,246 (70.7)		228,882 (71.6)	23,899 (71.6)		239,366 (71.9)	13,415 (66.8)		200,079 (71.3)	52,702 (72.8)
Rural	100,192 (28.4)	94,740 (28.4)	5,452 (27.6)		96,369 (28.4)	3,823 (29.3)		90,709 (28.4)	9,483 (28.4)		93,524 (28.1)	6,668 (33.2)		80,506 (28.7)	19,686 (27.2)
BMI in early pregnancy (kg/m2), mean (SD)	23.15 (3.53)	23.10 (3.50)	23.93 (3.92)	< 0.001	23.14 (3.52)	23.39 (3.87)	< 0.001	23.03 (3.50)	24.27 (3.60)	< 0.001	23.20 (3.53)	22.20 (3.46)	< 0.001	22.87 (3.46)	24.23 (3.62)	< 0.001
< 18.5 kg/m2, n (%)	21,534 (6.1)	20,518 (6.2)	1,016 (5.2)	< 0.001	20,572 (6.1)	962 (7.4)	< 0.001	20,691 (6.5)	843 (2.5)	< 0.001	19,192 (5.8)	2,342 (11.7)	< 0.001	19,540 (7.0)	1,994 (2.8)	< 0.001
18.5–23.9 kg/m2, n (%)	204,953 (58.1)	195,040 (58.5)	9,913 (50.3)		197,951 (58.2)	7,002 (53.6)		188,510 (59.0)	16,443 (49.3)		192,534 (57.8)	12,419 (61.8)		169,179 (60.3)	35,774 (49.4)
24-27.9 kg/m2, n (%)	91,067 (25.8)	85,382 (25.6)	5,685 (28.8)		87,675 (25.8)	3,392 (26.0)		80,217 (25.1)	10,850 (32.5)		87,107 (26.2)	3,960 (19.7)		67,728 (24.1)	23,339 (32.2)
≥ 28 kg/m2, n (%)	35,419 (10.0)	32,307 (9.7)	3,112 (15.8)		33,706 (9.9)	1,713 (13.1)		30,173 (9.4)	5,246 (15.7)		34,057 (10.2)	1,362 (6.8)		24,138 (8.6)	11,281 (15.6)
Smoker, n (%)	291 (0.1)	268 (0.1)	23 (0.1)	0.085	276 (0.1)	15 (0.1)	0.189	269 (0.1)	22 (0.1)	0.269	265 (0.1)	26 (0.1)	0.017	230 (0.1)	61 (0.1)	0.848
Parity, n (%)				< 0.001			< 0.001			< 0.001			< 0.001			< 0.001
Primiparous	155,570 (44.1)	147,223 (44.2)	8,347 (42.3)		149,026 (43.8)	6,544 (50.1)		141,760 (44.4)	13,810 (41.4)		143,495 (43.1)	12,075 (60.1)		130,715 (46.6)	24,855 (34.3)
Multiparous	197,403 (55.9)	186,021 (55.8)	11,379 (57.7)		190,878 (56.2)	6,525 (49.9)		177,831 (55.6)	19,572 (58.6)		189,395 (56.9)	8,008 (39.9)		149,870 (53.4)	47,533 (65.7)
Delivery, n (%)				< 0.001			< 0.001			< 0.001			< 0.001			< 0.001
Vaginal	183,108 (51.9)	177,343 (53.2)	5,765 (29.2)		179,637 (52.8)	3,471 (26.6)		168,815 (52.8)	14,293 (42.8)		171,897 (51.6)	11,211 (55.8)		153,970 (54.9)	29,138 (40.3)
Cesarean	169,865 (48.1)	155,904 (46.8)	13,961 (70.8)		160,267 (47.2)	9,598 (73.4)		150,776 (47.2)	19,089 (57.2)		160,993 (48.4)	8,872 (44.2)		126,615 (45.1)	43,250 (59.7)
Sex of newborn, n (%)				< 0.001			< 0.001			< 0.001			< 0.001			0.972
Male	184,109 (52.2)	173,139 (52.0)	10,970 (55.6)		177,898 (52.3)	6,211 (47.5)		163,270 (51.1)	20,839 (62.4)		173,024 (52.0)	11,085 (55.2)		146,356 (52.2)	37,753 (52.2)
Female	168,864 (47.8)	160,108 (48.0)	8,756 (44.4)		162,006 (47.7)	6,858 (52.5)		156,321 (48.9)	12,543 (37.6)		159,866 (48.0)	8,998 (44.8)		134,229 (47.8)	34,635 (47.8)

Abbreviations: LBW low birth weight, SGA small-for-gestational age, LGA large-for-gestational age

Table 2 presents the distribution of fluoride concentration in drinking water, gestational age, birth weight, and birth weight Z-score. The average fluoride concentration was found to be 0.41 ± 0.16 mg/L. The mean values for gestational age, birth weight, and birth weight Z-score were 38.72 ± 1.49 weeks, 3,344.67 ± 482.94 g, and 0.24 ± 1.07, respectively. Additionally, the characteristics of fluoride exposure in drinking water across various birth outcomes are detailed in Table S1.

Table 2.

Descriptive statistics of fluoride, gestational age, birth weight, and birth weight Z-score

Variables	Mean ± SD	P25	P50	P75
Fluoride (mg/L)	0.41 ± 0.16	0.32	0.39	0.48
Urban	0.40 ± 0.12	0.33	0.39	0.46
Rural	0.46 ± 0.22	0.30	0.42	0.58
Gestational age (weeks)	38.72 ± 1.49	38.42	39.29	40.14
Birth weight (g)	3,344.67 ± 482.94	3,050	3,350	3,650
Birth weight Z-score	0.24 ± 1.07	-0.5	0.18	0.91

Associations of fluoride with birth outcomes

Figure 1A illustrates the relationships between fluoride exposure and gestational age, revealing positive associations across all models. In adjusted model 2, each unit increase in fluoride was associated with a 0.36-week (95% CI: 0.34, 0.39) increase in gestational age. Figure 1B indicates that each unit increase in fluoride corresponded to a 43.56 g (95% CI: 33.56, 53.57) increase in birth weight. Figure 1C depicts the relationship between fluoride and birth weight Z-scores. Both the crude model and adjusted model 1 indicated positive associations between fluoride and birth weight Z-scores; however, in adjusted model 2, no association was found (β: -0.01, 95% CI: -0.04, 0.01). Figure 1D and 1E illustrate the relationships between fluoride and the risks of preterm birth and LBW, respectively, showing a 26% (OR: 0.74, 95% CI: 0.67, 0.82) decrease in preterm birth risk and a 31% (OR: 0.69, 95% CI: 0.61, 0.78) decrease in LBW risk. Figure 1F and 1G, and 1H present the relationships between fluoride and the risks of macrosomia, SGA, and LGA, respectively, indicating a 156% (OR: 2.56, 95% CI: 2.40, 2.74) increase in macrosomia risk, a 72% (OR: 1.72, 95% CI: 1.58, 1.88) increase in SGA risk, and a 27% (OR: 1.27, 95% CI: 1.20, 1.34) increase in LGA risk.

Fig. 1.

Association of fluoride in drinking water with birth outcomes. (A) Gestational age (weeks); (B) Birth weight (g); (C) Birth weight Z-score; (D) Preterm birth; (E) LBW; (F) Macrosomia; (G) SGA; (H) LGA. Notes: Crude model, without adjustment. Adjusted model 1, adjusted for age at delivery, educational level, and sex of newborn; Adjusted model 2, adjusted for age at delivery, educational level, sex of newborn, BMI in early pregnancy, parity, and smoker. Abbreviations: Z-score, birth weigh Z-score. LBW, low birth weight. SGA, small-for-gestational age. LGA, large-for-gestational age

As demonstrated in Fig. 2, the E-R relationships between fluoride exposure and birth outcomes predominantly exhibit nonlinear patterns (all P for nonlinear < 0.001), with the exception of LBW. Specifically, gestational age (Fig. 2A), birth weight (Fig. 2B), and birth weight Z-scores (Fig. 2C) exhibited sigmoidal trajectories. For gestational age and birth weight, fluoride-associated increases were modest at concentrations below 0.3 mg/L, accelerated significantly between 0.3 and 0.6 mg/L, and plateaued when concentrations exceeded 0.6 mg/L. This pattern suggests a potential biological saturation of fluoride’s effects on prolonging gestation and promoting fetal growth. Birth weight Z-scores displayed a similar sigmoidal trend, with minimal changes at fluoride concentrations below 0.3 mg/L, a gradual elevation between 0.3 and 0.5 mg/L, and stabilization above 0.5 mg/L. This indicates that fluoride’s influence on relative birth weight, compared to gestational age-specific norms, also exhibits a dose-dependent saturation. For preterm birth (Fig. 2D), macrosomia (Fig. 2F), SGA (Fig. 2G), and LGA (Fig. 2H), E-R curves demonstrated threshold-like characteristics. Risk reductions for preterm birth and elevations for macrosomia, SGA, and LGA became statistically significant only when fluoride concentrations exceeded 0.3 mg/L, with steeper slopes observed between 0.3 and 0.5 mg/L before flattening at higher concentrations. LBW (Fig. 2E) remained the sole exception, exhibiting a linear E-R relationship (P for nonlinear = 0.243), consistent with a uniform risk reduction per unit of fluoride exposure. Given the identified nonlinearity, subsequent mediation and sensitivity analyses—utilizing GLMs that assume linear exposure effects—were conducted with explicit acknowledgment of this assumption. To partially capture nonlinearity, quartile-based sensitivity analyses (Tables S6–S13) were prioritized, as they categorize fluoride into tiered exposure ranges (Q1: ≤ 0.32 mg/L, Q2-Q3: 0.32–0.48 mg/L, Q4: > 0.48 mg/L) that correspond with the observed sigmoidal and threshold patterns.

Fig. 2.

Exposure-response (E-R) function analysis of fluoride and birth outcomes. (A) Gestational age (weeks); (B) Birth weight (g); (C) Birth weight Z-score; (D) Preterm birth; (E) LBW; (F) Macrosomia; (G) SGA; (H) LGA. Abbreviations: Z-score, birth weight Z-score. LBW, low birth weight. SGA, small-for-gestational age. LGA, large-for-gestational age

The mediation effects of gestational age in the relationships of fluoride with birth weight, LBW risk and macrosomia risk

Significant mediating effects of gestational age were identified in the associations between fluoride exposure and birth weight levels, as well as the risks of LBW and macrosomia. The proportion mediated by gestational age was most pronounced for birth weight (150%; 95% CI: 116%, 187%), indicating that fluoride’s growth-promoting effects predominantly operate through the prolongation of pregnancy. For clinical risk endpoints, gestational age mediated 80% (95% CI: 64%, 130%) of fluoride’s protective association against LBW, but only 19% (95% CI: 17%, 21%) of its association with macrosomia risk (Fig. 3). Notably, while an extended gestation period reduced the risk of LBW and increased the risk of macrosomia, fluoride exposure was also associated with an elevated risk of SGA (Fig. 1G). This finding suggests that prolonging gestation does not uniformly enhance fetal growth and that other mechanisms may contribute to variations in growth.

Fig. 3.

Mediation effects of gestational age in the associations of fluoride with birth weight, LBW risk, and macrosomia risk. (A) Birth weight; (B) LBW; (C) Macrosomia. Abbreviations: ACME, average causal mediation effects (indirect effect); ADE, average direct effects. LBW, low birth weight. * P < 0.05, ** P < 0.01, and *** P < 0.001

Stratified analysis

In addition to the association between fluoride exposure and birth weight Z-scores, significant relationships between fluoride levels and various birth outcomes were observed among participants aged under 35 years and those aged 35 years or older. However, no significant differences in effect sizes were detected between these two age groups concerning the impact of fluoride on birth outcomes (P-interaction > 0.05) (Table S2 and Table S3). Stratified analyses by district indicated that, among urban participants, significant associations were observed between fluoride exposure and gestational age, birth weight, birth weight Z-scores, LBW, macrosomia risk, and LGA risk. However, no statistically significant associations were found between fluoride exposure and the risks of preterm birth or SGA within this group. Conversely, for rural participants, significant associations between fluoride and birth outcomes were identified across all metrics, with the exception of birth weight Z-scores. Furthermore, significant differences in effect sizes were observed between urban and rural participants for gestational age, birth weight Z-scores, the risk of preterm birth, the risk of macrosomia, the risk of SGA, and the risk of LGA (P-interaction < 0.001), suggesting potential environmental or socioeconomic confounding (Table S4 and Table S5).

Sensitivity analysis

Sensitivity analyses robustly confirmed these findings. Fluoride concentrations were categorized into four quartiles based on the distribution of drinking water fluoride levels within the study population (Table 2), with Q1 serving as the reference group. Notably, the quartile ranges correspond to the previously nonlinear E-R patterns: Q1 (≤ 0.32 mg/L) represents the low-effect phase of E-R curves (minimal outcome associations), Q2-Q3 (0.32–0.48 mg/L) corresponds to the steep-effect phase (accelerated outcome changes), and Q4 (> 0.48 mg/L) reflects the plateau phase (saturated effects). This alignment ensures that the quartile analysis captures nonlinearity to some extent, complementing the linear assumption used in mediation analyses. Compared to Q1, Q4 was significantly associated with an extended gestational age (β = 0.11 weeks, 95% CI: 0.10, 0.12; P < 0.001; Table S6) and an increased birth weight (β = 19.86 g, 95% CI: 15.39, 24.33; P < 0.001; Table S7). The birth weight Z-score also exhibited a significant positive association with Q4 (β = 0.01, 95% CI: 0.005, 0.02; P = 0.004; Table S8). In comparison to Q1, Q4 was significantly associated with reduced risks of preterm birth (OR = 0.93, 95% CI: 0.89, 0.97; P < 0.001; Table S9) and LBW (OR = 0.89, 95% CI: 0.85, 0.94; P < 0.001; Table S10). Conversely, Q4 was associated with significantly increased risks of macrosomia (OR = 1.48, 95% CI: 1.43, 1.52; P < 0.001; Table S11), SGA (OR = 1.17, 95% CI: 1.12, 1.22; P < 0.001; Table S12), and LGA (OR = 1.13, 95% CI: 1.10, 1.15; P < 0.001; Table S13).

The sensitivity analysis further demonstrated that the significant association between fluoride exposure and birth outcomes remained statistically robust, even after excluding participants with extreme maternal nutritional status—specifically, those with a BMI in early pregnancy of less than 18.5 kg/m² (underweight) and those with a BMI of 28 kg/m² or greater (obese) (Tables S14 and S15, and Figure S2). Notably, after excluding these groups, fluoride exposure was still associated with a 0.34-week increase in gestational age (95% CI: 0.31, 0.37; P < 0.001), a 36.49 g increase in birth weight (95% CI: 25.76, 47.22; P < 0.001), a 26% reduction in the risk of preterm birth (95% CI: 0.66, 0.82; P < 0.001), a 32% reduction in the risk of LBW (95% CI: 0.60, 0.78; P < 0.001), a 160% increase in the risk of macrosomia (95% CI: 2.41, 2.80; P < 0.001), an 81% increase in the risk of SGA (95% CI: 1.64, 1.99; P < 0.001), and a 26% increase in the risk of LGA (95% CI: 1.19, 1.33; P < 0.001). These findings suggest that the relationships between fluoride exposure and birth outcomes are independent of extreme maternal nutritional status.

To assess the potential impact of unmeasured confounding variables, we calculated the E-value for each association between fluoride exposure and birth outcomes (Table S16). The E-values for the key associations were significantly greater than 1, suggesting that it is highly improbable for any unmeasured confounder—such as unaccounted dietary factors or environmental exposures—to possess sufficient strength to completely attenuate or reverse the observed associations between fluoride and birth outcomes.

Discussion

This population-based study systematically evaluated the associations between drinking water fluoride exposure and birth outcomes among Chinese pregnant individuals. Our findings indicate a positive correlation between fluoride exposure in drinking water and both gestational age and birth weight. Specifically, each unit increase in fluoride concentration is associated with a 0.36-week increase in gestational age and a 43.56 g increase in birth weight. Notably, fluoride exposure demonstrated dual effects on birth outcomes: protective associations were observed for preterm birth and low birth weight, while concurrently elevating the risks of macrosomia, SGA, and LGA. Dose-response analyses further identified nonlinear exposure-effect relationships for most outcomes, suggesting potential threshold effects or biological saturation. Collectively, these findings suggest that fluoride plays a dual role in prolonging gestation while potentially disrupting fetal growth homeostasis. This illustrates the complex and multifaceted influences on fetal development in China, as fluoride not only extends gestation but also disrupts growth homeostasis through distinct mechanistic pathways. The observed duality of effects underscores the need for regionally tailored fluoride management strategies during pregnancy.

Our findings demonstrate a positive association between fluoride exposure and increased birth weight, aligning with multiple international cohort studies while revealing important nuances related to exposure sources and specific populations. The Swedish NICE study reported comparable effects, indicating that each 1 mg/L increase in urinary fluoride corresponds to an 84 g (95% CI: 30, 138) increase in birth weight and a 39% higher risk of LGA (95% CI: 1.03, 1.89) [23]. Research conducted in Mexico identified a positive correlation between urinary fluoride levels and birth weight Z-scores during the first trimester (β = 0.79, 95% CI: 0.10, 1.48, P = 0.02) [24]. While these studies assessed total fluoride exposure through urinary biomarkers, our analysis focusing specifically on drinking water aligns with their findings, supported by evidence that water fluoride levels strongly predict biological fluoride concentrations during pregnancy [37]. This consistency is further reinforced by Iranian data, which show significantly higher birth weights (3201 g vs. 2729 g, P < 0.01) in areas with high fluoride (> 1.5 mg/L) compared to low fluoride (< 0.7 mg/L) water [25], and Californian research linking reductions in water fluoride to decreased birth weight [15].

Notably, the National Health and Nutrition Examination Survey (NHANES) study presents an exception, reporting inverse associations exclusively among Hispanic mothers [22], highlighting potential ethnic and contextual modifiers. While the increase in birth weight associated with fluoride may reduce the prevalence of LBW, it concurrently elevate the risks of macrosomia (OR = 2.56 in our data) and LGA—conditions associated with acute obstetric complications such as shoulder dystocia and birth trauma [38, 39], as well as long-term cardiometabolic disorders in offspring [40]. From a mechanistic perspective, fluoride’s thyroid-disrupting properties [41] and the established interplay between thyroid function and diabetes [42] suggest endocrine pathways that could mediate both fetal overgrowth and gestational metabolic dysfunction, which may explain the observed risk of growth imbalances. To address existing gaps, future investigations should prioritize elucidating ethnic disparities in fluoride susceptibility, trimester-specific exposure windows, and potential interactions with gestational diabetes [15]—a critical gap given fluoride’s dual growth-promoting and metabolic-perturbing effects.

Our findings demonstrate that fluoride exposure in drinking water is significantly associated with prolonged gestational age and a reduced risk of preterm birth. This supports evidence from a Massachusetts cohort study, which identified particularly strong protective effects against preterm birth when fluoridation was combined with prenatal dental care (aRR = 0.74; 95% CI: 0.57, 0.95) [26]. Mediation analysis revealed that gestational age serves as a critical mechanistic pathway, accounting for 80% of fluoride’s protective effect against LBW and 19% of its association with macrosomia risk. This indicates that fluoride reduces LBW risk primarily by prolonging gestation, thereby allowing additional developmental time, whereas its effect on macrosomia is likely mediated through alternative pathways, such as enhanced placental nutrient transfer or endocrine disruption. The mediation effect exceeding 100% for birth weight further implies synergistic interactions between gestational duration and other fluoride-induced growth mechanisms. However, this benefit appears to be counterbalanced by elevated risks of SGA infants, potentially reflecting fluoride-induced placental dysfunction during prolonged pregnancies. This association is concerning, given the heightened susceptibility of SGA neonates to hypoxia and perinatal asphyxia [38]. The dual nature of these effects underscores significant gaps in our understanding of the biological mechanisms by which fluoride modulates gestational duration. Future studies should integrate longitudinal hemodynamic data with placental omics to elucidate fluoride’s dual role in extending gestation and disrupting fetal growth.

Our dose-response analyses further contributed to the understanding of the complexity of fluoride’s effects. The nonlinear E-R relationships observed for most outcomes suggest potential dose-dependent biological saturation or toxicity effects. In contrast, the linear association identified for LBW risk may indicate effects driven by cumulative exposure without biological buffering capacity. These divergent patterns suggest distinct mechanistic pathways: nonlinear outcomes may involve receptor-mediated or homeostatic processes, while LBW risk is likely influenced by unbuffered cumulative fluoride exposure.

The most striking finding of this study is the paradoxical association between fluoride and both SGA and LGA, which challenges the conventional view of fluoride as a unidirectional modulator of fetal growth. This dual effect likely arises from distinct, and potentially competing, biological pathways disrupted by fluoride exposure. First, fluoride’s endocrine-disrupting properties may contribute to growth imbalances. Research has demonstrated that fluoride can alter thyroid hormone levels in pregnant individuals [16, 17], and thyroid function is critical for regulating fetal growth; both hypothyroidism and hyperthyroidism are linked to SGA and LGA, respectively [42]. Second, fluoride-induced oxidative stress and placental dysfunction may play significant roles. Oxidative stress is implicated in reduced placental nutrient transport, leading to SGA, as well as impaired placental insulin signaling, which can result in fetal hyperinsulinemia and LGA [21, 38]. This paradox has important clinical implications. While fluoride has been shown to reduce preterm birth and LBW, which are beneficial outcomes, its association with both SGA and LGA increases the risk of perinatal complications. SGA neonates face a higher mortality risk due to hypoxia and asphyxia [37], while LGA is associated with shoulder dystocia and birth trauma [39]. In the long term, both conditions are linked to cardiometabolic and neurodevelopmental disorders in offspring [40], underscoring the need for targeted monitoring of fetal growth in fluoride-endemic regions.

Our stratified analyses revealed distinct urban-rural disparities in the associations between fluoride exposure and birth outcomes, despite the absence of significant age-based effect modification. Notably, rural populations exhibited a stronger positive correlation between fluoride exposure and gestational age, while the association with birth weight did not differ significantly between the two groups. Additionally, urban populations showed no significant protective effect against preterm birth and no association with SGA risk. In contrast, rural populations demonstrated a significant protective effect of fluoride against preterm birth and a positive association with SGA risk. Furthermore, urban populations exhibited more pronounced associations between fluoride and macrosomia as well as LGA risk. This geographical divergence likely reflects systemic differences in water infrastructure; rural areas predominantly utilize variable-fluoride groundwater sources (shallow and deep wells), with shallow well water exhibiting a wide fluoride range (0.025-2.300 mg/L, with 31.76% exceeding 1 mg/L) and deep well water showing a narrower but still variable range (0.002-2.700 mg/L, with 3.82% exceeding 1 mg/L) [43]. In contrast, urban centers maintain stable fluoride concentrations through centralized treatment (< 1 mg/L) [44]. Furthermore, variations in maternal nutritional status and access to medical resources between urban and rural populations may also influence the effects of fluoride. These findings contrast with reports from high-fluoride regions (> 4 mg/L) in India and Africa, where fluoride exposure is consistently linked to adverse birth outcomes, such as increased LBW and preterm risk [45, 46]. This discrepancy highlights a dose-dependent duality in the effects of fluoride. In our study, which focused on moderate fluoride exposure (Jinan mean: 0.41 mg/L), the gestation-prolonging and LBW-reducing protective effects of fluoride appear to dominate. Conversely, in high-fluoride regions (> 4 mg/L), the toxic effects of fluoride, such as disruption of fetal growth homeostasis, become prominent, particularly when combined with maternal nutritional deficits that may further impair the body’s ability to mitigate fluoride-induced damage. Notably, this duality is also influenced by the triad of “infrastructure-nutrition-exposure”: rural areas, characterized by variable fluoride levels from groundwater sources, and urban areas, where stable fluoride concentrations are maintained through centralized treatment, exhibit divergent associations between fluoride and birth outcomes. This divergence likely arises because the stability of infrastructure modulates exposure consistency, while maternal nutritional status, often lower in rural settings, interacts with fluoride metabolism. This underscores the necessity for region-specific risk assessments that integrate environmental (water infrastructure), social (access to nutrition), and biological (fluoride dose) cofactors to fully elucidate fluoride’s perinatal effects.

This study has several limitations. Firstly, the observational nature of the research design inherently restricts the ability to establish causal relationships. Secondly, the mediation and sensitivity analyses were conducted using GLM with linear exposure assumptions, which contrast with the nonlinear E-R relationships observed for most outcomes. To address this limitation, we performed quartile-based sensitivity analyses aligned with E-R phases to partially capture nonlinearity. We framed the mediation results for strongly nonlinear outcomes as exploratory, with the consistency between mediation findings and quartile associations supporting the robustness of our core conclusions. However, future nonlinear mediation frameworks, such as spline-based models, could further validate these pathways. Additionally, our assessment of fluoride exposure was based on area-level annual average concentrations of drinking water fluoride rather than individual-level data on daily tap-water consumption. We were also unable to directly measure individual fluoride exposure levels and did not account for fluoride sources beyond drinking water, such as food and dental products. Furthermore, the fluoride concentrations in drinking water may have included measurement errors due to our reliance on annual averages. While we utilized annual mean fluoride concentrations during dry and wet seasons to mitigate short-term variability, inherent fluctuations—such as seasonal changes and rural groundwater instability—may have led to minor exposure misclassification. Future studies should incorporate more frequent sampling to refine exposure estimates. Although we geocoded participants based on their registered addresses, some individuals may have relocated during pregnancy, a factor we could not capture. Furthermore, we did not account for other potential sources of water, such as bottled and purified water. These limitations may contribute to discrepancies between the measured fluoride exposure levels and actual exposure. Our study involved a database analysis utilizing data from the Jinan Maternal and Child Health Information Management System. However, we could not obtain data on individuals who were not registered, which may introduce selection bias and obscure the associations between fluoride exposure and birth outcomes. We were unable to access data on maternal income, pre-existing chronic conditions (e.g., diabetes, hypertension), or co-contaminants in drinking water (e.g., arsenic, manganese, lead), which are recognized as potential confounders. Future studies that utilize prospectively collected data and incorporate these variables are essential for further disentangling these potential effects and strengthening the causal evidence regarding the relationship between fluoride in drinking water and birth outcomes.

Conclusion

This study provides evidence that prenatal fluoride exposure in drinking water has dual effects on birth outcomes in China. Our findings indicate that fluoride exposure is significantly associated with both prolonged gestational duration and increased birth weight, likely mediated by an extension of gestation. However, it is essential to remain vigilant regarding its potential link to fetal growth imbalances, such as macrosomia, LGA, and SGA, suggesting a complex disruption of growth homeostasis. We recommend enhancing the surveillance of drinking water quality while simultaneously monitoring health indicators for mothers and newborns, as well as developing targeted perinatal intervention strategies. Furthermore, longitudinal cohort studies are necessary to establish exposure thresholds that maximize gestational benefits while minimizing growth disturbances. This study underscores fluoride as both a modifiable protective factor for the prevention of prematurity and a potential environmental risk that necessitates targeted management during critical developmental windows.

Abbreviations

ACME

Average causal mediation effects

ADE

Average direct effects

Birth weight Z-score

Birth weight-for-gestational age Z-score

BMI

Body mass index

CI

Confidence intervals

DAG

Directed acyclic graph

E-R

Exposure-response

GLM

Generalized linear models

LBW

Low birth weight

LGA

Large-for-gestational age

NHANES

National Health and Nutrition Examination Survey

OR

Odds ratio

RCS

Restricted cubic spline

SD

Standard deviation

SGA

Small-for-gestational age

Authors’ contributions

L.Y.: Conceptualization, Formal analysis, Writing-original draft, Writing-review & editing. F.Y.: Data curation. X.P.: Data curation, Project administration. Z.M.: Data Curation. X.G.: Supervision. W.W.: Writing-review and editing, Conceptualization, Funding acquisition.

Funding

This work was supported by the National Natural Science Foundation of China (grant number 82373699 and 82574217).

Data availability

The data supporting the findings of this study are not publicly available due to concerns regarding data security; however, they are available from the corresponding author upon reasonable request.

Declarations

Ethical approval and consent to participate

The study was approved by the Ethical Review Committee of Center for Endemic Disease Control, Chinese Center for Disease Control and Prevention, Harbin Medical University (No. hrbmuecdc20250326). Since all participants’ data were anonymous, informed consent was not required.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.