This study was a school-based longitudinal cohort conducted in Xiamen, Fujian Province, China. A multistage cluster sampling design was used to recruit school-aged children from four nine-year compulsory education schools. The initial recruitment phase was carried out in October 2017, during which approximately 2,298 students in grades 2 to 4 participated in baseline screening, including physical examinations and questionnaire surveys. According to the study protocol, children with severe systemic diseases (e.g., cardiovascular, hepatic, or renal diseases), diagnosed endocrine disorders, those who had already entered puberty, or those unwilling to participate in long-term follow-up and biospecimen collection were excluded. In May 2018, the longitudinal cohort was formally established, and a total of 772 children who met the inclusion criteria were enrolled. The participant selection process is illustrated in Supplementary Figure 1. The year 2018 was defined as the baseline for the present analysis. Participants were followed annually from 2018 to 2020, with each follow-up visit including standardized physical examinations, questionnaire surveys, and biological sample collection. All data were collected according to uniform standardized procedures with strict quality control.

The study was conducted in accordance with the Declaration of Helsinki and was approved by the Medical Ethics Committee of Peking University (approval number: IRB00001052-17026). Written informed consent was obtained from all participants and their legal guardians prior to enrollment.

Urine samples were collected by trained research staff at schools using polypropylene containers free of phthalates. Samples were immediately stored under low-temperature conditions and transported to the laboratory, where they were stored at -80°C until analysis. Phthalate metabolites were measured at the Peking University Medical Laboratory using an ultra-performance liquid chromatography-tandem triple quadrupole mass spectrometry (UPLC-MS/MS) method. This method has been validated and widely applied in population-based studies and is known for its high sensitivity and reproducibility^[20-22].

Seven commonly detected phthalate metabolites were measured, including monomethyl phthalate (MMP), monoethyl phthalate (MEP), mono-n-butyl phthalate (MBP), monoisobutyl phthalate (MiBP), mono-2-ethylhexyl phthalate (MEHP), mono-2-ethyl-5-hydroxyhexyl phthalate (MEHHP), and mono-2-ethyl-5-oxohexyl phthalate (MEOHP). Sample pretreatment and quality control procedures, including enzymatic deconjugation, purification, and the use of blank and quality control samples, were conducted according to previously validated protocols. The limits of detection (LOD) and intra- and inter-batch coefficients of variation for all metabolites met established analytical quality control standards^[23,24]. Concentrations below the LOD were imputed as LOD divided by the square root of two. Detailed information on LOD values and detection frequencies is provided in Supplementary Table 1.

To account for inter-individual variability in urinary dilution, a covariate-adjusted standardization approach proposed by O’Brien et al. was applied at the data processing and statistical analysis^[25]. In this approach, urinary metabolite concentrations were adjusted using regression-based correction incorporating urinary creatinine and key demographic covariates (e.g., age, sex, and body mass index), rather than simple creatinine normalization, to reduce potential bias when creatinine is associated with metabolic outcomes^[26]. In addition, a sensitivity analysis was conducted using conventional creatinine correction (ng/mmol creatinine) based on three-year averaged concentrations, with identical modeling procedures. Based on molecular weight and alkyl chain structure, phthalate metabolites were further classified into low-molecular-weight phthalates (LMWP; MMP, MEP, MBP, and MiBP) and high-molecular-weight phthalates (HMWP; primarily DEHP metabolites, including MEHP, MEHHP, and MEOHP). Molar concentrations within each group were summed to represent overall exposure and used in subsequent analyses. This grouping approach has been widely used in environmental epidemiology to evaluate exposure mixtures while acknowledging heterogeneity in biological effects across individual metabolites. Because urinary phthalate metabolites have relatively short biological half-lives and exhibit substantial within-person variability, concentrations measured at each annual visit (2018-2020) were first processed separately and then averaged across the three survey waves to characterize habitual exposure levels. This approach has been widely adopted in environmental epidemiological studies to reduce random measurement error associated with single-spot urine samples.

At each follow-up visit, participants underwent standardized physical examinations and fasting venous blood collection to assess cardiometabolic risk factors. All physical examinations were conducted at schools by trained research staff following uniform protocols. Height and weight were measured using portable stadiometers and calibrated electronic scales, respectively. Waist circumference was measured at the end of normal expiration at a standardized anatomical site using a non-elastic tape. Blood pressure was measured with a standard mercury sphygmomanometer. All anthropometric and blood pressure measurements were performed twice, and the average values were used for analysis.

Fasting venous blood samples were collected between 7:00 and 9:00 a.m. after at least 10 h of overnight fasting. Samples were promptly centrifuged to separate serum and stored at -80 °C until analysis. Fasting blood glucose (FBG), triglycerides (TG), and high-density lipoprotein cholesterol (HDL-C) were measured using an automated biochemical analyzer, with all assays conducted in accordance with standardized laboratory quality control procedures.

Cardiometabolic risk factors were defined based on established literature and the latest Chinese pediatric diagnostic guidelines for cardiovascular health^[27-30]. Central obesity was defined as waist circumference at or above the age- and sex-specific 90th percentile. Elevated blood pressure was defined as systolic and/or diastolic blood pressure at or above the age- and sex-specific 95th percentile. Hypertriglyceridemia was defined as TG ≥ 1.47 mmol/L, low HDL-C as HDL-C < 1.03 mmol/L, and elevated fasting glucose as FBG ≥ 5.6 mmol/L. The presence of any one of these five abnormalities at a given follow-up visit was considered indicative of abnormal cardiometabolic risk status. This definition was intended to capture early-stage metabolic dysregulation in children, where even a single abnormality may signal increased future cardiometabolic vulnerability.

Based on repeated assessments across follow-up visits, participants were further classified into five cardiometabolic risk progression patterns: (1) Participants who exhibited no abnormalities at all visits were classified as the stable healthy pattern. (2) Those with abnormalities at baseline or early follow-up who subsequently reverted to and maintained a normal status were classified as the improving pattern. (3) Participants without abnormalities at baseline who developed abnormalities during follow-up that persisted thereafter were classified as the worsening pattern. (4) Participants whose cardiometabolic status alternated between healthy and abnormal states across visits, without a sustained improving or worsening trend, were classified as the fluctuating pattern. (5) Participants who presented with at least one abnormality at baseline and at all subsequent visits were classified as the persistently abnormal pattern. This classification framework integrates information across multiple time points and captures the dynamic evolution of cardiometabolic risk during childhood. Compared with single-time-point assessments, such pattern-based approaches provide more informative insights into early cardiometabolic risk accumulation and its potential long-term implications.

All covariates were collected using standardized questionnaires administered at baseline and during follow-up visits, completed jointly by children and their parents or legal guardians. Potential confounding variables included child sex (boy/girl), age (years), place of residence (urban/rural), only-child status (yes/no), mode of delivery (vaginal delivery/cesarean section), highest parental educational attainment (junior high school or below/high school or technical secondary school/college or above), and family history of cardiometabolic diseases (yes/no). These variables were selected based on prior literature suggesting their potential associations with cardiometabolic risk and environmental exposure patterns. In addition, pubertal development at baseline was assessed during routine school health examinations by trained clinicians using Tanner staging, based on the evaluation of secondary sexual characteristics and pubic hair development.

Lifestyle factors included physical activity, sedentary behavior, and sleep duration. Relevant information was collected using structured questionnaires at each follow-up visit. Physical activity was assessed as the average daily time spent in moderate-to-vigorous physical activity (MVPA). Sedentary behavior was evaluated based on daily recreational screen time, and sleep duration was calculated as the average nightly sleep time. To account for potential dietary confounding, several diet-related variables were also collected, including sugar-sweetened beverage intake, snack consumption, fried food consumption, and meat intake. Participants reported their average intake over the past seven days. Each variable was categorized into three levels (0 = low intake, 1 = moderate intake, and 2 = high intake) according to predefined intake ranges and included as ordinal covariates in sensitivity analyses.

Based on these three behavioral components, a composite indicator reflecting adherence to the 24-h movement guidelines was constructed. Participants were considered adherent if they simultaneously met all three recommendations: at least 60 min of MVPA per day, no more than 2 h of recreational screen time per day, and age-specific recommended sleep duration (9-11 h per night for school-aged children and 8-10 h per night for adolescents)^[16,31]. This composite indicator was used to characterize overall lifestyle patterns and to evaluate potential effect modification in the association between phthalate exposure and cardiometabolic risk progression patterns.

Continuous variables were summarized as mean ± standard deviation or median (interquartile range), depending on distributional characteristics, and categorical variables were presented as counts and percentages. Differences in baseline characteristics between boys and girls were assessed using the t-test or Wilcoxon rank-sum test for continuous variables and the χ² test for categorical variables.

Urinary phthalate metabolite concentrations were right-skewed and therefore natural log-transformed prior to analysis. To better reflect long-term exposure and reduce short-term variability inherent to spot urine samples, metabolite concentrations measured across the three follow-up waves (2018-2020) were averaged to construct three-year mean exposure indices. These averaged concentrations were further categorized into quartiles (Q1-Q4), with the lowest quartile serving as the reference group.

Multinomial logistic regression models were used to examine associations between phthalate exposure and cardiometabolic risk progression patterns, with the stable healthy pattern as the reference outcome. Odds ratios (ORs) and 95% confidence intervals (CIs) were estimated for the improving, fluctuating, worsening, and persistently abnormal patterns. Three models with increasing levels of adjustment were constructed: Model 1 was unadjusted; Model 2 adjusted for child sex, age, and residence; and Model 3 further adjusted for family history of cardiometabolic diseases, only-child status, mode of delivery, and parental educational attainment. Baseline cardiometabolic status was not included as a covariate because it is inherently incorporated into the definition of progression patterns, and adjusting for it would result in over-adjustment.

To evaluate potential sex-specific differences, both sex-stratified analyses and formal interaction tests (exposure × sex) were conducted based on Model 3. In addition, interaction terms were included to assess potential effect modification by adherence to the 24-h movement guidelines. The statistical significance of interaction terms was evaluated using Wald tests based on the multinomial logistic regression models. Weighted quantile sum (WQS) regression was further applied to assess the overall association between mixed phthalate exposure and cardiometabolic risk progression patterns. Phthalate metabolites were categorized into quartiles before inclusion in the WQS model. The dataset was randomly divided into training (40%) and validation (60%) subsets, and 1,000 bootstrap samples were used to estimate stable mixture weights. A positive-direction constraint was specified based on prior evidence suggesting predominantly adverse metabolic effects of phthalate exposure. The WQS index was incorporated into a multinomial logistic regression framework and adjusted for the same covariates as in the single-exposure models^[18]. Given the relatively modest sample size, WQS results should be interpreted as exploratory.

All statistical analyses were conducted using R (version 4.3.2). All tests were two-sided, and P values < 0.05 were considered statistically significant.

Table 1 summarizes the demographic and family characteristics of the study participants, which were generally comparable between boys and girls. The median age was slightly higher in boys than in girls (9.4 vs. 8.4 years, P < 0.001). No significant sex differences were observed for place of residence, only-child status, mode of delivery, parental educational attainment, or family history of cardiometabolic diseases (all P > 0.05). The distribution of cardiometabolic risk progression patterns during follow-up is presented in Figure 1. The stable healthy pattern was the most prevalent in both sexes. Although some differences in the distribution of progression patterns between boys and girls were observed descriptively, these differences were not formally tested and should be interpreted with caution. The baseline prevalence of individual cardiometabolic risk components is shown in Supplementary Table 2. Elevated fasting glucose was relatively uncommon in this cohort (1.2%), whereas central obesity was the most prevalent component (11.2%).

The associations between individual phthalate exposures and cardiometabolic risk progression patterns are presented in Figure 2, with detailed estimates from different adjustment models provided in Supplementary Tables 3 and 4. Results from Models 1 and 2 are shown in Supplementary Figure 2, whereas Figure 2 presents the fully adjusted results from Model 3. Overall, the most consistent associations were observed for the persistently abnormal pattern, while associations with the improving and fluctuating patterns were generally weaker and less consistent across metabolites. In Model 3, most phthalate metabolites showed weak or inconsistent associations with the improving and fluctuating patterns, with odds ratios close to 1 and confidence intervals including the null value. These findings suggest limited evidence of stable associations for these patterns.

In contrast, more pronounced associations were observed for the persistently abnormal pattern. Higher exposure levels of several high-molecular-weight phthalate metabolites were significantly associated with increased odds of this pattern. For example, MEOHP exposure in the second and fourth quartiles was associated with higher odds of the persistently abnormal pattern (Q2 vs. Q1: OR = 3.132, 95% CI: 1.484-6.609, P = 0.003; Q4 vs. Q1: OR = 2.837, 95% CI: 1.311-6.136, P = 0.008). Similar associations were observed for MEHHP, with both Q2 and Q4 exposures significantly increasing the odds (Q2 vs. Q1: OR = 2.172, 95% CI: 1.027-4.594, P = 0.042; Q4 vs. Q1: OR = 2.508, 95% CI: 1.179-5.333, P = 0.017). In addition, MEP exposure in the third quartile was associated with increased odds of the persistently abnormal pattern (Q3 vs. Q1: OR = 2.031, 95% CI: 1.012-4.075, P = 0.046).

For mixed exposure metrics, higher cumulative exposure to high-molecular-weight phthalates (ΣHMWP) was also associated with the persistently abnormal pattern. Specifically, participants in the second and fourth quartiles had increased odds compared with those in the lowest quartile (Q2 vs. Q1: OR = 2.763, 95% CI: 1.302-5.864, P = 0.008; Q4 vs. Q1: OR = 2.258, 95% CI: 1.029-4.954, P = 0.042). Results from sensitivity analyses, including additional dietary adjustment, alternative model specifications, and alternative exposure categorization to address sparse data issues, were generally consistent with the primary findings (Supplementary Tables 5-7). Sensitivity analyses using conventional creatinine correction showed similar effect directions, although several estimates were attenuated and confidence intervals were wider. Detailed results are presented in Supplementary Table 8.

For the worsening pattern, statistically significant associations were observed for only a limited number of metabolites, and the directions of association were inconsistent. Higher exposure levels of MBP and MEHHP were associated with lower odds of the worsening pattern (MBP Q2 vs. Q1: OR = 0.583, 95% CI: 0.343-0.990, P = 0.046; MEHHP Q3 vs. Q1: OR = 0.542, 95% CI: 0.305-0.962, P = 0.037). However, most metabolites and mixed-exposure indices did not show consistent associations with this pattern, and these findings should be interpreted with caution.

Sex-stratified analyses based on the fully adjusted model (Model 3) are presented in Figure 3, and quartile-specific interaction estimates (exposure × sex) are provided in Supplementary Table 9. Overall, several positive associations between phthalate exposure and the persistently abnormal cardiometabolic risk pattern were observed among boys, whereas corresponding associations among girls were generally weaker and less consistent.

Among boys, higher exposure levels of several phthalate metabolites were associated with increased odds of the persistently abnormal pattern. For example, MEP exposure in the third quartile was associated with higher odds (Q3 vs. Q1: OR = 2.417, 95% CI: 1.074-5.438, P = 0.033). Similarly, MEOHP exposure across the second to fourth quartiles showed consistently elevated odds (OR range: 3.414-5.698, all P ≤ 0.012). Comparable patterns were also observed for MiBP, with Q2-Q4 exposure groups demonstrating increased odds of the persistently abnormal pattern (OR range: 2.505-4.179, all P ≤ 0.038). For mixed exposure indices, ΣHMWP exposure in the second quartile was positively associated with this pattern (Q2 vs. Q1: OR = 3.342, 95% CI: 1.430-7.809, P = 0.005).

Among girls, most associations between phthalate exposure and the persistently abnormal pattern were not statistically significant, and effect estimates were generally less stable. Formal interaction tests indicated limited statistical evidence for effect modification by sex for most metabolites [Supplementary Table 9]. Therefore, these findings should be interpreted with caution.

For the worsening pattern, only a small number of inverse associations were observed among girls, and no consistent patterns were identified across exposure metrics.

As shown in Figure 4, adherence to the 24-h movement guidelines varied considerably among participants. Only 23.1% of children met all three recommendations for physical activity, recreational screen time, and sleep duration, whereas 5.6% met none. The remaining participants met either one (29.8%) or two (42.8%) components.

Potential effect modification by adherence to the 24-h movement guidelines was evaluated in the fully adjusted model (Model 3) by including multiplicative interaction terms between phthalate exposure quartiles and adherence status. Overall, limited evidence of interaction was observed for several high-molecular-weight phthalate metabolites in relation to the persistently abnormal cardiometabolic risk pattern [Figure 5 and Supplementary Table 10].

Specifically, interaction estimates for some DEHP-related metabolites (e.g., MEOHP, MEHHP, and ΣHMWP) suggested differential associations according to adherence to the 24-h movement guidelines. However, several interaction estimates were accompanied by wide confidence intervals or extreme point estimates, likely reflecting limited sample sizes within certain exposure-behavior strata. Therefore, these findings should be interpreted cautiously.

Similar patterns were observed in additional analyses evaluating individual components of the 24-h movement guidelines separately [Supplementary Table 10], although the estimates were generally imprecise and did not consistently support strong effect modification.

Weighted quantile sum (WQS) regression analyses were conducted to further evaluate associations between phthalate mixture exposure and cardiometabolic risk progression patterns [Figure 6]. In the fully adjusted model, the WQS index was significantly associated with the persistently abnormal cardiometabolic risk pattern. Compared with the stable healthy pattern, each one-quartile increase in the mixture index was associated with higher odds of the persistently abnormal pattern (OR = 1.56, 95% CI: 1.05-2.33, P = 0.029).

The estimated mixture weights suggested that MiBP, MBP, and MEOHP contributed relatively more to the overall association. However, these weights should be interpreted cautiously because WQS regression prioritizes identifying dominant contributors within a constrained mixture model rather than estimating independent causal effects.

No statistically significant associations were observed between the mixture index and the improving, fluctuating, or worsening patterns. Overall, the mixture analysis was consistent with the single-exposure results and supports a potential role for phthalate co-exposure in the accumulation of persistent cardiometabolic risk.