This cross-sectional study used data derived from the NHANES. This nationally representative survey systematically collects comprehensive nutritional and health indicators of the American population through standardized physical examinations and interviews. The National Center for Health Statistics (NCHS) provides detailed documentation of NHANES study protocols, including standardized data collection manuals for questionnaires and laboratory tests, through its official portal at the Centers for Disease Control and Prevention.¹ Ethical clearance for the NHANES survey protocol was granted by the NCHS Research Ethics Review Board. Prior to enrollment, documented informed consent forms were voluntarily signed by participants or legally authorized representatives (17).

This study analyzed data from the NHANES 2003–2006 and 2017–2020 cycles. This selection was made to enhance the validity and generalizability of the statistical inferences, as these were the only cycles where data on both iron biomarkers and dietary iron intake were simultaneously available. The screening process is depicted in Figure 1. From an initial cohort of 36,030 participants, a total of 16,393 individuals with available data on at least one iron biomarker and BMI were enrolled. Subsequently, the study excluded 4,539 underage participants, 638 pregnant participants, and 5,762 participants with missing covariate data. As a result, a total of 5,454 participants were ultimately included in the final analysis. The complete-case analysis approach was adopted to avoid potential biases introduced by imputation methods.

Exposure variables were defined as dietary iron intake and indicators of iron status, including serum iron, TIBC, TSAT, and ferritin. Dietary iron intake was assessed using a 24-h dietary recall survey. The initial dietary recall interview was conducted in-person at the Mobile Examination Center, and the second occurred by telephone 3 to 10 days later, administered by trained interviewers. Dietary intakes were calculated using the standardized United States Department of Agriculture (USDA) Food and Nutrient Database for Dietary Studies. The average of the two 24-h dietary recall information was used for analysis. In cases where participants lacked the second dietary interview, data from the first dietary recall were utilized. Serum iron was measured via a meticulous three-step procedure using the ferrozine reagent on the Roche Cobas 6,000 analyzer series. In this process, Fe³⁺ was liberated from transferrin, reduced to Fe²⁺, and complexed with ferrozine to form a colorimetric product, enabling its quantitative determination. TIBC was determined indirectly using measurements of serum iron and unsaturated iron-binding capacity obtained by the Roche method. TSAT was computed by dividing the measured value of frozen serum iron by the calculated TIBC and then multiplying the result by 100. Ferritin measurement was carried out using a sandwich immunoassay.

BMI was calculated as weight in kilograms divided by height in meters squared (kg/m²). Both height and weight were measured using a calibrated weight scale and a stadiometer by trained examiners and recorders. Adult weight status was classified according to the World Health Organization (WHO) classification criteria, where non-overweight/obesity was defined as a BMI of < 24.9 kg/m², overweight as a BMI of ≥ 25 kg/m², and obesity as a BMI of ≥ 30 kg/m² (18).

Demographic and lifestyle factors used as covariates included age, sex, ethnicity, physical activity (PA), poverty–income ratio (PIR), and alcohol consumption. Physical activity was quantified as metabolic equivalent (MET) minutes of moderate-to-vigorous physical activity per week. Respondents were classified into three categories based on adult physical activity guidelines (19): highly active (≥1,200 MET-min/week, equivalent to ≥300 min/week of moderate-intensity or ≥150 min/week of vigorous-intensity activity), active (600–1,199 MET-min/week, equivalent to 150–299 min/week of moderate-intensity or 75–149 min/week of vigorous-intensity activity), and inactive (<600 MET-min/week). Poverty was measured using the PIR, which divided participants into low-income (PIR < 1.3) and high-income (PIR ≥ 1.3) categories. Alcohol consumption patterns were categorized into three distinct groups based on self-reported intake: non-drinkers were defined as individuals with no lifetime alcohol consumption; low-to-moderate drinkers, who consumed alcohol less than daily; and heavy drinkers, classified as individuals who consumed alcohol daily.

Clinical indicators linked to metabolic parameters were also included, such as diabetes status, total calorie intake, serum creatinine (SCR), alanine aminotransferase (ALT), aspartate aminotransferase (AST), total cholesterol (TC), and high-density lipoprotein cholesterol (HDL-C) (20–23). Diabetes mellitus was defined as a self-reported diagnosis via validated questionnaires or a glycosylated hemoglobin (HbA1c) level of ≥6.5%. The body roundness index (BRI), an indicator of visceral adiposity distribution, was calculated as: BRI = 364.2–365.5 × √[1 − (waist circumference/(2π))²/(0.5 × height)²], where waist circumference and height are in centimeters (24).

Given the critical role of dietary vitamin C intake in enhancing non-heme iron absorption and influencing iron metabolism (25), it was also included as a nutritional covariate. Total calorie intake was adjusted for, as it is a major determinant of body weight and may confound the association between iron status and obesity. Dietary vitamin C and calorie intake were quantified using 24-h dietary recall surveys. The analysis incorporated the mean of two 24-h dietary assessments. When participants lacked a second dietary recall, data from the first recall were used.

To evaluate the differences in various variables across weight status groups, we employed the following statistical methods: continuous variables were presented as weighted median values with interquartile ranges (IQRs), while categorical variables were described using weighted frequencies and percentages. Group differences for continuous variables were assessed using the design-based Kruskal–Wallis test, which accounts for the complexities of the survey design and ensures the accuracy and reliability of the results. For categorical variables such as ethnicity, alcohol consumption, and physical activity, Pearson’s chi-squared test with Rao–Scott adjustment was used. This adjustment is designed to address potential biases due to the survey design, thereby making the chi-squared test results more robust and reliable.

Associations between iron status, dietary iron intake, and BMI were evaluated using weighted linear regression models. Weighted logistic regression models were used to examine the relationship between iron status, dietary iron intake, and the risk of overweight/obesity. To explore potential non-linear relationships between iron status, dietary iron intake, and the risk of obesity/overweight, we used restricted cubic spline (RCS) models with four knots positioned at the 5th, 35th, 65th, and 95th percentiles of the exposure distribution. This configuration was selected to balance the smoothness of the fitted curve and avoid loss of precision due to overfitting (26).

Furthermore, stratified multivariate regression was utilized for subgroup analyses, adjusting for all covariates except the stratification variable. We performed subgroup analyses stratified by categorical variables, including sex, ethnicity, PIR, physical activity, alcohol use, and diabetes status. To evaluate subgroup heterogeneity in associations, interaction terms were assessed via the likelihood ratio test (LRT). Specifically, LRT compared models with and without interaction terms (between exposure and stratifying covariates). A significant p-value for interaction indicated that the alternative model with interaction terms fitted the data better, confirming the presence of an interaction effect. All statistical tests were two-tailed, with statistical significance defined as a p-value of <0.05 for primary analyses (excluding subgroup analyses). For subgroup analyses involving multiple comparisons across stratified variables, the Bonferroni correction was applied to adjust the significance threshold: the original α = 0.05 was divided by the total number of independent subgroup tests (n = 6), resulting in a corrected significance level of α’ ≈ 0.008. All analyses were performed using R software (https://www.r-project.org/; version 4.4.2).

Among the 5,454 enrolled adults, 1,345 (25.51%) were categorized as non-overweight/non-obese, 1,655 (30.93%) as overweight, and 2,454 (43.56%) as obese. Table 1 presents the baseline characteristics and weighted estimates of adults with or without overweight/obesity in this study. Iron biomarkers differed significantly across weight categories: serum iron and TSAT decreased progressively with higher adiposity, from a median of 90.0 μg/dL and 29.00% in the non-overweight/obesity group to 79.0 μg/dL and 25.00% in the obesity group (p < 0.001), while ferritin increased from 80.20 ng/mL (IQR, 39.60–143.00) to 116.00 ng/mL (IQR, 57.00–198.00) (p < 0.001). TIBC, however, showed no statistically significant variation (p = 0.204). In addition, demographic and metabolic differences (sex, age, ethnicity, PIR, body roundness index, physical activity, and dietary vitamin C intake) were statistically significant across weight categories (All p < 0.001). Individuals with overweight/obesity exhibited a higher prevalence of diabetes, higher SCR, ALT, TC, and lower HDL-C. Nevertheless, no differences were observed in alcohol consumption, dietary iron intake, or caloric intake among these groups (all p > 0.05).

Multivariable-adjusted logistic regression models assessed the odds of overweight or obesity (BMI ≥ 25 kg/m²) in relation to iron biomarkers (Table 2). After completely adjusting for covariates, higher dietary iron intake was associated with 2% lower odds of overweight/obesity (OR = 0.98; 95% CI: 0.96–1.00; p = 0.026). Transferrin saturation demonstrated consistent inverse associations in three models, with each 1% increase corresponding to a 2% reduction in odds in model 3 (OR = 0.98; 95% CI: 0.96–0.99; p = 0.003). TIBC showed a marginal association (OR = 1.00; 95% CI: 0.99–1.00; p = 0.003), as the effect estimate and its confidence interval encompassed the null value. Serum iron also exhibited a non-significant inverse trend (OR = 1.00; 95% CI: 0.99–1.00; P = 0.004), while serum ferritin levels were unassociated with overweight/obesity (P > 0.05).

Multivariable linear regression analyses (Table 3) revealed significant inverse associations between serum iron and TSAT with BMI across all models, whereas TIBC demonstrated a positive association. These relationships persisted in the fully adjusted model (model 3): each 1 μg/dL increase in serum iron was associated with a 0.02-unit decrease in BMI (β = −0.02; 95% CI: −0.03 to −0.02; p < 0.001), and each 1% increase in TSAT corresponded to a 0.09-unit reduction in BMI (β = −0.09; 95% CI: −0.11 to −0.06; p < 0.001). In the linear regression analysis examining the association between TIBC and BMI, a marginally statistically significant positive association was observed (β = 0.01, 95% CI: 0.00 to 0.02; p = 0.022). Higher dietary iron intake was also linked to lower BMI in model 3 (β = −0.06; 95% CI: −0.12 to 0.00; p = 0.045). In contrast, no significant association was observed between serum ferritin levels and BMI (P > 0.05).

As illustrated in Figure 2, non-linear associations between iron status biomarkers and overweight/obesity risk were evaluated using RCS regression models adjusted for age, sex, ethnicity, PIR, PA, alcohol use, diabetes status, total calorie intake, and vitamin C intake, SCR, ALT, AST, TC, and HDL-C. Notably, a threshold-dependent effect was observed, with a significant reduction in obesity risks when TSAT values ranged between 16.76 and 39.81% (p-overall < 0.001; p-non-linear < 0.001). An inverted U-shaped relationship was identified between SI and obesity risks (p-overall < 0.001; p-non-linear = 0.011), with the OR peaking at a turning point of 57.39 μg/dL. Beyond this level, the OR declined with increasing SI concentrations. Similarly, TIBC showed a non-linear trend (p-overall < 0.001; p-non-linear = 0.002), with the highest OR observed at approximately 385.86 μg/dL and decreasing risks thereafter. In addition, a continuously decreasing linear trend between total dietary iron intake and the prevalence of overweight/obesity was observed (p-overall < 0.015; p-non-linear = 0.421). No significant relationship was observed between ferritin and overweight/obesity (p-overall = 0.021; p-non-linear = 0.212).

Supplementary Tables 1, 2 present the results of subgroup analyses stratified by sex, ethnicity, PIR, PA, alcohol use, and diabetes status. After Bonferroni correction, no significant interaction effects were observed between iron biomarkers and stratified variables on either overweight/obesity risk or BMI in the multivariable-adjusted models (all interaction p-values > 0.008), indicating that the overall association pattern was consistent across the examined subgroups.