Document resource
WHAT IS ALREADY KNOWN ON THIS TOPIC Environmental exposure to toxic metals such as cadmium, lead, and mercury has been linked to impaired glucose metabolism and increased risk of type 2 diabetes. However, most studies have focused on single metals or simple mixtures, and little is known about the role of metal-related inflammatory indices in diabetes risk prediction.WHAT THIS STUDY ADDS This study demonstrates that the Metal Mixture Inflammatory Index (MMII) is independently associated in a linear fashion with diabetes risk in US adults using nationally representative National Health and Nutrition Examination Survey data from 1999 to 2020. We also developed and internally validated a least absolute shrinkage and selection operator-based nomogram incorporating urinary metals and conventional risk factors, which showed excellent discrimination and calibration.HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY Our findings highlight mixed-metal exposures as a modifiable but under-recognized contributor to diabetes risk. The MMII may serve as a practical biomarker for risk stratification, while incorporating metal exposures into prediction models could improve screening and prevention. Policymakers may also consider stricter regulation of metal contamination to reduce future diabetes burden.Introduction Diabetes mellitus (DM) is increasing at an unprecedented pace. The most recent edition of the International Diabetes Federation Atlas estimates that 537 million adults worldwide were living with diabetes in 2021, and projects that number will rise to 783 million by 2045, with the steepest growth expected in low-income and middle-income nations. 1 This expansion already translates into more than US$900 billion in annual healthcare expenditures and has widened treatment gaps in resource-constrained settings.2 Although population aging, obesity, and sedentary behavior remain dominant drivers, marked regional heterogeneity in incidence suggests that additional, modifiable risk factors merit attention.Over the past decade, environmental epidemiology has identified non-essential metals as credible metabolic disruptors. Prospective and cross-sectional research links higher cadmium, lead, and mercury burdens to impaired glucose tolerance, incident type 2 diabetes, and faster progression of diabetic complications.3 4 Mixture-oriented analyses strengthen this view: weighted-quantile-sum and Bayesian kernel-machine models applied to recent National Health and Nutrition Examination Survey (NHANES) waves show that combined urinary metals dominated by cadmium, lead, and thallium correspond with abnormal glucose metabolism and metabolic syndrome features.5 6 Experimental data provide plausible mechanisms—these metals accumulate in β-cells, generate mitochondrial reactive oxygen species, activate the nuclear factor kappa B (NF-κB) pathway and the NOD-like receptor family pyrin domain containing 3 (NLRP3) inflammasome, and disrupt amino acid and lipid pathways integral to insulin action.7 8Despite this progress, few studies have evaluated mixture-based inflammatory indices in relation to diabetes or incorporated metal biomarkers into individual risk-prediction tools. The Metal Mixture Inflammatory Index (MMII) was recently shown to track systemic inflammation and predict all-cause mortality in US adults,9 yet its association with diabetes and its incremental value over traditional risk factors remains uncertain. Emerging machine-learning work suggests that adding urinary metals can improve cardiometabolic risk stratification beyond conventional predictors.10 11 Against this background, we analyzed NHANES data spanning 1999–2020 to quantify the association between MMII and prevalent diabetes, characterize the dose–response relationship and subgroup heterogeneity and develop and internally validate a least absolute shrinkage and selection operator (LASSO)-based nomogram that integrates metal exposures with established risk factors to improve diabetes risk assessment.Materials and methods Study design and population NHANES is a continuous, biennial, cross-sectional survey designed to provide nationally representative estimates of the health and nutritional status of the non-institutionalized US. population, using a complex, multistage probability sampling design. 12 Conducted by the National Center for Health Statistics (NCHS), the program collects detailed data through standardized interviews, physical examinations, and laboratory tests. The collected data encompass sociodemographic information, dietary intake, physical and laboratory measurements, and self-reported health history.13We analyzed data from the NHANES spanning 1999–2020.14 Of the 107 622 initially eligible participants, 83 837 were excluded due to missing MMII data, leaving 23 785 participants with complete MMII measurements. Subsequently, 497 individuals with incomplete diabetes information were excluded. Thus, the final analytical sample comprised 23 288 adults aged ≥18 years with available MMII and diabetes data. We did not exclude participants with renal insufficiency; instead, serum creatinine (SCR) was incorporated as a covariate in multivariable models to minimize potential confounding (figure 1).Figure 1Participant selection process in NHANES 1999–2020. NHANES, National Health and Nutrition Examination Survey; MMII, Metal Mixture Inflammatory Index.Assessment of the MMII The systemic inflammatory burden from concurrent metal exposure was quantified with the MMII following Wang et al.15 In that work, reduced rank regression was first applied to 10 urinary metals to find the linear combination that explained the largest proportion of variance in C reactive protein and the platelet-to-lymphocyte ratio. A subsequent stepwise procedure showed that six elements—mercury, cadmium, cobalt, molybdenum, lead, and tungsten—captured almost all of the inflammation-related signal, and only these were retained in the final algorithm. Consistent with the published procedure, we Z-transformed the six metal concentrations and calculated MMII as their weighted sum using the reported regression coefficients; higher scores denote greater proinflammatory potential. The index has been validated in earlier NHANES analyses and correlates positively with systemic inflammatory markers and all-cause mortality.15Definition of DM DM was defined according to the diagnostic criteria of the American Diabetes Association (ADA). Participants were classified as having diabetes if they met any of the following conditions: self-reported physician diagnosis of diabetes, current use of insulin or oral glucose-lowering medications, fasting plasma glucose≥126 mg/dL, or glycated hemoglobin≥6.5%. These criteria are consistent with the ADA Standards of Medical Care in Diabetes.16Covariate selection and assessment Covariates were selected based on prior evidence linking demographic, socioeconomic, behavioral, clinical, and dietary factors to diabetes risk. All data were collected through standardized physical examinations or structured interviews in accordance with the NHANES protocol. Race/ethnicity was categorized as Mexican American, non-Hispanic white, non-Hispanic black, or other. Educational attainment was classified as less than high school, high school graduate or general educational development, and college or above. Marital status was recorded as married or living with a partner, single, widowed, divorced, or separated. The poverty-to-income ratio (PIR) was calculated by dividing household income by the federal poverty threshold, adjusted for household size and inflation. Behavioral factors included smoking and alcohol use. Smoking status was self-reported as never (fewer than 100 cigarettes in a lifetime), former (more than 100 cigarettes but not currently smoking), or current. Alcohol use was defined as current or non-current, with non-current drinkers including those who had consumed fewer than 12 drinks in their lifetime or none in the past year. Physical activity was assessed through a detailed questionnaire covering transport-related walking or cycling, household tasks, muscle-strengthening exercises, occupational activity, and leisure-time sports. Total weekly physical activity was expressed as the sum of metabolic equivalent minutes across activity categories. Clinical variables included body mass index (BMI), SCR, total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C). Height and weight were measured using standardized anthropometric procedures, and BMI was calculated as weight divided by height squared (kg/m²). Blood pressure was measured in accordance with the standardized NHANES protocol by trained personnel using a mercury sphygmomanometer with an appropriate cuff size, after the participant had been seated quietly for at least 5 min. Three or more measurements were obtained, and the average of available readings was used in the analysis. Dietary intake was assessed using 24-hour dietary recall interviews conducted by trained interviewers with the United States Department of Agriculture (USDA) Automated Multiple-Pass Method, and nutrient intakes—including total energy, protein, carbohydrates, saturated fat, and dietary fiber—were calculated based on the USDA Food and Nutrient Database for Dietary Studies. 17Statistical analysis All statistical analyses followed the NHANES analytic guidelines, incorporating sampling weights, strata, and primary sampling units to account for the survey’s complex, multistage probability design. 18 Participants were categorized into diabetes and non-diabetes groups. The distribution of continuous variables was examined using the Shapiro-Wilk test and Q–Q plots. Normally distributed variables were presented as weighted means±SD and compared using Student’s t-test, whereas non-normally distributed variables were expressed as weighted medians (IQR) and compared using the Mann-Whitney U test. Categorical variables were summarized as weighted frequencies (percentages) and compared using the χ2 test. Categorical covariates were modeled as dummy variables, while continuous covariates were modeled in their original continuous form. Given the small numerical values of MMII, we multiplied the values by 10 to facilitate interpretation. The rescaled MMII was evaluated both as a continuous variable and as a quartile-based categorical variable derived from the weighted population distribution: Q1 (MMII < –2.087), Q2 (–2.087≤MMII < 0.022), Q3 (0.022≤MMII < 2.007), and Q4 (MMII≥2.007), with Q1 serving as the reference group.Survey-weighted logistic regression models were used to estimate ORs and 95% CIs for the association between MMII and diabetes risk across three models: an unadjusted (crude) model; model 1 adjusted for age, sex, race/ethnicity, PIR, education level, and marital status; and model 2 further adjusted for smoking status, alcohol use, physical activity, diastolic blood pressure (DBP), systolic blood pressure (SBP), SCR, BMI, HDL-C, LDL-C, TC, TG, and dietary intakes of energy, protein, carbohydrates, saturated fat, and dietary fiber. Restricted cubic spline (RCS) regression was applied to explore potential nonlinear relationships between MMII and diabetes risk, with knots placed at the 10th, 50th, and 90th percentiles of MMII. Subgroup analyses were conducted according to sex, age (<60 or ≥60 years), race/ethnicity, smoking status, alcohol use, and hypertension status; interaction terms were tested using likelihood ratio tests.To identify the most predictive urinary metals and address potential multicollinearity among correlated covariates, we applied the LASSO regression. Although the number of predictors was moderate, LASSO remains advantageous because it performs simultaneous variable selection and shrinkage, thereby limiting overfitting and enhancing model stability. 10-fold cross-validation was used to select the penalization parameter (λ), and a more conservative λ within 1 SE of the minimum deviance was chosen to optimize generalizability.19 A nomogram was constructed based on LASSO-selected predictors to estimate individual diabetes risk. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration plots, and decision curve analysis (DCA). Missing covariate data (<5%) were imputed using predictive mean matching.20 All analyses were conducted using R software (V.4.3.2), and two-sided p values<0.05 were considered statistically significant.Results Baseline characteristics of the study population A total of 23 288 participants were included in the analysis, of whom 2837 (12.2 %) had diabetes. Compared with individuals without diabetes, those with diabetes were older, had higher BMI, blood pressure (SBP and DBP), SCR, LDL-C, TG, and MMII scores, and exhibited lower HDL-C levels (all p<0.001). The diabetes group also showed lower total energy, protein, carbohydrate, and saturated fat intake (all p<0.05). No significant difference was observed in dietary fiber intake between groups. Sociodemographic patterns differed between groups. A higher proportion of individuals with diabetes were widowed, divorced, or separated, had lower educational attainment, and reported former smoking. In contrast, non-diabetic participants were more likely to be never smokers and more physically active. The distribution of race/ethnicity also differed (p<0.001), with a greater proportion of non-Hispanic black participants in the diabetes group. No significant differences were observed in alcohol consumption between groups ( table 1).Table 1Baseline characteristics of participants according to diabetes statusVariablesTotalNon-DMDMP valuen23 28820 4512837Age, years39.26±0.2636.99±0.2658.33±0.38<0.001Gender, n (%)0.005 Female12 008 (53.18)10 636 (53.65)1372 (49.23) Male11 280 (46.82)9815 (46.35)1465 (50.77)Race, n (%)<0.001 Mexican American4470 (10.40)3920 (10.39)550 (10.52) Non-Hispanic black5636 (12.69)4905 (12.40)731 (15.13) Non-Hispanic white8592 (62.82)7624 (63.34)968 (58.49) Other race4590 (14.09)4002 (13.87)588 (15.86)Education level, n (%)<0.001 Below high school6594 (19.33)5596 (18.55)998 (25.85) High school graduate or GED5563 (24.12)4887 (23.99)676 (25.20) Some college or above11 131 (56.56)9968 (57.46)1163 (48.95)Marital status, n (%)<0.001 Married or living with a partner9595 (49.60)7914 (48.48)1681 (59.03) Never married10 270 (34.96)9946 (37.67)324 (12.14) Widowed, divorced, or separated3423 (15.44)2591 (13.85)832 (28.83)PIR2.81±0.032.83±0.032.61±0.05<0.001DBP, mm Hg68.28±0.1967.94±0.1971.17±0.44<0.001SBP, mm Hg118.31±0.20116.88±0.20130.33±0.68<0.001Physical activity total METs/week3738±683787±713327±1530.004SCR, mg/dL0.861±0.0020.852±0.0010.943±0.009<0.001BMI, kg/m227.15±0.0826.48±0.0932.76±0.20<0.001HDL-C, mg/dL53.73±0.1954.43±0.2047.91±0.50<0.001LDL-C, mg/dL91.52±0.4490.91±0.4796.64±1.16<0.001TC, mg/dL187.22±0.45186.98±0.48189.25±1.420.14TG, mg/dL213.10±2.01211.29±2.08228.33±6.030.01Energy intake, kcal2130.38±8.752151.14±9.571955.83±26.93<0.001Protein intake, g80.18±0.4280.48±0.4677.71±1.050.02Carbohydrates intake, g/day259.25±1.13262.93±1.22228.33±3.24<0.001Saturated fat intake, g/day27.11±0.1727.30±0.1925.51±0.46<0.001Dietary fiber intake, g/day16.08±0.1316.02±0.1416.58±0.310.1Rescaled MMII (×10)0.133±0.033−0.031±0.0361.518±0.076<0.001Smoking, n (%)<0.001 Former5503 (23.93)4590 (22.78)913 (33.58) Never13 763 (56.88)12 317 (57.78)1446 (49.30) Now4022 (19.19)3544 (19.44)478 (17.12)Drinking, n (%)0.68 Current alcohol user9659 (54.18)8265 (54.11)1394 (54.78) Non-current alcohol user13 629 (45.82)12 186 (45.89)1443 (45.22)Continuous variables are expressed as weighted means±SD or weighted medians (IQR), depending on distribution. The normality of continuous variables was assessed using the Shapiro-Wilk test. Normally distributed variables were compared using Student’s t-test, and non-normally distributed variables were compared using the Mann-Whitney U test. Categorical variables are summarized as weighted frequencies (percentages) and compared using the χ2 test.BMI, body mass index; DBP, diastolic blood pressure; DM, diabetes mellitus; GED, general educational development; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; MET, metabolic equivalent; MMII, Metal Mixture Inflammatory Index; PIR, poverty income ratio; SBP, systolic blood pressure; SCR, serum creatinine; TC, total cholesterol; TG, triglycerides.Association between MMII and risk of diabetes Higher MMII was associated with an increased risk of diabetes across all models. In the fully adjusted model, each 10-fold increase in MMII was associated with a 2% higher odds of diabetes (OR 1.02, 95% CI 1.00 to 1.04; p=0.02). When MMII was modeled in quartiles, the association exhibited a clear dose–response trend (p for trend=0.003). Compared with the lowest quartile (Q1), participants in Q2, Q3, and Q4 had progressively higher odds of diabetes, with fully adjusted ORs of 1.21 (1.00–1.47), 1.23 (1.02–1.48), and 1.26 (1.04–1.52), respectively ( table 2).Table 2Logistic regression models for the association between MMII and the risk of diabetesCharacteristicCrude modelModel 1Model 2OR (95% CI)P valueOR (95% CI)P valueOR (95% CI)P valueRescaled MMII (×10)1.23 (1.22 to 1.25)<0.0011.04 (1.02 to 1.05)<0.0011.02 (1.00 to 1.04)0.020Q1ReferenceReferenceReferenceQ22.87 (2.44 to 3.39)<0.0011.36 (1.13 to 1.62)<0.0011.21 (1.00 to 1.47)0.046Q34.83 (4.12 to 5.65)<0.0011.43 (1.20 to 1.71)<0.0011.23 (1.02 to 1.48)0.028Q47.26 (6.23 to 8.47)<0.0011.5 (1.27 to 1.79)<0.0011.26 (1.04 to 1.52)0.016P for trend<0.001<0.0010.003Crude model: unadjusted. Model 1: adjusted for age, gender, race, PIR, education level, and marital status. Model 2: further adjusted for smoking status, alcohol use, physical activity, DBP, SBP, SCR, BMI, HDL-C, LDL-C, TC, TG, and dietary intake of energy, protein, carbohydrates, saturated fat, and dietary fiber.BMI, body mass index; DBP, diastolic blood pressure; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; MMII, Metal Mixture Inflammatory Index; PIR, poverty-income ratio; SBP, systolic blood pressure; SCR, serum creatinine; TC, total cholesterol; TG, triglycerides.Dose–response association between MMII and diabetes risk The RCS analysis revealed a linear positive association between rescaled MMII and diabetes risk in the overall population (p for overall<0.001; p for non-linearity=0.461; figure 2A). The association remained consistent across sex-specific analyses (figure 2B). Among men, higher MMII was significantly associated with increased diabetes risk (p for overall=0.003), whereas no significant association was observed in women (p for overall=0.631). No evidence of non-linearity was detected in either subgroup.Figure 2Dose–response association between rescaled MMII and diabetes risk in the overall population (A) and by sex (B), based on restricted cubic spline analysis. ORs and 95% CIs were estimated using survey-weighted logistic regression models adjusted for demographic, lifestyle, and clinical covariates. Shaded areas represent 95% CIs. MMII, Metal Mixture Inflammatory Index.Subgroup analyses In subgroup analyses, the positive association between rescaled MMII and diabetes risk was generally consistent across most strata, with significant associations observed in men (OR 1.04, 95% CI 1.01 to 1.07; p=0.005), non-Hispanic white participants (OR 1.08, 95% CI 1.05 to 1.12; p<0.0001), former smokers (OR 1.08, 95% CI 1.04 to 1.12; p<0.0001), current alcohol users (OR 1.03, 95% CI 1.00 to 1.06; p=0.028), and participants without hypertension (OR 1.04, 95% CI 1.00 to 1.07; p=0.032). Significant interactions were detected for age (p for interaction=0.027) and hypertension status (p for interaction<0.0001). No significant associations were observed in women or other race/ethnicity groups ( online supplemental figure 1).SP110.1136/bmjdrc-2025-005366.supp1Supplementary dataLASSO-based variable selection and development of the diabetes risk prediction model A LASSO-penalized logistic regression model was developed to predict diabetes risk based on a comprehensive set of candidate variables. Specifically, the model incorporated 28 predictors, including 22 covariates from model 2 and 6 individual urinary metals (mercury, cadmium, cobalt, molybdenum, lead, and tungsten), which replaced the composite MMII Score. As shown in online supplemental figure 2) LASSO regression introduced an L1 penalty term to the loss function, shrinking less informative coefficients toward zero to enable variable selection and reduce multicollinearity. The coefficient path plot (online supplemental figure 2A) visualizes how each variable enters or exits the model as the penalty parameter (λ) changes. The optimal λ was chosen via 10-fold cross-validation (online supplemental figure 2B), balancing model fit and parsimony. Thirteen variables with non-zero coefficients were retained: tungsten exposure, lead exposure, molybdenum exposure, cadmium exposure, dietary fiber intake, carbohydrate intake, HDL-C, TC, SCR, SBP, BMI, PIR, and age. These predictors were incorporated into a nomogram to estimate individualized diabetes risk (online supplemental figure 3A). Model discrimination was excellent, with an area under the receiver operating characteristic curve (AUC) of 86.89% (95% CI 86.29% to 87.49%) as shown in the ROC curve (online supplemental figure 3B). Calibration performance was also strong: the calibration curve closely followed the ideal diagonal, with a mean absolute error of 0.01 and a mean squared error of 0.00027 (online supplemental figure 4A), indicating high agreement between predicted and observed probabilities. The DCA further demonstrated the clinical utility of the model: across the full range of risk thresholds, the LASSO-based model consistently yielded the highest standardized net benefit (online supplemental figure 4B). Although net benefit declined gradually as the risk threshold increased, it remained superior to both the “all” and “none” strategies throughout.SP210.1136/bmjdrc-2025-005366.supp2Supplementary dataSP310.1136/bmjdrc-2025-005366.supp3Supplementary dataSP410.1136/bmjdrc-2025-005366.supp4Supplementary dataDiscussion Principal findings In this nationally representative cross-sectional study, we found that higher MMII scores were positively associated with increased diabetes risk in US adults. The association was linear across the full range of MMII values and remained robust after adjusting for demographic, behavioral, clinical, and dietary factors. Subgroup analyses revealed that the association was more pronounced among men, non-Hispanic white participants, former smokers, current alcohol users, and individuals without hypertension. Using LASSO regression, we further identified 13 key predictors of diabetes risk, including four urinary metals (tungsten, lead, molybdenum, cadmium), dietary factors (fiber and carbohydrate intake), clinical markers (HDL-C, TC, SCR, SBP, BMI), PIR, and age. A nomogram-based prediction model incorporating these variables demonstrated excellent discrimination (AUC=86.89%) and good calibration. The DCA indicated favorable clinical utility across a wide range of risk thresholds. Together, these findings suggest that metal-related inflammatory burden may contribute to diabetes risk and highlight the potential of integrating metal exposures into diabetes risk prediction models.Comparison with previous literature Our findings align with—and extend—a growing body of evidence linking both single-metal and mixture exposures to disordered glucose metabolism. Prospective data from China first showed dose-dependent increases in fasting glucose and incident diabetes across cadmium quartiles, 21 while a large occupational cohort confirmed higher diabetes incidence with combined cadmium-related and lead-related work exposures.22 Mixture-oriented approaches have reached similar conclusions: in NHANES 2015–2016, Bayesian kernel-machine, weighted-quantile-sum, and LASSO models all demonstrated a positive joint effect of urinary metals—dominated by cadmium and thallium—on abnormal glucose metabolism,6 and an Environmental Risk Score derived from 18 blood and urinary metals in NHANES 2003–2014 associated higher cumulative exposure with greater odds of type 2 diabetes.23 Beyond cross-sectional analyses, a prospective study of mid-life US women reported that higher combined metal burdens predicted incident metabolic syndrome, a key diabetes precursor.24 Metal-specific cohorts reinforce these trends: adipose cadmium predicted future diabetes in southern Spain,25 and a 2021 meta-analysis estimated a 13% rise in type 2 diabetes risk for each doubling of cadmium exposure.26 Mixed-metal effects are not confined to general populations; among Mexican-American adults, urinary arsenic, molybdenum, and copper jointly disrupted glycemic traits even in non-diabetics.27 Recent mechanistic work further implicates chronic cadmium-driven oxidative stress and inflammation in the excess mortality observed among US adults with diabetes and pre-diabetes.7 Finally, studies of broader metabolic endpoints show that tungsten, molybdenum, and thallium materially contribute to metal-mixture associations with metabolic-associated fatty liver disease—another phenotype tightly coupled to insulin resistance.28 Collectively, these reports corroborate our observation that a higher MMII—reflecting the pro-inflammatory potential of a cadmium-enriched, lead-enriched, molybdenum-enriched, and tungsten-enriched mixture—confers excess diabetes risk and supports the integration of metal mixture metrics into modern cardiometabolic risk models.Biological pathways linking metals, clinical factors, and diet to diabetes risk Multiple, non-mutually exclusive pathways may explain how a proinflammatory metal mixture promotes diabetes. First, cadmium, lead, and mercury accumulate in pancreatic β-cells, where they increase mitochondrial reactive oxygen species production, collapse membrane potential, and blunt glucose-stimulated insulin secretion; mouse and in vitro data show marked mitochondrial bioenergetic failure after low-micromolar Cd, Pb, or Hg exposure. 29 30 The resulting oxidative stress activates NF-κB and the NLRP3 inflammasome, amplifying islet inflammation and accelerating β-cell apoptosis, phenomena documented in longitudinal mixture studies of mid-life women and in animal models.31 32 Second, untargeted metabolomics of adult NHANES participants reveals that a cadmium-enriched, lead-enriched, and tungsten-enriched signature perturbs branched-chain amino acid, fatty acid, and sugar pathways that are tightly coupled to insulin signaling.33 Third, whole-life metal exposure reshapes the gut microbiota: high urinary metal loads correlate with reduced short-chain-fatty-acid producers and greater adiposity and hyperglycemia in multicenter cohorts, and mechanistic reviews implicate cadmium-induced dysbiosis in metabolic derailment.34 35 Epigenetic reprogramming provides an additional layer—meta-analyses show prenatal methyl-mercury alters cord-blood DNA-methylation at loci governing glucose and lipid metabolism, with signatures persisting into childhood.36 At the organ level, exposome studies demonstrate that chronic mixed-metal stress impairs endocrine-pancreas plasticity and β-cell compensation to metabolic demand, predisposing to later dysglycemia.37 Systemically, tungsten-driven and cadmium-driven alterations in plasma lipid handling intersect with high-fat-diet signals, intensifying hepatic steatosis and peripheral insulin resistance in murine “multihit” models.33 38 Finally, selenium excess illustrates a complementary mechanism: misincorporation of selenocysteine into the insulin receptor destabilizes ligand binding and precipitates insulin resistance in silico and cellular assays.39 In addition to individual metal toxicities, metal mixtures are hypothesized to exert joint effects on diabetes risk through both convergent and complementary pathways. Convergent effects arise when multiple metals concurrently induce oxidative stress, inflammation, and β-cell dysfunction through shared molecular targets (eg, ROS production, NF-κB activation).40 Complementary effects occur when toxic metals (eg, Cd, Pb, Hg) impair insulin secretion or promote β-cell damage, while disruption of essential trace metals (eg, Zn, Se, Mn, Cu) further compromises antioxidant defenses and insulin signaling, amplifying metabolic dysregulation.41 Emerging epidemiological evidence and mixture-specific models (eg, weighted quantile sum (WQS) regression and quantile g-computation (QGC)) also support that combined exposure to multiple metals elevates diabetes and metabolic disease risk beyond individual effects.42 In our fully adjusted models, the attenuation—but not elimination—of the metal–diabetes association following adjustment for elevated blood pressure and dyslipidemia suggests that these metabolic dysfunction markers may partially mediate or modify the relationship. Blood pressure elevation and lipid abnormalities are known to co-occur with oxidative stress, endothelial dysfunction, and inflammatory processes, which are also implicated in metal toxicity and glucose dysregulation.43 Concurrently, impaired renal function may reduce the excretion of metals, increasing systemic metal burden; oxidative stress is a recognized contributor to kidney injury, which further feeds into this cycle.44 On the dietary side, higher intake of dietary fiber has been inversely associated with the risk of type 2 diabetes in cohort studies and meta-analyses.45 Fiber may blunt postprandial glucose spikes, improve insulin sensitivity, reduce systemic inflammation, and modulate gut microbiota.46 These mechanisms suggest that diet quality may buffer or interact with metal-induced metabolic stress. Taken together, these observations underscore that clinical status and dietary exposures do not merely confound but may modulate the strength of associations between metal mixture inflammation and diabetes risk.Heterogeneity of the MMII–diabetes relationship across subgroups Sex-stratified toxicokinetic data suggest that men accumulate higher body burdens of lead and cadmium because of greater occupational contact and lower iron stores, which may partly explain why the MMII–diabetes association was significant in men but not in women; similar sex differences have been reported for cadmium-related renal and metabolic outcomes in NHANES and other cohorts. 47 48 The stronger effect seen in non-Hispanic white participants mirrors prior mixture studies showing that this group experiences the steepest exposure–response slope for metal-driven metabolic dysfunction, possibly owing to distinctive diet and residential sources of tungsten and molybdenum.48 Former smokers showed the clearest association, consistent with evidence that cadmium continues to leach from bone and lung long after cessation, so that smoking history remains a dominant determinant of internal dose even when current exposure is absent.49 Alcohol may exacerbate metal toxicity by up-regulating hepatic cytochrome P-450 enzymes and depleting glutathione; an occupational cohort found a synergistic rise in diabetes incidence when heavy drinking co-occurred with mixed-metal exposure, supporting the interaction we observed.49 Finally, the null finding in participants with hypertension accords with mechanistic work showing that chronic vascular inflammation and oxidative stress from elevated blood pressure may mask or dilute additional inflammatory signals attributable to metals, whereas subjects without hypertension retain sufficient metabolic reserve for the incremental effect to manifest.23 Collectively, these patterns highlight how sex-specific toxicodynamic, behavioral reservoirs of exposure (smoking, drinking), and pre-existing cardiometabolic conditions can modulate the diabetogenic impact of metal mixtures.Clinical and public health implications Our findings suggest that the MMII could serve as a practical biomarker for identifying adults at heightened risk of diabetes. Because MMII is derived from routinely collected urinary metals, it could be incorporated into existing screening platforms alongside traditional metabolic markers. At a population level, the linear dose–response we observed reinforces the need for stricter regulation of cadmium, lead, tungsten, and molybdenum in consumer products, soil, and drinking water. Targeted mitigation—such as remediating contaminated housing stock, reducing dietary sources of heavy metals, and monitoring high-risk occupations—may help lower community‐wide inflammatory burdens and curb diabetes incidence.Strengths and limitations of the study Key strengths include the large, nationally representative NHANES sample, comprehensive adjustment for demographic, lifestyle, clinical, and dietary covariates, and the use of validated mixture and machine-learning methods to capture complex exposure patterns. Nonetheless, the cross-sectional design precludes causal inference, spot urine measurements may misclassify long-term metal exposure, and residual confounding from unmeasured environmental or genetic factors cannot be excluded. Self-reported behaviors such as smoking and alcohol use remain susceptible to misclassification. Moreover, the large sample size increases statistical power, so even modest effect sizes may reach statistical significance; therefore, our findings were interpreted in terms of effect estimates and CIs rather than p values alone, and caution is warranted when considering their clinical significance.Future research directions Prospective studies with repeated metal measurements are needed to clarify temporality and dose–response kinetics. Although the observed per-unit effect sizes of MMII are modest, future research should evaluate whether these small individual-level associations accumulate to meaningful risks at the population level. Validation of the nomogram in independent populations with available metal exposure data, particularly Asian cohorts, is an essential next step, along with comparison against established diabetes risk scores to determine clinical utility.Conclusion In a nationally representative US cohort, the MMII was independently associated with higher diabetes risk. At the same time, a LASSO-based model identified four key urinary metals together with nine conventional risk factors, yielding strong discrimination and calibration. These findings suggest that both mixed-metal exposure and established cardiometabolic determinants jointly shape diabetes susceptibility, underscoring the importance of incorporating environmental exposures alongside traditional risk factors in future risk-assessment frameworks.