Incremental value of <scp>DXA</scp> ‐derived fat distribution over anthropometrics for classifying metabolically unhealthy obesity: A population‐based analysis

Anthropometric indices such as body mass index (BMI), waist circumference (WC), and waist-to-hip ratio (WHR) are surrogates of body composition that incompletely capture central/ectopic adiposity and cardiometabolic risk heterogeneity.1 Visceral adipose tissue (VAT) is more strongly linked to insulin resistance, dyslipidaemia, and hypertension than subcutaneous adipose tissue (SAT), yet routine clinic measures do not accurately quantify these compartments.2 Consequently, individuals with similar BMI or WC can differ markedly in metabolic health, including the distinction between metabolically healthy obesity (MHO) and metabolically unhealthy obesity (MUO).3 Distinguishing these two subtypes is important as individuals with MHO generally exhibit a more favourable cardiometabolic profile, with lower risk of cardiovascular disease, and better response to therapies compared with those with MUO. Dual-energy x-ray absorptiometry (DXA) provides low-dose quantification of VAT and SAT and may complement simple anthropometrics for risk stratification.4 However, the incremental value of DXA-derived compartments beyond BMI, WC, and WHR in population-based data remains uncertain. This study sought to evaluate whether adding DXA-derived VAT and SAT measurements improves the classification of MUO beyond standard anthropometric models that reflect routine clinical practice (BMI, waist circumference, and waist-to-hip ratio). We conducted a cross-sectional, population-based analysis using the National Health and Nutrition Examination Survey (NHANES) 2011–2016 examination cycles, a nationally representative survey with standardized comprehensive whole-body DXA and cardiometabolic phenotyping. This study used de-identified, publicly available data from the Centers for Disease Control and Prevention. Participants with complete data on anthropometry, DXA fat compartments, and metabolic markers were included. Participants with missing anthropometric, DXA, or biochemical data were excluded; no imputation was performed. While the definition of MUO has varied significantly across studies, among participants with obesity (BMI ≥30 kg/m2), MUO was defined as meeting at least one of the following metabolic risk criteria, to capture early metabolic dysfunction3, 5: systolic or diastolic blood pressure ≥130/80 mmHg, HbA1c ≥6.5%, triglycerides ≥150 mg/dL, or HDL cholesterol <40 mg/dL (men)/<50 mg/dL (women). Participants with obesity who met none of these criteria were classified as MHO. We fitted nested survey-weighted logistic regression models for MUO (vs MHO): (1) BMI alone; (2) BMI + WC; (3) BMI + WC + WHR; and (4) anthropometrics + DXA-derived VAT, SAT, and the VAT/SAT ratio. DXA-derived models included VAT, SAT, and the VAT–SAT ratio in order to capture both absolute and relative fat compartment contributions. This sequential approach reflects the hierarchical manner in which adiposity metrics are typically applied in clinical and epidemiologic assessment; from general to regional fat measures. Model discrimination was assessed using the area under the receiver operating characteristic curve (AUC) with 95% CIs (DeLong), with variance estimates incorporating the complex survey design. Baseline characteristics were compared across ethnic groups using survey-weighted ANOVA for continuous variables and χ2 tests for categorical variables, accounting for NHANES sampling. All analyses were performed in R 4.5.1 using appropriate packages. Of a total of 39 156 participants in the 2011–2016 cycle, 15 217 with complete DXA, biochemistry and anthropometric measures were included. In this cohort, 4188 participants had obesity (BMI ≥30). Sensitivity analysis showed that obesity prevalence varied by ethnicity (White 27.9%, Hispanic 30.2%, Black 33.4%, Other/Multi 14.0%; p < 0.001). Among participants with obesity, the prevalence of MHO was similar by sex (female 24.2% vs. male 22.3%, p = 0.18). Anthropometric and compartmental adiposity differed by ethnicity: White participants had the highest WC and VAT, while Black participants had the highest SAT and the lowest VAT/SAT ratio. The prevalence of MHO and MUO varied by ethnicity, with lower proportions in the Other/Multi group. Table 1. The prevalence of MUO increased with age: 40.8% at 18–29 years, 53.8% at 30–39 years, 65.2% at 40–49 years, and 76.3% at 50–59 years (p < 0.001). Across ethnic groups, WC correlated more strongly with VAT (r = 0.79–0.83) than BMI (r = 0.71–0.80), all p < 0.01. For MUO classification among participants with obesity, discrimination improved stepwise across nested models: BMI alone (AUC 0.543 [95% CI 0.502–0.583]), BMI + WC (0.594 [0.555–0.632], p < 0.01 vs. BMI alone), BMI + WC + WHR (0.631 [0.592–0.669], p < 0.01 vs. BMI + WC), and anthropometrics + DXA (0.706 [0.672–0.741], p < 0.01 vs. BMI + WC + WHR); the DXA-augmented model's CIs did not overlap those of BMI alone. Continuous NRI indicated an overall 48.3% net improvement in correct classification (95% CI 42.1–54.6%, p < 0.001). After adjustment for age and sex, discrimination improved modestly (+0.03–0.05 across simpler models) but remained highest for the DXA-augmented model (AUC 0.73) in comparison to BMI alone. There was no evidence of problematic multicollinearity among VAT, SAT, and VAT/SAT ratio (variance inflation factors <2 for all). In stratified analyses, discrimination with the DXA model was consistent across ethnic subgroups, with AUCs ranging from 0.66 (Black, lowest) to 0.74 (Hispanic, highest). Overall adjusted cohort model AUCs are provided in Figure 1. This population-based analysis demonstrates that DXA-derived measures of visceral and subcutaneous adiposity provide incremental discrimination for metabolically unhealthy obesity beyond conventional anthropometric indices. While the absolute improvements in model performance were modest, they were consistent across subgroups and clinically meaningful, particularly in identifying individuals who may be misclassified using traditional metrics such as BMI, waist circumference, or waist-to-hip ratio alone. These findings highlight the limitations of anthropometric measures in capturing heterogeneity in adipose tissue distribution and underscore the potential value of more refined body composition assessment in cardiometabolic risk stratification. Although DXA is not routinely used in clinical practice, it has been extensively validated against CT for quantifying visceral and subcutaneous fat and offers a robust and reproducible method for assessing regional adiposity across diverse populations.6 Several limitations warrant consideration. First, the cross-sectional design precludes causal inference or assessment of longitudinal transitions between MUO and MHO. Second, inclusion was restricted to participants with complete DXA data, introducing potential selection bias, as individuals undergoing imaging may differ systematically from those without such data. Third, the definition of MUO used in this study, based on the presence of at least one metabolic abnormality, represents a relatively permissive criterion and may overestimate MUO prevalence compared with stricter definitions requiring two or more abnormalities. However, this approach is consistent with prior population-based studies and aligns with established frameworks that aim to capture early metabolic dysfunction. Finally, although DXA is more accessible and involves lower radiation exposure than CT or MRI, it remains less widely available than routine anthropometric measures, which may limit immediate clinical implementation. DXA-derived measures of visceral and subcutaneous adiposity consistently improve discrimination of MUO beyond traditional anthropometrics. These gains are clinically meaningful, particularly for identifying individuals whose cardiometabolic risk may be underestimated by conventional measures. Our findings support the use of refined body composition assessment in research and selected clinical settings and highlight the need for longitudinal studies to determine their prognostic value. Shaun Khanna: Conceptualization; methodology; formal analysis; data curation; software; validation; visualization; writing—original draft; writing—review and editing; project administration. Nelson Wang: Methodology; formal analysis; supervision; writing—review and editing. Aditya Bhat: Methodology; data interpretation; writing—review and editing. Clare Arnott: Conceptualization; supervision; funding acquisition; writing—review and editing. Nitesh Nerlekar: Conceptualization; methodology; supervision; data interpretation; writing—review and editing. The authors acknowledge the National Health and Nutrition Examination Survey (NHANES), conducted by the National Center for Health Statistics of the Centers for Disease Control and Prevention, for providing the publicly available data used in this study. The authors also acknowledge the contributions of the NHANES participants and staff whose efforts made this research possible. None. The authors declare no conflicts of interest. The examination survey obtained institutional ethics approval and participant consent; this secondary analysis of de-identified public data was considered exempt/not human research by our institution. The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/dom.70479. Publicly available de-identified survey data and documentation can be accessed via the CDC repository. The data that support the findings of this study are publicly available from the National Health and Nutrition Examination Survey (NHANES) at https://www.cdc.gov/nchs/nhanes/. All data used in this analysis are publicly accessible and require no special permissions.