Body Composition Phenotypes and Their Metabolic Associations in Treatment-Naïve Patients With Newly Diagnosed Type 2 Diabetes: A DXA-Based Cluster Analysis

Dual-energy X-ray absorptiometry identified four body-composition patterns among 185 people with newly diagnosed, treatment-naive type 2 diabetes in a cross-sectional study at Beijing Shijitan Hospital, with the clusters showing different associations with metabolic dysfunction-associated steatotic liver disease, hyperuricemia, bone mineral density and osteoporosis. Two adiposity-heavy clusters had markedly higher odds of steatotic liver disease than the low-fat, low-lean-mass group, while the latter had the highest observed osteoporosis prevalence. The investigators cautioned that the four-group solution represented a pragmatic partition of a continuum rather than definitive biological categories, particularly because two clusters were composed almost entirely of one sex.

The study recruited consecutive patients from January 2020 through June 2024. Participants were eligible if they were 18 to 80 years old, had type 2 diabetes and had not received glucose-lowering medication. Researchers excluded people with type 1, gestational or secondary diabetes, as well as conditions independently capable of disturbing body composition or bone metabolism, including active malignancy, end-stage organ failure and malnutrition.

The final clustering analysis included 185 participants after observations with missing body-composition data and statistical outliers were removed. Participants were also assessed for glucose regulation, pancreatic function, liver enzymes, kidney function, uric acid, blood pressure and several bone-related markers. The analytical model classified the cohort into four groups.

Cluster 1 contained 42 participants and combined low fat mass with low lean mass. Cluster 2 was the largest group, with 71 participants, and was 98.6% male; it combined preserved muscle mass with a high percentage of body fat and the highest android-to-gynoid ratio, indicating a central-adiposity pattern. Cluster 3 comprised 49 participants, 98.0% of them female, and had high adiposity with relatively low lean mass.

Because muscle function was not measured, the investigators did not characterize this pattern as a clinical diagnosis of sarcopenia. Cluster 4 contained 23 participants and had the highest total, fat and lean mass as well as the greatest visceral adipose tissue. K-means clustering performed better than hierarchical clustering in internal validation, but its silhouette width was only 0.272, indicating limited separation between groups.

All four clusters achieved bootstrap Jaccard indices above 0.85, while re-clustering with 30 less-correlated variables reproduced the original solution with an adjusted Rand index of 0.912. Sex was a dominant feature of the classification and complicated interpretation. Cluster 2 was overwhelmingly male and Cluster 3 overwhelmingly female, making phenotype and sex statistically difficult to separate in those groups.

The researchers therefore standardized every DXA variable separately within men and women and repeated the clustering analysis. This removed the original association between cluster membership and sex, with the adjusted Rand index falling from 0.378 to 0.002, although agreement with the initial four-cluster solution was moderate, at 0.351. The sex-standardized analysis retained a central-adiposity group with an android-to-gynoid ratio of 1.3 that remained associated with alanine aminotransferase, gamma-glutamyl transferase, hip T-score, blood urea nitrogen, HDL cholesterol, triglycerides, HOMA-insulin resistance, osteoporosis, steatotic liver disease and hyperuricemia.

Separate analyses within men and women also found differences in adiposity, lean mass and metabolic outcomes, supporting the presence of residual heterogeneity beyond simple sex differences. Variance decomposition indicated that cluster labels captured information beyond sex for most of the DXA variables examined. For total fat mass, adding cluster to a sex-only model increased explained variance by 0.746, whereas adding sex to a cluster-only model increased it by 0.002.

The corresponding additional cluster contribution for android fat mass was 0.692. Contributions were more evenly divided for total lean mass, with values of 0.251 for cluster and 0.104 for sex, and for the android-to-gynoid ratio, with 0.047 and 0.068, respectively. Because the cluster labels were generated directly from the DXA variables, these calculations quantify the relative contribution of labels and sex rather than independently validate the existence of discrete clusters.

Regression adjustment for age, sex, smoking and alcohol consumption changed the cluster contribution to uric acid, alanine aminotransferase, hip bone density and blood urea nitrogen by no more than 0.016 in R-squared. Significant sex-by-cluster interactions nevertheless emerged for hip T-score and Z-score, showing that some bone-density associations differed between men and women. The four phenotypes did not differ significantly in HbA1c, fasting glucose, post-load glucose or HOMA-beta, and the observed HOMA-insulin-resistance difference did not survive Bonferroni correction.

During 75-g oral glucose tolerance testing, peak insulin and C-peptide secretion occurred later than 60 minutes in every cluster. Cluster 4 had the highest insulin and C-peptide areas under the curve and Cluster 1 the lowest, but neither area-under-the-curve measure remained significant after correction. Liver and renal markers provided more pronounced differences.

Alanine aminotransferase and gamma-glutamyl transferase differed significantly after correction and were highest in Cluster 4, whereas aspartate aminotransferase did not. Blood urea nitrogen also varied across clusters, while creatinine, estimated glomerular filtration rate, urinary albumin-to-creatinine ratio and urinary albumin excretion did not, indicating that the renal signal did not extend to a broad difference in intrinsic kidney function. Hip and femoral-neck bone mineral density differed across clusters after correction, with Cluster 4 recording the highest values and Cluster 1 the lowest; lumbar-spine T- and Z-scores did not differ.

Osteoporosis was present in 46.3% of Cluster 1. In a logistic model that pooled 20 multiply imputed datasets and adjusted for age, sex, smoking and alcohol consumption, Cluster 4 had lower odds of osteoporosis than Cluster 1, with an odds ratio of 0.16 and a 95% confidence interval of 0.03 to 0.92. Age, male sex and smoking were also independent correlates in that model.

The authors described the pattern as possibly consistent with differences in mechanical loading and endocrine influences associated with muscle and fat mass, but the cross-sectional data could not establish a mechanism. None of the measured bone-turnover markers—N-mid osteocalcin, P1NP, 25-hydroxyvitamin D or parathyroid hormone—retained significance after Bonferroni correction, so the density differences did not establish a distinct bone-turnover mechanism. After adjustment for age, sex, smoking and alcohol consumption, Clusters 2 and 4 had sharply higher odds of metabolic dysfunction-associated steatotic liver disease than Cluster 1.

The pooled odds ratio was 9.69 for Cluster 2, with a 95% confidence interval of 3.34 to 28.09, and 8.55 for Cluster 4, with an interval of 2.18 to 33.51. Cluster 3 did not differ significantly from the reference group, with an odds ratio of 2.44 and p = 0.142. Cluster 4 also had higher odds of hyperuricemia, with an odds ratio of 6.36 and a 95% confidence interval of 1.42 to 28.57, as well as higher serum uric acid levels.

Cluster 2 also had elevated serum uric acid, although its hyperuricemia estimate did not meet the significance threshold at p = 0.066. In the univariate analyses, cluster membership was associated with osteoporosis, steatotic liver disease and hyperuricemia; diabetic kidney disease, diabetic retinopathy, lower-extremity arteriosclerosis, carotid plaque, hypertension, coronary artery disease and dyslipidemia did not show significant univariate cluster associations. The researchers used multiple imputation to address substantial gaps in complication data, including 43.5% missingness for diabetic retinopathy and 41.4% for carotid plaque.

Results from complete-case analyses were directionally consistent for the three outcomes that initially showed associations. The diabetic-retinopathy analysis nevertheless included only eight confirmed cases among 105 participants with known status, precluding meaningful inference from that comparison. In nested models, adding cluster membership to BMI alone, BMI plus waist circumference, or BMI plus visceral adipose tissue area improved model fit for steatotic liver disease and blood urea nitrogen.

Cluster membership did not add significant incremental value for the other examined outcomes, indicating that its additional information was limited in this treatment-naive, newly diagnosed cohort. Important limitations included the single-center design, the small Cluster 4, modest separation between clusters, absence of external validation, sex-cluster non-separability for Clusters 2 and 3, and reliance on ultrasound rather than biopsy for diagnosing steatotic liver disease. Taken together, the findings show that DXA-based body-composition profiling distinguished clinically different patterns among people at the time of type 2 diabetes diagnosis, but the evidence does not establish a new diagnostic taxonomy or a causal pathway.

The most pronounced adjusted associations were higher odds of steatotic liver disease in the central-adiposity and high-fat-high-lean groups, along with higher odds of hyperuricemia in the high-fat-high-lean group. The low-fat-low-lean group had the highest osteoporosis prevalence, whereas the high-fat-high-lean group had lower adjusted odds of osteoporosis. These relationships persisted through statistical adjustment and imputation, yet they intersected with pronounced sex differences and, in several analyses, wide confidence intervals.

The cross-sectional results cannot determine whether body-composition patterns preceded metabolic disease or its complications, and prospective multicenter studies with larger sex-specific samples, systematic complication assessment and longer follow-up would be required to evaluate their prospective value beyond BMI, waist circumference and visceral fat.

Subject of Research: DXA-derived body-composition patterns and their metabolic associations in newly diagnosed type 2 diabetes

Article Title: Body Composition Phenotypes and Their Metabolic Associations in Treatment‐Naïve Patients With Newly Diagnosed Type 2 Diabetes: A DXA‐Based Cluster Analysis

Article References: Zhang, X., Zhang, R., & Jiang, T. (2026). Body Composition Phenotypes and Their Metabolic Associations in Treatment‐Naïve Patients With Newly Diagnosed Type 2 Diabetes: A DXA ‐Based Cluster Analysis. Endocrinology, Diabetes & Metabolism, 9(6), Article e70366. https://doi.org/10.1002/edm2.70366

Image Credits: AI Generated

DOI: 10.1002/edm2.70366

Keywords: type 2 diabetes, DXA, body composition, visceral fat, lean mass, metabolic dysfunction-associated steatotic liver disease, hyperuricemia, osteoporosis, cluster analysis, sex differences, Body, Composition