Abstract
Background
Possible Sarcopenic obesity (PSO) carries higher risks than either condition alone, but whether its risk factors differ by sex and age remains unclear. We investigated the prevalence, risk factors, and sex/age differences of PSO using traditional regression, Bayesian analysis, and machine learning.
Methods
This cross-sectional study included 427 community-dwelling adults aged 60–98 years. PSO was defined as concurrent possible sarcopenia (AWGS 2019: low handgrip strength + slow gait speed) and central obesity (waist circumference ≥ 90 cm in men, ≥ 85 cm in women). Multivariable logistic regression, sex- and age-stratified analyses, dose–response analysis, mediation analysis, and E-value sensitivity analysis were performed. Bayesian hierarchical models and XGBoost with SHAP were used for validation.
Results
PSO prevalence was 8.7% (37/427). Overall, age (OR = 1.12, 95%CI:1.05–1.19) was a risk factor and hemoglobin (OR = 0.97, 95%CI:0.94–0.99) was protective. Sex-stratified analysis revealed marked differences: in men, BMI was a strong risk factor (OR = 1.54, 95%CI:1.19–2.01); in women, low hemoglobin (OR = 0.96, 95%CI:0.92–0.99) and high white blood cell count (OR = 1.43, 95%CI:1.04–1.98) were significant. Age-stratified analysis showed that in those < 75 years, glucose (OR = 1.26) and white blood cell count (OR = 1.23) were important; in those ≥ 75 years, BMI showed a strong effect (OR = 1.91, 95%CI:1.33–2.76). Bayesian analysis produced consistent findings with narrow credible intervals. The XGBoost model achieved good discrimination (AUC = 0.86), with SHAP confirming age, BMI, and hemoglobin as top predictors. Dose–response analysis showed a linear relationship between hemoglobin and PSO (p for nonlinearity = 0.743). E-value analysis confirmed robustness to unmeasured confounding (E-values ≥ 1.47).
Conclusions
Risk factors for PSO differ substantially by sex and age. These findings support targeted prevention strategies: weight control in men, nutritional and inflammatory monitoring in women, metabolic management in younger-old adults, and weight control in the oldest-old.
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1186/s12877-026-08088-0.
Keywords: Possible sarcopenic obesity, Sarcopenia, Aging, Sex differences, Bayesian analysis, Machine learning, Hemoglobin
Introduction
Sarcopenia, the age-related loss of muscle mass and function, affects a growing proportion of older adults worldwide [1]. When sarcopenia coexists with obesity, the condition is termed Possible sarcopenic obesity (PSO), and evidence suggests its clinical consequences—including cardiovascular disease, disability, and mortality—exceed those of either condition alone [2, 3]. Beyond demographic shifts, sarcopenia and obesity may coexist through interconnected biological pathways, including age-related hormonal changes (declining testosterone, growth hormone, and estrogen), chronic low-grade inflammation, insulin resistance, and reduced physical activity that creates a vicious cycle of muscle loss and fat accumulation. These interrelated mechanisms contribute to the disproportionate burden of metabolic and functional decline in older adults. As populations age, understanding the epidemiology and risk factors of PSO has become a public health priority.
Most studies on PSO have focused on Western populations, and data from Chinese community-dwelling older adults remain relatively limited [4]. Prevalence estimates vary widely depending on diagnostic criteria and population characteristics, ranging from 6 to 12% in Chinese samples [5, 6]. More importantly, whether risk factors for PSO differ between men and women or between younger and older age groups remains poorly understood. Some evidence suggests that men may be more affected by overall adiposity, while women may be more vulnerable to nutritional deficits and inflammation [7], but these observations have not been systematically tested using rigorous stratified analyses.However, whether these sex-specific patterns extend to sarcopenic obesity, and how they interact with age, has not been systematically quantified.
The present study extends previous evidence in several key aspects. First, we formally test sex- and age-based interactions using interaction terms in regression models, rather than relying on subgroup point estimates alone. Second, we validate the differential risk profiles using Bayesian hierarchical models that provide posterior probabilities for subgroup effects, thereby quantifying the uncertainty around these estimates. Third, we employ machine learning (XGBoost with SHAP) to confirm variable importance rankings without presuming linear relationships. Fourth, we complement the primary analysis with dose–response characterization (hemoglobin–PSO relationship), mediation analysis (inflammation as a potential mediator), and E-value sensitivity analysis to assess robustness to unmeasured confounding. This multi-method approach provides more rigorous evidence for sex- and age-specific risk stratification than previously available.
Hemoglobin, a simple and inexpensive marker of nutritional status, has been linked to sarcopenia in previous studies [8]. However, its relationship with PSO—particularly whether a threshold effect exists—has not been characterized. White blood cell count, a marker of systemic inflammation, may also play a role in PSO pathogenesis, but whether it mediates the effect of body mass index (BMI) on PSO has not been examined.Hemoglobin was selected as a candidate biomarker for several reasons. First, it reflects systemic oxygen-carrying capacity, which is critical for skeletal muscle metabolism and physical performance. Lower hemoglobin levels may impair oxygen delivery to muscle tissue, contributing to reduced mitochondrial oxidative capacity and accelerated muscle atrophy—both central features of sarcopenia. Second, hemoglobin is closely linked to nutritional status; low levels may indicate deficiencies in iron, vitamin B12, or folate, or reflect chronic disease and systemic inflammation. Third, inflammation is a key driver of both anemia (via hepcidin-mediated iron sequestration) and obesity-related metabolic dysfunction, making hemoglobin a potentially useful integrative marker of the intersection between nutritional, inflammatory, and metabolic pathways in PSO. Importantly, hemoglobin is a simple, inexpensive, and routinely available clinical measure, which makes it a practical screening tool in primary care and community settings.
Recent methodological advances offer new opportunities to deepen our understanding of PSO. Bayesian hierarchical models provide a framework for quantifying uncertainty and borrowing strength across subgroups, which is particularly valuable when sample sizes are limited [9]. Machine learning approaches, such as XGBoost with SHAP, complement traditional regression by identifying non-linear relationships and providing interpretable variable importance measures [10, 11]. XGBoost (eXtreme Gradient Boosting) is a tree-based ensemble algorithm that builds a sequence of decision trees, with each subsequent tree correcting errors from the previous ones. This approach captures non-linear relationships and complex interactions among predictors without requiring prespecified functional forms. SHAP (SHapley Additive exPlanations) is a game-theoretic framework that assigns each predictor a contribution value for individual predictions, enabling both global interpretation (ranking variables by overall importance) and local interpretation (understanding which factors may contribute to risk for specific patients). Together, these methods complement traditional regression by providing a transparent, clinically interpretable model that can identify patterns not readily captured by linear assumptions.Studies have demonstrated the utility of these methods in predicting sarcopenia and PSO, with algorithms providing both predictive accuracy and model transparency [12, 13].
In this study, we used data from 427 community-dwelling older adults to: (1) describe the prevalence and clinical characteristics of PSO; (2) identify risk factors for PSO using multivariable regression; (3) test whether these risk factors differ by sex and age; (4) validate findings using Bayesian hierarchical models; (5) examine the dose–response relationship between hemoglobin and PSO; (6) test whether white blood cell count mediates the BMI-PSO association; (7) assess robustness to unmeasured confounding using E-value analysis; and (8) enhance interpretability using XGBoost with SHAP.
Methods
Study design and participants
This cross-sectional study included consecutive community-dwelling older adults who underwent routine health examinations at the geriatric department of a tertiary hospital between January 2023 and February 2024. All eligible individuals during this period were invited to participate; no active selection or sampling was applied beyond the inclusion/exclusion criteria. Inclusion criteria were: age 60 years or older, ability to complete physical and functional assessments independently, and provision of written informed consent. Exclusion criteria were: severe cardiac, hepatic, or renal dysfunction; active malignancy; acute infection or inflammation; and cognitive impairment that prevented cooperation with assessments.
A total of 427 participants were included in the final analysis (152 men, 275 women; age range 60–98 years). The study was approved by the hospital ethics committee, and all participants provided written informed consent.
Data collection
Demographic characteristics and anthropometric measurements, including age, sex, height, weight, waist circumference, and calf circumference, were recorded for all participants. BMI was calculated as weight divided by height squared (kg/m2).
Handgrip strength was measured using an electronic dynamometer. Participants performed two trials with their dominant hand, with a 30-s rest interval between trials to minimize fatigue, and the maximum value was recorded. Low muscle strength was defined according to the Asian Working Group for Sarcopenia (AWGS) 2019 criteria as grip strength < 28 kg in men and < 18 kg in women [14].Two trials were chosen as a pragmatic balance between obtaining a reliable maximum value and minimizing participant burden, particularly given the older age of the study population. This protocol follows standard practice in large-scale epidemiological studies and is consistent with the AWGS 2019 recommended assessment procedure.
Gait speed was assessed using the 6-m walk test at usual pace. Slow gait speed was defined as < 0.8 m/s, also following AWGS 2019 criteria [14].
Fasting venous blood samples were collected from all participants after an overnight fast. Hemoglobin was measured by the cyanmethemoglobin method, white blood cell count by automated flow cytometry (Sysmex XN-9000, Kobe, Japan), and glucose, triglycerides, and HDL-C by enzymatic colorimetric methods using a Hitachi 7600 series autoanalyzer (Tokyo, Japan). All assays were performed in the hospital's clinical laboratory following standardized protocols. Quality control was maintained through daily calibration and participation in external proficiency testing programs. The coefficients of variation for all assays were within acceptable ranges (< 5% for glucose and lipids, < 3% for hemoglobin).
Definitions and grouping
Sarcopenia
Defined as the presence of both low handgrip strength and slow gait speed according to AWGS 2019 criteria [14]. Muscle mass data were not available, so this represents "possible sarcopenia" as defined in the AWGS framework. We acknowledge that this definition may overestimate sarcopenia prevalence compared to muscle-mass-based definitions. However, we justify its use for the following reasons. First, the AWGS 2019 guidelines explicitly recognize that possible sarcopenia is a valid and pragmatic first-step screening tool, particularly in community and primary care settings where advanced imaging (e.g., dual-energy X-ray absorptiometry or bioelectrical impedance analysis) is not routinely available. Second, handgrip strength and gait speed are established independent predictors of adverse outcomes, including disability, falls, and mortality, and are widely used in epidemiological studies of sarcopenia. Third, the primary objective of this study was to identify clinically actionable risk factors using a feasible, low-cost, and widely accessible approach that can be readily implemented across diverse healthcare settings. Finally, the Bayesian and machine learning validation provides additional robustness to our findings despite this limitation.
Obesity
Central obesity was defined as waist circumference ≥ 90 cm in men and ≥ 85 cm in women, based on Chinese criteria [15].
Study groups
Participants were classified into four mutually exclusive groups: normal (no sarcopenia, no obesity), obese-only (no sarcopenia, with obesity), sarcopenia-only (sarcopenia, no obesity), and possible sarcopenic obesity (both possible sarcopenia and obesity).
Other definitions
Anemia (hemoglobin < 120 g/L in men, < 110 g/L in women) [16], elevated white blood cell count (> 9.5 × 10⁹/L), hyperglycemia (fasting glucose ≥ 6.1 mmol/L) [17], hypertriglyceridemia (triglycerides ≥ 1.7 mmol/L) [17], low HDL-C (< 1.0 mmol/L in men, < 1.3 mmol/L in women) [17]. Metabolic syndrome was defined as central obesity plus at least one of hyperglycemia, hypertriglyceridemia, or low HDL-C.
Statistical analysis
Analyses were performed using R version 4.5.1. Continuous variables were reported as mean ± standard deviation, and categorical variables as counts and percentages. Differences among the four groups were compared using one-way ANOVA or chi-square tests.
Traditional regression analysis
Multivariable logistic regression was used to identify factors independently associated with PSO, adjusting for age, BMI, hemoglobin, white blood cell count, fasting glucose, and metabolic syndrome. Results were reported as odds ratios (OR) with 95% confidence intervals (CI). Pre-specified subgroup analyses included sex-stratified models (men vs. women) and age-stratified models using 75 years as the cutpoint (< 75 vs. ≥ 75 years). The 75-year cut-point was selected based on previous studies demonstrating accelerated functional decline and increased frailty prevalence after this age, as well as to ensure balanced subgroup sample sizes for meaningful statistical comparison. We also performed a sensitivity analysis treating age as a continuous variable with an interaction term (age × each predictor), which yielded consistent results, supporting the robustness of the age-stratified findings.
Bayesian analysis
Bayesian hierarchical models were fitted using the brms package with weakly informative priors [18]. Four chains were run with 2000 iterations each. Posterior distributions were summarized using means and 95% credible intervals (CrI).
Dose–response analysis
Restricted cubic splines with three knots were used to assess the potential non-linear relationship between hemoglobin and PSO.
Mediation analysis
The mediation package in R was used to test whether white blood cell count mediated the association between BMI and PSO, with 1000 bootstrap iterations.
E-value analysis
E-values were calculated to assess robustness to unmeasured confounding, following VanderWeele and Ding [19].
Machine learning analysis
An XGBoost model was developed to predict PSO and validate the regression findings. The dataset was randomly split into training (70%) and testing (30%) sets, stratified by PSO status. This 70/30 split is a common and well-established practice in machine learning for moderate-sized datasets, balancing sufficient data for model training with an adequate sample size for performance evaluation [10, 13]. Hyperparameters were optimized using five-fold cross-validation. SHAP was used to interpret model predictions and quantify variable importance [11].
Given the limited number of outcome events (n = 37), we took several measures to mitigate overfitting. First, the number of predictors in the multivariable model was restricted to six prespecified variables based on clinical relevance and previous literature. Second, we did not perform stepwise selection. Third, for subgroup analyses, the results are explicitly interpreted as exploratory due to the small sample sizes (men with PSO: n = 14; women with PSO: n = 23). Fourth, the Bayesian analysis used weakly informative priors to regularize estimates. Fifth, the machine learning component is presented as an exploratory validation, not a definitive predictive model.Multicollinearity was assessed using variance inflation factors (VIF). All VIF values were below 2.0, indicating no substantial multicollinearity among the predictors in the regression model.
Results
Baseline characteristics
Of the 427 participants, 204 (47.8%) were classified as normal, 78 (18.3%) as obese-only, 108 (25.3%) as sarcopenia-only, and 37 (8.7%) as having sarcopenic obesity. Mean age was 72.4 ± 7.1 years, and 64.4% were women.
Table 1 shows the baseline characteristics of the four groups. The PSO group was the oldest (76.0 ± 6.9 years) and had the most pronounced metabolic disturbances. Waist circumference and BMI were significantly higher in the obese-only and PSO groups (p < 0.001). Handgrip strength and gait speed were lowest in the sarcopenia-only and PSO groups (p < 0.001).
Table 1.
Baseline characteristics of study participants
| Characteristic | Normal (n = 204) | Obese-only (n = 78) | Sarcopenia-only (n = 108) | Possible Sarcopenic obesity (n = 37) | p-value |
|---|---|---|---|---|---|
| Age, years | 70.13 (5.55) | 71.15 (5.31) | 73.56 (7.26) | 75.97 (6.92) | < 0.001 |
| BMI, kg/m2 | 23.56 (3.51) | 25.90 (3.03) | 24.22 (3.67) | 25.97 (2.50) | 0.545 |
| Waist circumference, cm | 76.76 (10.69) | 94.58 (10.39) | 76.63 (10.10) | 91.78 (4.68) | < 0.001 |
| Calf circumference, cm | 32.18 (2.94) | 35.13 (2.65) | 31.30 (2.22) | 33.70 (2.58) | < 0.001 |
| Gait speed, m/s | 1.10 (0.21) | 1.10 (0.22) | 0.90 (0.26) | 0.96 (0.27) | < 0.001 |
| Handgrip strength, kg | 28.52 (9.98) | 34.48 (17.46) | 18.30 (5.38) | 17.99 (5.98) | < 0.001 |
| SARC-CalF score | 15.00 (0.00) | 15.00 (0.00) | 15.00 (0.00) | 15.00 (0.00) | — |
| Hemoglobin, g/L | 136.49 (14.54) | 139.36 (12.93) | 133.37 (13.99) | 131.49 (16.55) | 0.023 |
| White blood cell count, × 10⁹/L | 6.45 (1.74) | 6.90 (2.34) | 6.49 (1.77) | 6.88 (1.64) | 0.339 |
| Platelet count, × 10⁹/L | 253.98 (56.34) | 265.64 (166.23) | 247.20 (57.91) | 237.05 (78.22) | 0.525 |
| Alanine aminotransferase, U/L | 19.01 (9.51) | 20.91 (12.12) | 17.70 (12.21) | 16.35 (5.40) | 0.133 |
| Aspartate aminotransferase, U/L | 23.54 (11.99) | 21.19 (7.27) | 20.76 (7.40) | 19.41 (4.60) | 0.065 |
| Total bilirubin, μmol/L | 12.50 (4.80) | 11.11 (3.54) | 11.00 (4.15) | 11.80 (5.60) | 0.115 |
| Serum creatinine, μmol/L | 77.88 (17.85) | 85.73 (32.20) | 75.98 (16.28) | 104.67 (64.56) | < 0.001 |
| Blood urea nitrogen, mmol/L | 5.57 (1.61) | 6.17 (5.36) | 6.01 (1.82) | 6.86 (2.85) | 0.205 |
| Fasting glucose, mmol/L | 5.28 (1.77) | 5.55 (1.64) | 5.25 (1.10) | 5.96 (2.44) | 0.163 |
| Total cholesterol, mmol/L | 5.24 (1.22) | 5.64 (6.96) | 5.03 (1.21) | 4.63 (1.27) | 0.588 |
| Triglycerides, mmol/L | 1.55 (1.56) | 1.85 (1.00) | 1.29 (0.83) | 1.48 (0.51) | 0.069 |
| LDL-C, mmol/L | 3.27 (1.05) | 2.99 (0.94) | 3.18 (0.99) | 2.85 (1.14) | 0.106 |
| HDL-C, mmol/L | 1.44 (0.41) | 1.22 (0.31) | 1.45 (0.44) | 1.27 (0.28) | < 0.001 |
Data are presented as mean (SD). BMI Body mass index, LDL-C Low-density lipoprotein cholesterol, HDL-C High-density lipoprotein cholesterol. P-values were derived from one-way ANOVA for continuous variables
*P-values were derived from one-way ANOVA for continuous variables. Post-hoc pairwise comparisons were performed using Tukey's Honest Significant Difference test. Significant differences between groups are indicated in the text where relevant
Hemoglobin levels were lowest in the PSO group (131.5 ± 16.5 g/L, p = 0.023). The prevalence of hyperglycemia was highest in the PSO group (32.4%, p = 0.061), and metabolic syndrome was significantly more common in the obese-only (29.5%) and PSO (27.0%) groups compared to the normal (1.0%) and sarcopenia-only (0.9%) groups (p < 0.001).
Risk factors for sarcopenic obesity: traditional regression
In univariable analysis, age (OR = 1.11, 95% CI: 1.05–1.16, p < 0.001) and metabolic syndrome (OR = 5.19, 95% CI: 2.19–11.64, p < 0.001) were positively associated with PSO, while hemoglobin was inversely associated (OR = 0.98, 95% CI: 0.95–1.00, p = 0.044) (Table 2).
Table 2.
Multivariable logistic regression analysis for possible sarcopenic obesity
| Variable | Univariable | Multivariable | ||
|---|---|---|---|---|
| OR (95% CI) | p-value | OR (95% CI) | p-value | |
| Age (per year) | 1.11 (1.05–1.16) | < 0.001 | 1.12 (1.05–1.19) | < 0.001 |
| BMI (per kg/m2) | 1.01 (0.98–1.03) | 0.499 | 1.02 (0.99–1.05) | 0.162 |
| Hemoglobin (per g/L) | 0.98 (0.95–1.00) | 0.044 | 0.97 (0.94–0.99) | 0.020 |
| White blood cell count (per 10⁹/L) | 1.07 (0.90–1.26) | 0.412 | 1.15 (0.93–1.38) | 0.152 |
| Fasting glucose (per mmol/L) | 1.16 (0.98–1.36) | 0.063 | 1.18 (0.96–1.41) | 0.094 |
| Metabolic syndrome (yes vs. no) | 5.19 (2.19–11.64) | < 0.001 | 2.01 (0.73–5.22) | 0.160 |
CI Confidence interval, BMI Body mass index. Model adjusted for all variables shown
In multivariable analysis (Table 2, Fig. 1), age remained an independent risk factor (OR = 1.12, 95% CI: 1.05–1.19, p < 0.001) and hemoglobin remained an independent protective factor (OR = 0.97, 95% CI: 0.94–0.99, p = 0.02). Fasting glucose showed a borderline association (OR = 1.18, 95% CI: 0.96–1.41, p = 0.094). The model had moderate predictive ability, with an area under the ROC curve of 0.76 (Fig. 2).
Fig. 1.
Forest plot of risk factors for sarcopenic obesity. Multivariable logistic regression analysis adjusted for age, BMI, hemoglobin, white blood cell count, fasting glucose, and metabolic syndrome. Points represent odds ratios; horizontal lines represent 95% confidence intervals
Fig. 2.
ROC curve from traditional regression model. Area under the curve (AUC) = 0.76
Sex-stratified analysis
Sex-stratified analysis revealed striking differences (Supplementary Table 1, Supplementary Fig. 2). In men, age (OR = 1.15, 95% CI: 1.03–1.28) and BMI (OR = 1.54, 95% CI: 1.19–2.01) were independent risk factors. In women, age showed a borderline association (OR = 1.09, 95% CI: 1.00–1.19), while hemoglobin (OR = 0.96, 95% CI: 0.92–0.99) and white blood cell count (OR = 1.43, 95% CI: 1.04–1.98) were significant independent factors.
Age-stratified analysis
When stratified by age (Supplementary Table 2), different patterns emerged. In participants younger than 75 years, hemoglobin (OR = 0.96, 95% CI: 0.92–0.99), white blood cell count (OR = 1.23, 95% CI: 1.01–1.51), and fasting glucose (OR = 1.26, 95% CI: 1.02–1.56) were significant. In those aged 75 years and older, BMI showed a very strong association (OR = 1.91, 95% CI: 1.33–2.76), and age remained a risk factor (OR = 1.28, 95% CI: 1.05–1.56).
Bayesian validation
Bayesian analysis confirmed the traditional regression findings with narrow posterior credible intervals (Supplementary Table 3, Supplementary Fig. 3). The posterior mean for age was 1.12 (95% CrI: 1.05–1.19), for hemoglobin was 0.97 (95% CrI: 0.94–0.99), and for BMI was 1.02 (95% CrI: 0.99–1.05). Sex-stratified Bayesian models showed posterior probabilities > 0.99 for BMI in men and > 0.95 for hemoglobin and white blood cell count in women. Age-stratified models showed posterior probabilities > 0.99 for BMI in the ≥ 75 years group and > 0.95 for glucose and white blood cell count in the < 75 years group.
Interaction analyses
No statistically significant interactions were detected: age × metabolic syndrome (p = 0.916), BMI × glucose (p = 0.144), or hemoglobin × age (p = 0.261) (Supplementary Fig. 1). However, the BMI × glucose interaction showed a trend (p = 0.144), suggesting a possible synergistic effect.
Dose–response relationship
Restricted cubic spline analysis showed a linear relationship between hemoglobin and PSO (p for nonlinearity = 0.743), with no evidence of a threshold effect (Fig. 3). Each 1 g/L decrease in hemoglobin was associated with approximately a 3% increase in SO risk.
Fig. 3.
Dose–response relationship between hemoglobin and sarcopenic obesity. Restricted cubic spline analysis adjusted for age and BMI. Shaded area represents 95% confidence interval. p for nonlinearity = 0.743
Mediation analysis
White blood cell count did not significantly mediate the association between BMI and PSO (indirect effect p = 0.944, mediation proportion 1.05%) (Supplementary Table 4, Supplementary Fig. 4). This suggests that the effect of BMI on PSO operates through other pathways, such as direct mechanical effects, adipokine secretion, or insulin resistance.
Sensitivity analysis
E-value analysis showed that an unmeasured confounder would need to be associated with both age and PSO by an odds ratio of at least 1.47 to explain away the age effect, and at least 1.21 to explain away the hemoglobin effect (Supplementary Table 5). These values suggest the main findings are reasonably robust to unmeasured confounding.
Machine learning validation
The XGBoost model achieved good discrimination, with an AUC of 0.86 on the test set (Supplementary Fig. 5). This performance exceeds that of the traditional regression model (AUC = 0.76), suggesting that machine learning captures additional complexity in the data.The small number of positive cases (n = 37) means that the XGBoost results should be interpreted as exploratory; no claim of clinical prediction model readiness is made.
SHAP analysis revealed the relative importance of predictors (Supplementary Fig. 6). Age, BMI, and hemoglobin emerged as the three most important variables, consistent with the traditional regression and Bayesian findings. The SHAP summary plot (Supplementary Fig. 7) shows the direction of effects: older age and higher BMI increased predicted risk, while higher hemoglobin decreased predicted risk. Sex-specific patterns were also evident, with BMI showing stronger effects in men and hemoglobin showing stronger effects in women, corroborating the stratified regression results.
The 427 participants were split into training (n = 299, of which 26 had PSO) and testing (n = 128, of which 11 had PSO) sets. To address class imbalance, we applied the Synthetic Minority Over-sampling Technique (SMOTE) to the training data. The final XGBoost model was trained with the following hyperparameters: max_depth = 4, eta = 0.05, nrounds = 200, subsample = 0.8, colsample_bytree = 0.8, min_child_weight = 1. Performance on the test set was: AUC = 0.86 (95% CI: 0.78–0.94), sensitivity = 0.82, specificity = 0.75, PPV = 0.24, NPV = 0.98. To reduce optimistic bias, we used fivefold cross-validation within the training set for hyperparameter tuning rather than relying on a single test set for final evaluation.
Discussion
In this cross-sectional study of 427 community-dwelling older adults, we found that sarcopenic obesity affects nearly one in twelve individuals, and its risk factors differ importantly by sex and age. Our main findings are: (1) overall, age is a risk factor and hemoglobin is protective; (2) in men, BMI is the dominant risk factor; (3) in women, low hemoglobin and high white blood cell count are more important; (4) in people under 75, inflammation and glucose metabolism matter most; (5) in people 75 and older, BMI has a very strong effect; and (6) Bayesian and machine learning approaches produced consistent findings.
Prevalence in context
The 8.7% prevalence of PSO in our study is consistent with previous reports from Chinese community samples, which range from 6 to 12% [5, 6]. It is somewhat lower than figures from Western populations (10–15%) [2, 3], which may reflect differences in body composition, dietary patterns, or obesity definitions. A recent nationwide study using CHARLS data reported a prevalence of 6.40% among Chinese adults aged ≥ 45 years [20]. Our estimate of 8.7% in an older sample aligns with the expected age-related increase in PSO prevalence. The larger group—sarcopenia without obesity (25.3%)—is worth noting: while PSO carries the highest metabolic risk, sarcopenia alone is more common and also merits clinical attention.
The central role of age and hemoglobin
Age was the most consistent risk factor across all analyses, with each additional year increasing PSO risk by about 12%. This is not surprising: aging brings loss of muscle mass, redistribution of fat to central depots, and declines in metabolic function [21]. What is perhaps more interesting is the consistent protective effect of hemoglobin. Each 1 g/L increase in hemoglobin was associated with a 3% lower risk of PSO, and this relationship was linear across the entire range of values. This means that even mild anemia—not just clinically defined anemia—may increase PSO risk. Hemoglobin is easy to measure in routine clinical practice and could serve as a simple screening tool for nutritional vulnerability. The Bayesian analysis confirmed this finding with high posterior probability, and the dose–response analysis ruled out non-linear threshold effects.
Several mechanisms may explain the inverse association between hemoglobin and PSO observed in this study. First, low hemoglobin levels may reflect chronic systemic inflammation or nutritional deficits (e.g., iron, vitamin B12, or folate deficiency), both of which are independently linked to muscle wasting and metabolic dysfunction. Second, reduced oxygen-carrying capacity may impair mitochondrial oxidative phosphorylation in skeletal muscle, contributing to reduced physical performance and accelerating age-related muscle atrophy. Third, hemoglobin may serve as an integrative marker of overall health status, capturing the cumulative burden of multiple comorbidities and subclinical disease states rather than acting as a direct causal factor. While our cross-sectional design does not allow causal inference, the linear dose–response relationship and robustness to unmeasured confounding (E-value = 1.21) support the plausibility of a meaningful association that warrants further investigation in prospective studies.
Sex differences: a key finding
The most striking finding of this study is that the risk factor profile for PSO is fundamentally different in men and women. In men, BMI was a strong risk factor (OR = 1.54). In women, BMI had no significant effect; instead, low hemoglobin and high white blood cell count emerged as the key factors. Bayesian analysis produced consistent findings with posterior probabilities > 0.99 for BMI in men and > 0.95 for hemoglobin and white blood cell count in women.
The distribution of adipose tissue differs markedly between sexes. Men tend to accumulate visceral fat, which is more metabolically harmful and promotes systemic inflammation and insulin resistance, is associated with muscle catabolism [22]. Women, particularly after menopause, accumulate more subcutaneous fat, which may be less directly harmful to muscle [23]. Instead, women may be more vulnerable to nutritional deficits and low-grade inflammation. Estrogen has anti-inflammatory effects, and its decline after menopause may be associated with inflammatory pathways that correlate with muscle loss [24]. This aligns with research showing that the relationship between vitamin D status and sarcopenia also differs by sex [25].
These findings suggest that sex- and age-specific risk stratification may improve clinical decision-making. In men, elevated BMI—even in the absence of other risk factors—may warrant targeted weight management interventions. In women, normal BMI does not preclude PSO risk; routine assessment of hemoglobin and white blood cell count could identify otherwise unrecognized vulnerability. These data support a precision medicine approach to PSO prevention and management, wherein screening and intervention strategies are tailored to an individual's sex and age group.
Age differences: a second layer
Age stratification added another layer of complexity. In people under 75, the dominant factors were white blood cell count and glucose—both markers of metabolic and inflammatory stress. In people 75 and older, BMI emerged as a very strong risk factor (OR = 1.91). Bayesian analysis confirmed this with posterior probabilities > 0.99 for BMI in the oldest-old group.
This suggests that the pathways to PSO may change with age. In the "younger old," metabolic dysregulation and inflammation may contribute to the syndrome. In the "older old," the cumulative mechanical and metabolic burden of excess weight may become more important. This makes biological sense: as people age into their late 70 s and 80 s, the association between BMI and PSO appears to be particularly strong in older age groups, possibly reflecting the cumulative mechanical and metabolic burden of excess weight through joint stress, reduced mobility, and adiposity-related metabolic dysfunction.
What about inflammation as a mediator?
We conducted an exploratory mediation analysis to assess whether white blood cell count was associated with the BMI-PSO relationship, but found no evidence of mediation (proportion 1.05%). Bayesian mediation analysis confirmed this null finding with high precision. This negative finding is informative. It suggests that BMI does not cause PSO primarily through inflammatory pathways (at least as measured by white blood cell count). Other mechanisms—perhaps direct mechanical effects, adipokine secretion, or insulin resistance—may be more important. This aligns with research showing that the pro-inflammatory state associated with obesity is complex and may not be fully captured by a single marker [26].Given the cross-sectional design, these mediation findings should be interpreted as exploratory and hypothesis-generating rather than confirmatory.
Methodological contributions
The use of multiple analytic approaches strengthens confidence in our findings. Traditional regression provided interpretable effect sizes and confidence intervals. Bayesian hierarchical models quantified uncertainty and provided posterior probabilities for subgroup effects. Machine learning with SHAP validated the variable importance rankings and revealed potential non-linearities. The convergence of findings across these distinct methodological frameworks suggests that the observed sex and age differences are robust and not artifacts of any single analytic approach.
To our knowledge, this is the first study to comprehensively examine sex- and age-specific risk factors for sarcopenic obesity in a Chinese community sample using multiple analytic approaches—traditional regression, Bayesian hierarchical models, and machine learning with SHAP. We found that risk factor profiles differ fundamentally: BMI dominates in men and in those ≥ 75 years, while hemoglobin and white blood cell count are more important in women and in those < 75 years. The dose–response relationship between hemoglobin and PSO was linear, with no threshold effect.
Strengths and limitations
This study has several strengths. First, we used standardized, widely accepted definitions for sarcopenia (AWGS 2019) and central obesity. Second, we performed rigorous stratified analyses to uncover sex and age differences—a critical contribution given that previous studies have largely treated older adults as a homogeneous group. Third, we went beyond basic regression to examine dose–response relationships, mediation, and robustness to unmeasured confounding. Fourth, we validated our findings using Bayesian hierarchical models and machine learning with SHAP, providing both methodological rigor and clinical accessibility. Fifth, the convergence of findings across distinct analytic approaches—traditional regression, Bayesian analysis, and machine learning—substantially increases confidence in the observed sex- and age-specific patterns.
Several limitations should be acknowledged. First, the cross-sectional design means we cannot establish causality. We can identify associations, but we cannot determine whether low hemoglobin is associated with lower odds PSO or is simply a marker of poor overall health. However, the linear dose–response relationship and robustness to unmeasured confounding (E-value ≥ 1.47) support the plausibility of causal inference. Second, we lacked direct measures of muscle mass (by DXA or BIA), so our definition of sarcopenia relies on strength and physical performance only. This represents "possible sarcopenia" in the AWGS framework—a limitation that may correlated with misclassification of some individuals. Third, the study was conducted at a single center, which may limit generalizability. Nevertheless, our findings are consistent with previous reports from Chinese community samples, suggesting external validity. Fourth, we did not measure physical activity, medication use, hormone levels, or detailed inflammatory markers (such as CRP, IL-6)—all of which could influence SO risk. Fifth, some subgroup analyses had small sample sizes, particularly the PSO group itself (n = 37) and sex-specific subgroups (men with PSO: n = 14; women with SO: n = 23), which may have limited statistical power. However, Bayesian analysis partially mitigates this concern by borrowing strength across subgroups and providing posterior probabilities that confirm the robustness of key findings.
The small number of outcome events (n = 37) limited our ability to include more predictors in regression models and may have increased the risk of overfitting, particularly in subgroup analyses. Therefore, subgroup findings should be interpreted with caution and considered hypothesis-generating rather than confirmatory.
Clinical implications
These findings suggest that prevention and management strategies for sarcopenic obesity should be sex- and age-specific: weight control for men and the oldest-old, nutritional and inflammatory monitoring for women and the younger-old. Hemoglobin, a simple and inexpensive marker, could serve as a screening tool for nutritional vulnerability.
Conclusions
Possible sarcopenic obesity affects 8.7% of community-dwelling older adults in this Chinese sample. Age is a consistent risk factor, and hemoglobin is a consistent protective factor. However, these overall associations mask important heterogeneity. In men, BMI is the dominant risk factor. In women, low hemoglobin and high white blood cell count are more important. In people under 75, metabolic and inflammatory markers dominate. In people 75 and older, BMI has a very strong effect. Bayesian and machine learning approaches produced consistent findings. These results suggest that preventing and managing sarcopenic obesity will require sex- and age-specific strategies, not a one-size-fits-all approach.
Supplementary Information
Supplementary Material 2. Supplementary Table 1. Sex-Stratified Multivariable Logistic Regression Analysis for Possible Sarcopenic Obesity. CI: confidence interval; BMI: body mass index. Models adjusted for all variables shown. Supplementary Table 2. Age-Stratified Multivariable Logistic Regression Analysis for Possible Sarcopenic Obesity. CI: confidence interval; BMI: body mass index. Models adjusted for all variables shown. Supplementary Table 3. Bayesian Validation of Subgroup Findings. Bayesian models were fitted with weakly informative priors. Credible intervals are based on posterior distributions. Supplementary Table 4. Mediation Analysis: White Blood Cell Count as a Mediator of the BMI- Possible Sarcopenic Obesity Association. ACME: average causal mediation effect; ADE: average direct effect; CI: confidence interval. Analysis was performed with 1000 bootstrap iterations. The proportion mediated represents the percentage of the total effect of BMI on Possiblesarcopenic obesity that operates through white blood cell count. Supplementary Table 5. E-Value Sensitivity Analysis for Main Findings. E-value represents the minimum strength of association that an unmeasured confounder would need to have with both the exposure and outcome to fully explain away the observed exposure-outcome association, above and beyond the measured covariates. For the age effect, an unmeasured confounder would need to be associated with both age and sarcopenic obesity by an odds ratio of at least 1.47 to nullify the finding. For hemoglobin, the corresponding E-value was 1.21. Supplementary Figure 1. Interaction analysis for selected variable pairs. Points represent odds ratios for interaction terms; horizontal lines represent 95% confidence intervals. No statistically significant interactions were detected: age × metabolic syndrome (OR = 1.01, 95% CI: 0.92–1.10, p = 0.916), BMI × glucose (OR = 1.19, 95% CI: 0.94–1.50, p = 0.144), and hemoglobin × age (OR = 1.00, 95% CI: 0.99–1.01, p = 0.261). The BMI × glucose interaction showed a trend (p = 0.144), suggesting a possible synergistic effect that warrants further investigation in larger samples. Supplementary Figure 2. Sex-stratified forest plot. Results from multivariable logistic regression stratified by sex. BMI was a strong risk factor in men; low hemoglobin and high white blood cell count were significant in women. Supplementary Figure 3. Bayesian validation of subgroup findings. Point estimates and 95% credible intervals from Bayesian hierarchical models, confirming the robustness of stratified analyses. Supplementary Figure 4. Mediation analysis path diagram. White blood cell count as a mediator of the BMI-sarcopenic obesity association. Indirect effect p = 0.944, mediation proportion = 1.05%.Supplementary Figure 5. ROC curve from XGBoost model. Area under the curve (AUC) = 0.86. Supplementary Figure 6. SHAP feature importance. Age, BMI, and hemoglobin were the top three predictors, consistent with traditional regression findings. Supplementary Figure 7. SHAP summary plot. Red indicates higher feature values; blue indicates lower feature values. Older age and higher BMI increased predicted risk; higher hemoglobin decreased predicted risk.
Acknowledgements
We would like to thank the medical and nursing staff of Nanning Second People's Hospital for their dedicated patient care and support in data collection. We also express our gratitude to the Department of Respiratory and Critical Care Medicine at Guangxi Jiangbin Hospital for facilitating collaborative efforts in this research.
Data sharing
Deidentified participant data collected for this study, including individual participant data and a data dictionary defining each field, will be made available to others upon reasonable request to the corresponding author. The study protocol and statistical analysis plan are available from the corresponding author upon request. Data will be available beginning 3 months and ending 5 years after publication. Data will be shared with researchers who provide a methodologically sound proposal, after approval by the institutional ethics committee and with a signed data access agreement.
Authors’ contributions
C.C. and Y.D. conceptualized and designed the study. C.C. collected the data, performed the statistical analysis, and prepared the original draft and all figures. C.C. and Y.D. accessed and verified the underlying data. Y.D. supervised the study, acquired funding, and reviewed and edited the manuscript. All authors approved the final version and are accountable for all aspects of the work.
Funding
This work was supported by the Self-funded Scientific Research Project of the Health Commission of Guangxi Zhuang Autonomous Region, China (Grant No. Z-A20250219, Z-A20221163). The funding body had no role in the design of the study, data collection, analysis, interpretation of data, or in writing the manuscript.
Data availability
The datasets generated and/or analyzed during the current study are not publicly available due to institutional restrictions but are available from the corresponding author upon reasonable request.
Declarations
Ethics approval and consent to participate
This retrospective study was conducted in accordance with the Declaration of Helsinki. The study protocol has been submitted for review and is pending approval by the Ethics Committee of Nanning Second People's Hospital (Approval No. Y2021100).
Informed consent was obtained from all participants.
Competing interests
The authors declare no conflicts of interest.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Cruz-Jentoft AJ, Bahat G, Bauer J, Boirie Y, Bruyère O, Cederholm T, et al. Sarcopenia: revised European consensus on definition and diagnosis. Age Ageing. 2019;48(1):16–31. 10.1093/ageing/afy169. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Donini LM, Busetto L, Bischoff SC, Cederholm T, Ballesteros-Pomar MD, Batsis JA, et al. Definition and diagnostic criteria for sarcopenic obesity: ESPEN and EASO consensus statement. Clin Nutr (Edinburgh, Scotland). 2022;41(4):990–1000. 10.1016/j.clnu.2021.11.014. [DOI] [PubMed] [Google Scholar]
- 3.Batsis JA, Villareal DT. Sarcopenic obesity in older adults: aetiology, epidemiology and treatment strategies. Nat Rev Endocrinol. 2018;14(9):513–37. 10.1038/s41574-018-0062-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Gao Q, Mei F, Shang Y, Hu K, Chen F, Zhao L, et al. Global prevalence of sarcopenic obesity in older adults: a systematic review and meta-analysis. Clin Nutr (Edinburgh, Scotland). 2021;40(7):4633–41. 10.1016/j.clnu.2021.06.009. [DOI] [PubMed] [Google Scholar]
- 5.Hai S, Cao L, Wang H, Zhou J, Liu P, Yang Y, et al. Association between sarcopenia and nutritional status and physical activity among community-dwelling Chinese adults aged 60 years and older. Geriatr Gerontol Int. 2017;17(11):1959–66. 10.1111/ggi.13001. [DOI] [PubMed] [Google Scholar]
- 6.Sánchez-Sánchez JL, He L, Morales JS, de Souto Barreto P, Jiménez-Pavón D, Carbonell-Baeza A, et al. Association of physical behaviours with sarcopenia in older adults: a systematic review and meta-analysis of observational studies. Lancet Healthy Longevity. 2024;5(2):e108–19. 10.1016/S2666-7568(23)00241-6. [DOI] [PubMed] [Google Scholar]
- 7.Lee HJ, Choi JY, Hong D, Kim D, Min JY, Min KB. Sex differences in the association between sarcopenia and mild cognitive impairment in the older Korean population. BMC Geriatr. 2023;23(1):332. 10.1186/s12877-023-03911-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Penninx BW, Guralnik JM, Onder G, Ferrucci L, Wallace RB, Pahor M. Anemia and decline in physical performance among older persons. Am J Med. 2003;115(2):104–10. 10.1016/s0002-9343(03)00263-8. [DOI] [PubMed] [Google Scholar]
- 9.Beck ED, Yoneda T, James BD, Bennett DA, Hassenstab J, Katz MJ, et al. Personality predictors of dementia diagnosis and neuropathological burden: An individual participant data meta-analysis. Alzheimer’s Dementia. 2024;20(3):1497–514. 10.1002/alz.13523. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Chen X, Li L. Prediction of sarcopenia at different time intervals: an interpretable machine learning analysis of modifiable factors. BMC Geriatr. 2025;25(1):133. 10.1186/s12877-025-05792-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Lundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, et al. From local explanations to global understanding with explainable AI for trees. Nat Mach Intell. 2020;2(1):56–67. 10.1038/s42256-019-0138-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Lian R, Tang H, Chen Z, Chen X, Luo S, Jiang W, et al. Development and multi-center cross-setting validation of an explainable prediction model for sarcopenic obesity: a machine learning approach based on readily available clinical features. Aging Clin Exp Res. 2025;37(1):63. 10.1007/s40520-025-02975-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Xu M, Liu J, Hu S, Luan T, Duan Y, Wang A, et al. Construction of a prediction model for sarcopenic obesity based on machine learning. Front Public Health. 2025;13:1576338. 10.3389/fpubh.2025.1576338. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Chen LK, Woo J, Assantachai P, Auyeung TW, Chou MY, Iijima K, et al. Asian working group for sarcopenia: 2019 consensus update on sarcopenia diagnosis and treatment. J Am Med Directors Assoc. 2020;21(3):300-307.e2. 10.1016/j.jamda.2019.12.012. [DOI] [PubMed] [Google Scholar]
- 15.Zhou BF, Cooperative Meta-Analysis Group of the Working Group on Obesity in China. Predictive values of body mass index and waist circumference for risk factors of certain related diseases in Chinese adults–study on optimal cut-off points of body mass index and waist circumference in Chinese adults. Biomed Environ Sci : BES. 2002;15(1):83–96. [PubMed] [Google Scholar]
- 16.World Health Organization. Haemoglobin concentrations for the diagnosis of anaemia and assessment of severity. Geneva: WHO; 2011. [Google Scholar]
- 17.Chinese Elderly Type 2 Diabetes Prevention and Treatment of Clinical Guidelines Writing Group, Geriatric Endocrinology and Metabolism Branch of Chinese Geriatric Society, Geriatric Endocrinology and Metabolism Branch of Chinese Geriatric Health Care Society, Geriatric Professional Committee of Beijing Medical Award Foundation, & National Clinical Medical Research Center for Geriatric Diseases (PLA General Hospital) (2022). Zhonghua nei ke za zhi, 61(1), 12–50. 10.3760/cma.j.cn112138-20211027-00751 [DOI] [PubMed]
- 18.Bürkner PC. brms: An R package for Bayesian multilevel models using Stan. J Stat Softw. 2017;80(1):1–28. [Google Scholar]
- 19.VanderWeele TJ, Ding P. Sensitivity analysis in observational research: introducing the E-value. Ann Intern Med. 2017;167(4):268–74. 10.7326/M16-2607. [DOI] [PubMed] [Google Scholar]
- 20.Li X, Li Y, Shen M, Zhou Z, Liu S, Yin L, et al. Prevalence, influencing factors, and dementia outcome of sarcopenic obesity in China. Aging Clin Exp Res. 2026;38(1):49. 10.1007/s40520-025-03318-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Bilski J, Pierzchalski P, Szczepanik M, Bonior J, Zoladz JA. Multifactorial mechanism of sarcopenia and sarcopenic obesity. Role of physical exercise, microbiota and myokines. Cells. 2022;11(1):160. 10.3390/cells11010160. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Knuever J, Persa OD, Gerecht L, Schlaak M, Theurich S. Body composition and inflammation as emerging sex-specific prognostic factors in patients with advanced cutaneous melanoma undergoing immune checkpoint-inhibition therapy: a prospective pilot study. Oncol Res Treat. 2025;48(11):729–39. 10.1159/000546797. [DOI] [PubMed] [Google Scholar]
- 23.Link JC, Reue K. Genetic basis for sex differences in obesity and lipid metabolism. Annu Rev Nutr. 2017;37:225–45. 10.1146/annurev-nutr-071816-064827. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Zore T, Palafox M, Reue K. Sex differences in obesity, lipid metabolism, and inflammation-A role for the sex chromosomes? Mol Metab. 2018;15:35–44. 10.1016/j.molmet.2018.04.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Jia S, Zhao W, Hu F, Zhao Y, Ge M, Xia X, et al. Sex differences in the association of physical activity levels and vitamin D with obesity, sarcopenia, and sarcopenic obesity: a cross-sectional study. BMC Geriatr. 2022;22(1):898. 10.1186/s12877-022-03577-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Franceschi C, Garagnani P, Vitale G, Capri M, Salvioli S. Inflammaging and “garb-aging.” Trends Endocrinol Metab TEM. 2017;28(3):199–212. 10.1016/j.tem.2016.09.005. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplementary Material 2. Supplementary Table 1. Sex-Stratified Multivariable Logistic Regression Analysis for Possible Sarcopenic Obesity. CI: confidence interval; BMI: body mass index. Models adjusted for all variables shown. Supplementary Table 2. Age-Stratified Multivariable Logistic Regression Analysis for Possible Sarcopenic Obesity. CI: confidence interval; BMI: body mass index. Models adjusted for all variables shown. Supplementary Table 3. Bayesian Validation of Subgroup Findings. Bayesian models were fitted with weakly informative priors. Credible intervals are based on posterior distributions. Supplementary Table 4. Mediation Analysis: White Blood Cell Count as a Mediator of the BMI- Possible Sarcopenic Obesity Association. ACME: average causal mediation effect; ADE: average direct effect; CI: confidence interval. Analysis was performed with 1000 bootstrap iterations. The proportion mediated represents the percentage of the total effect of BMI on Possiblesarcopenic obesity that operates through white blood cell count. Supplementary Table 5. E-Value Sensitivity Analysis for Main Findings. E-value represents the minimum strength of association that an unmeasured confounder would need to have with both the exposure and outcome to fully explain away the observed exposure-outcome association, above and beyond the measured covariates. For the age effect, an unmeasured confounder would need to be associated with both age and sarcopenic obesity by an odds ratio of at least 1.47 to nullify the finding. For hemoglobin, the corresponding E-value was 1.21. Supplementary Figure 1. Interaction analysis for selected variable pairs. Points represent odds ratios for interaction terms; horizontal lines represent 95% confidence intervals. No statistically significant interactions were detected: age × metabolic syndrome (OR = 1.01, 95% CI: 0.92–1.10, p = 0.916), BMI × glucose (OR = 1.19, 95% CI: 0.94–1.50, p = 0.144), and hemoglobin × age (OR = 1.00, 95% CI: 0.99–1.01, p = 0.261). The BMI × glucose interaction showed a trend (p = 0.144), suggesting a possible synergistic effect that warrants further investigation in larger samples. Supplementary Figure 2. Sex-stratified forest plot. Results from multivariable logistic regression stratified by sex. BMI was a strong risk factor in men; low hemoglobin and high white blood cell count were significant in women. Supplementary Figure 3. Bayesian validation of subgroup findings. Point estimates and 95% credible intervals from Bayesian hierarchical models, confirming the robustness of stratified analyses. Supplementary Figure 4. Mediation analysis path diagram. White blood cell count as a mediator of the BMI-sarcopenic obesity association. Indirect effect p = 0.944, mediation proportion = 1.05%.Supplementary Figure 5. ROC curve from XGBoost model. Area under the curve (AUC) = 0.86. Supplementary Figure 6. SHAP feature importance. Age, BMI, and hemoglobin were the top three predictors, consistent with traditional regression findings. Supplementary Figure 7. SHAP summary plot. Red indicates higher feature values; blue indicates lower feature values. Older age and higher BMI increased predicted risk; higher hemoglobin decreased predicted risk.
Data Availability Statement
The datasets generated and/or analyzed during the current study are not publicly available due to institutional restrictions but are available from the corresponding author upon reasonable request.