The data for this cross-sectional study were sourced from the National Health and Nutrition Examination Survey (NHANES), a program of the National Center for Health Statistics (NCHS) designed to collect demographic, health, and nutrition data on the U.S. population. Ethical approval for all NHANES protocols was granted by the NCHS Research Ethics Review Board, and written informed consent was obtained from every participant. This secondary analysis adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cross-sectional studies (17). Comprehensive information on NHANES methodology and ethical considerations is publicly available through the CDC and NCHS websites.

Data were obtained from 39,156 participants across four NHANES cycles (2011–2018). Initial eligibility criteria (age ≥ 20 years, not pregnant) were met by 22,369 individuals. We then excluded participants with incomplete SP data (n = 11,625), missing values for the LC9 indicator (n = 2,935), missing eGDR data (n = 40), or missing data for any included covariates (n = 761). A total of 7,769 participants were included in the final analysis (Supplementary Figure S1).

To account for the complex, multi-stage probability sampling design of NHANES, all analyses incorporated appropriate sampling weights, primary sampling units (PSUs), and stratification variables. Multi-year weights for the combined 2011–2018 cycles were calculated by dividing the original 2-year Mobile Examination Center (MEC) weights by the number of cycles (4), following NHANES analytical guidelines. This approach ensures national representativeness and appropriate variance estimation for the combined dataset. All statistical analyses used the survey package in R to properly account for the complex sampling design.

The assessment of sarcopenia utilized the ratio of appendicular lean mass to body mass index (ALM/BMI). Appendicular lean mass (ALM), comprising the lean tissue mass of the limbs, was determined by dual-energy X-ray absorptiometry (DXA). BMI was obtained through professional measurement. Based on the Foundation for the National Institutes of Health (FNIH) Sarcopenia Project consensus definitions, sarcopenia was identified in individuals with an ALM/BMI below the sex-specific thresholds of 0.789 for males and 0.512 for females (18). It should be noted that DXA data were not available for all NHANES participants, which may affect the representativeness of the sample. However, our missing data analysis showed that excluded participants represented only 0.5% of the eligible population, minimizing potential selection bias.

Estimated Glucose Disposal Rate (eGDR) was calculated for each participant using data obtained from the NHANES database. The calculation employed the following formula:

eGDR (mg/kg/min) = 21.158 − (0.09 × WC) − (3.407 × Hypertension) − (0.551 × HbA1c)

In this formula, WC is waist circumference (cm), Hypertension status is coded (1 for yes, 0 for no), and HbA1c is measured in percent (%). For the purpose of this calculation, hypertension was defined as systolic blood pressure (SBP) ≥ 140 mmHg and/or diastolic blood pressure (DBP) ≥ 90 mmHg, or self-reported physician diagnosis of hypertension, or current intake of prescribed medication for high blood pressure (14). This formula was originally developed by Williams et al. and validated against hyperinsulinemic-euglycemic clamp measurements in patients with type 1 diabetes (12). While the formula has not been specifically validated in the current NHANES population, it has shown strong correlations with gold-standard insulin sensitivity measures across diverse populations and has been successfully applied in epidemiological studies of general populations.

We calculated an LC9 score for each participant, reflecting nine health aspects: healthy diet, physical activity, smoking cessation, healthy sleep, weight management, cholesterol control, blood glucose management, blood pressure management, and psychological health. As detailed in Supplementary Table S1, each metric was scored from 0 to 100 based on NHANES data. All nine LC9 metrics were standardized to a 0–100 scale to ensure comparability across different measurement units and distributions. The diet component (HEI-2015) was scored on a 0–100 scale and directly incorporated, while other metrics were transformed to match this scale as detailed in Supplementary Table S1. This approach ensures that each metric contributes equally to the overall LC9 score, with consistent weighting across all components. The final LC9 score was computed as the average of these nine metric scores. For the healthy diet metric, assessment was based on the Healthy Eating Index 2015 (HEI-2015) (19); its scoring details are available in Supplementary Table S2.

Covariates included in this study were age, sex, race/ethnicity, marital status, educational attainment, Poverty Income Ratio (PIR), and alcohol consumption. Detailed information regarding the definitions and categorization of these covariates is provided in Supplementary Table S3. Covariates were selected based on established literature regarding sarcopenia risk factors and potential confounders of the LC9-sarcopenia relationship. Age and sex are fundamental demographic factors affecting muscle mass. Race/ethnicity was included due to known differences in body composition and sarcopenia prevalence across racial groups. Socioeconomic factors (education, marital status, PIR) were included as they influence health behaviors and access to healthcare. Alcohol consumption was included due to its effects on muscle metabolism and nutritional status.

Data analysis was conducted using R (version 4.3.1) with the survey package for complex sampling design analysis. All analyses properly incorporated NHANES sampling weights, primary sampling units (PSUs), and strata to ensure national representativeness and correct variance estimation. Continuous variables are reported as mean ± SE, and categorical variables as weighted percentages, compared using weighted chi-square tests (20). We used multivariable logistic regression to assess the relationship between LC9 score and sarcopenia (SP), and between eGDR and SP, in separate analyses. We developed three models: Model 1 (crude), Model 2 (adjusted for age, sex, race/ethnicity), and Model 3 (additionally adjusted for education, marital status, PIR, alcohol consumption). Complete lists of variables included in each model are as follows: Model 1 included only LC9; Model 2 additionally included age, sex, and race/ethnicity; Model 3 further included education, marital status, PIR, alcohol consumption, and BMI. Identical covariate sets were used across logistic regression, RCS, and mediation analyses to ensure consistency. Restricted Cubic Splines (RCS) were employed to assess potential non-linear associations of LC9 and eGDR with SP. Subgroup analyses stratified by age, sex, race/ethnicity, education, marital status, PIR, and alcohol consumption examined the LC9-SP association, including tests for interaction. Mediation analysis assessed the potential mediating role of eGDR in the LC9-SP relationship. Analyses were performed using R, EmpowerStats, and Free Statistics. Statistical significance was set at p < 0.05 (two-sided). Additional sensitivity analyses included glucose-adjusted models using fasting glucose instead of eGDR to test robustness of findings. HOMA-IR analysis was not feasible due to unavailability of fasting insulin data in the current dataset.

Missing Data Analysis: Participants with missing data (n = 40, 0.5%) were compared to included participants (n = 7,769, 99.5%) using weighted t-tests and chi-square tests to assess potential selection bias. Collinearity Assessment: Variance inflation factors (VIF) were calculated for all predictors, with VIF > 5 indicating potential multicollinearity concerns. Multiple Comparison Correction: For subgroup interaction tests, p-values were adjusted using the Bonferroni method to control family-wise error rate. Restricted Cubic Splines: RCS models used 3 degrees of freedom with knots placed at the 25th, 50th, and 75th percentiles of LC9 distribution. Non-linearity was tested using Wald tests. Mediation Analysis: Bootstrap resampling (n = 500) estimated confidence intervals for mediation effects. Given the cross-sectional design, results represent statistical rather than causal mediation. Model Diagnostics: Model performance was assessed using C-statistics (AUC) and Hosmer-Lemeshow goodness-of-fit tests. Multiple Imputation Analysis: For sensitivity analysis, missing data were imputed using multiple imputation by chained equations (MICE) with 5 imputed datasets, using predictive mean matching for continuous variables.

Comprehensive Sensitivity Analyses: Multiple sensitivity analyses were conducted to test the robustness of findings: (1) alternative weighting strategies including unweighted analysis; (2) different covariate adjustment strategies; (3) alternative weighting approaches; (4) standardized effect size calculations; and (5) subgroup analyses by age and sex and (6) subgroup analyses by age and diabetes status. Standardized Effect Sizes: Cohen’s d for continuous variables and standardized odds ratios (per 1-SD change) were calculated to facilitate clinical interpretation. Model Specification: Complete variable lists and coding schemes for all models are provided in Supplementary materials to ensure reproducibility.

Based on 7,769 participants aged ≥ 20 years (representing ~89 million U.S. adults), the estimated weighted prevalence of sarcopenia was 8.3%. Compared to excluded participants (n = 40), included participants showed significant differences in sex distribution (80% vs. 51% male, p < 0.001) and BMI (35.6 vs. 28.8 kg/m², p < 0.001), but were similar in age and LC9 scores. When comparing baseline characteristics, individuals with sarcopenia were significantly different from those without sarcopenia concerning age, race/ethnicity, educational attainment, PIR, and alcohol consumption. Notably, the sarcopenia group exhibited significantly lower mean LC9 scores and eGDR levels than the non-sarcopenia group (Table 1). Standardized mean differences revealed large effect sizes for BMI (SMD = 0.97), eGDR (SMD = 0.74), and LC9 (SMD = 0.64), moderate effects for age (SMD = 0.41) and education (SMD = 0.43), and small effects for other demographic variables. Missing Data Analysis: Excluded participants (n = 40) differed significantly from included participants in sex distribution (80% vs. 51% male, p < 0.001), BMI (35.6 ± 12.8 vs. 28.8 ± 6.7 kg/m², p < 0.001), and LC9 scores (65.3 ± 13.6 vs. 72.4 ± 13.5, p = 0.001), but were similar in age (39.9 ± 11.8 vs. 39.2 ± 11.6 years, p = 0.695). Multiple Imputation Analysis: To assess the robustness of findings, sensitivity analysis using multiple imputation (m = 5) was performed for the small proportion of missing data. The multiple imputation results were consistent with complete case analysis (OR: 0.996, 95% CI: 0.985–1.007, p = 0.354), confirming that missing data did not substantially bias the results.

Variance inflation factors for all predictors were below 5.0 (LC9: 1.50, eGDR: 2.56, age: 1.23, BMI: 2.34), indicating no severe multicollinearity concerns. Correlation analysis revealed moderate correlation between LC9 and eGDR (r = 0.606), while BMI showed stronger correlations with both LC9 (r = −0.499) and eGDR (r = −0.709) (Supplementary Table S6). The fully adjusted model showed adequate discrimination (C-statistic: 0.505) and explained 17.2% of the variance (McFadden pseudo-R² = 0.172). Comprehensive Sensitivity Analyses: Multiple sensitivity analyses confirmed the robustness of the null finding in fully adjusted models. Sensitivity analyses confirmed the robustness of findings. Unweighted analysis showed similar results (OR = 0.995, p = 0.179). Glucose-adjusted models yielded consistent patterns (OR = 0.985, p = 0.034) (Supplementary Table S4). Minimal adjustment models showed significant associations (OR = 0.954, p < 0.001), confirming that full covariate adjustment explains the attenuation of the LC9-sarcopenia relationship. The association remained non-significant across all sensitivity analyses with full covariate adjustment.

Table 2 displays the results from multivariable logistic regression models. A strong inverse association was observed in crude (OR: 0.951, 95% CI: 0.945–0.956) and partially adjusted models (OR: 0.952, 95% CI: 0.946–0.958). However, in the fully adjusted model including BMI, the association became non-significant (OR: 0.994, 95% CI: 0.984–1.005, p = 0.282), suggesting that body composition largely explains the LC9-sarcopenia relationship (Supplementary Table S7). Categorizing LC9 score into tertiles, participants in the highest tertile (T3) demonstrated 83.4% lower odds of SP relative to the lowest tertile (T1) (OR: 0.166, 95% CI: 0.114–0.242). An analogous inverse association was found between eGDR and the odds of SP in all three tested models (all p < 0.001). All analyses incorporated NHANES complex survey weights, primary sampling units, and strata to ensure national representativeness. Odds ratios represent the association per 10-point increase in LC9 score or per 1-unit increase in eGDR. For enhanced interpretability, standardized effect sizes revealed that each 1-standard deviation increase in LC9 (13.48 points) was associated with a 7.4% decrease in sarcopenia odds in fully adjusted models (OR: 0.926, based on standardized coefficient analysis).

A non-linear inverse association between the adjusted LC9 score and sarcopenia (SP) odds was confirmed using Restricted Cubic Splines (p for overall association < 0.001; p for non-linearity = 0.001; Figure 1). Subsequent threshold analysis using segmented regression identified a turning point at an LC9 score of 73.33, selected based on the maximum log-likelihood method, with statistical significance confirmed by log-likelihood ratio test (p < 0.001). As shown in Table 3, the analysis suggests that while higher LC9 scores are generally protective, this protective effect diminishes for scores above 73.33. This threshold corresponds to the median LC9 score (50th percentile) in our study population, representing a moderate level of cardiovascular health. The clinical significance of this finding suggests that individuals with LC9 scores below 73.33 may experience the greatest relative benefits from cardiovascular health improvements. This has important implications for targeted interventions, as efforts to improve cardiovascular health among those with the poorest baseline health status (LC9 < 73.33) may yield disproportionately larger reductions in sarcopenia risk compared to interventions targeting those with already moderate-to-high cardiovascular health scores.

As shown in Figure 2, the significant inverse association between LC9 score and sarcopenia (SP) was consistent across subgroups defined by age, sex, race/ethnicity, marital status, educational level, PIR, and alcohol consumption. The analysis revealed a statistically significant interaction between LC9 score and sex (p = 0.0005 after Bonferroni correction). Quantitative interpretation of the significant sex interaction revealed that among females, each 10-point increase in LC9 was associated with a 1.7% decrease in sarcopenia odds (OR: 0.983, 95% CI: 0.968–0.999, p = 0.039), while no significant association was observed in males (OR: 1.006, 95% CI: 0.993–1.021, p = 0.369) (Supplementary Table S5). This indicates that cardiovascular health metrics may be more predictive of sarcopenia risk in women than in men. This sex-specific effect remained significant after multiple comparison adjustment.

Mediation analysis was conducted to explore the role of eGDR in the LC9-sarcopenia (SP) pathway (Figure 3). After controlling for covariates, a significant positive association was found between LC9 score and eGDR (β = 1.085, 95% CI: 1.030, 1.140; Table 4). Importantly, the results demonstrated that eGDR partially mediated the relationship between LC9 score and SP. Both the indirect effect via eGDR (estimate = −0.02557, p < 0.001) and the direct effect of LC9 (estimate = −0.02941, p < 0.001) were statistically significant in the fully adjusted model. The estimated proportion mediated through eGDR was 48.5% (95% CI: −429.4 to 775.8%, p = 0.824). However, given the non-significant total effect in fully adjusted models and the cross-sectional design, this should be interpreted as statistical rather than causal mediation. The large confidence intervals reflect the uncertainty in mediation estimates when the total effect approaches null.