All analyses were based on de-identified data downloaded from the Dryad public repository¹, an open-access archive that permits secondary use of deposited datasets.

The exposure variable was the CHG index, calculated as (17, 18): CHG = ln TC ( mg / dL ) × FBG ( mg / dL ) 2 × HDL − C ( mg / dL )

This study was designed as a retrospective cross-sectional analysis of health check-up data. The study endpoint was prevalent MASLD detected at baseline health check-ups. In the Chinese cohort, hepatic steatosis was assessed according to the recommendations of the Chinese Society of Hepatology (19), and in the Japanese cohort according to the Japanese Society of Gastroenterology (20). Both guidelines rely on abdominal ultrasonography and apply the same operational rule: hepatic steatosis was diagnosed when at least two of the following three findings were present: diffuse hepatic hyperechogenicity, increased liver–kidney echo contrast, and posterior beam attenuation with poor visualization of deep intrahepatic vessels.

The MASLD was then defined in a guideline-consistent operational manner as ultrasound-detected hepatic steatosis accompanied by at least one cardiometabolic risk condition assessed at baseline, including overweight (BMI ≥ 23.0 kg/m²), dysglycaemia/diabetes (fasting blood glucose ≥5.6 mmol/L or self-reported diabetes when available), lipid abnormality (triglycerides ≥1.7 mmol/L or low HDL-C < 1.0 mmol/L in men or <1.3 mmol/L in women), or elevated blood pressure (systolic blood pressure ≥130 mmHg or diastolic blood pressure ≥85 mmHg). The same operational definition was applied consistently in both the Chinese and Japanese cohorts.

Participants were classified as MASLD cases if hepatic steatosis was present on ultrasonography and at least one cardiometabolic risk condition was met according to the above criteria; otherwise, they were classified as non-MASLD. Baseline characteristics were summarized using distribution-appropriate statistics: variables following an approximately normal pattern are expressed as means with standard deviations, whereas non-normal measures are described by medians and interquartile ranges. For qualitative characteristics, results are shown as numbers and percentages. Between-group differences were evaluated with methods matched to data type and shape, applying chi-square tests to categorical outcomes, two-sample t-tests to normally distributed continuous measures, and Mann–Whitney U tests to continuous measures with skewed distributions.

To estimate the association between CHG and MASLD, CHG was analyzed in multivariable logistic regression as both a continuous exposure and a quartile-based categorical exposure. Three hierarchical models were constructed: an unadjusted model including CHG only; a partially adjusted model controlling for age, sex, and BMI; and a fully adjusted model further accounting for major metabolic covariates and liver and kidney function indices. Effect estimates are shown as odds ratios with 95% confidence intervals. Trends across quartiles were tested by assigning each quartile its median CHG value and modeling this variable continuously.

Potential non-linear dose–response relationships were examined using restricted cubic spline (RCS) functions within the fully adjusted framework. Four knots were placed at the 5th, 35th, 65th, and 95th percentiles of the CHG distribution. Non-linearity was assessed using likelihood ratio tests comparing models with and without spline terms. Subgroup analyses stratified by age and sex (and additionally by BMI, SBP, and DBP) were performed using fully adjusted logistic regression models, and effect modification was evaluated by adding CHG × subgroup interaction terms and reporting p-values for interaction. Discriminative performance of CHG was compared with the TyG index and traditional markers (TC, FBG, and HDL-C) using receiver operating characteristic (ROC) analysis, with the area under the curve (AUC) used for quantification. For incremental discrimination analyses, the baseline model was defined as the fully adjusted multivariable logistic regression model including all covariates adjusted for in the main multivariable analysis, but excluding CHG, and the “baseline + CHG” model was defined as the same model with CHG added. Predicted probabilities from these models were used to generate ROC curves and calculate AUC. The incremental discriminative contribution of CHG beyond baseline risk models was evaluated using net reclassification improvement (NRI) and integrated discrimination improvement (IDI). Subgroup analyses stratified by age and sex [and additionally by BMI, systolic blood pressure (SBP), and diastolic blood pressure (DBP)] were performed using fully adjusted logistic regression models; effect modification was evaluated by adding CHG × subgroup interaction terms and reporting p-values for interaction. All statistical analyses were conducted using R software (version 4.2.0), with the rms package for restricted cubic spline modeling, pROC for ROC analyses, and PredictABEL for NRI and IDI calculations; The FreeStatistics software (version 2.4.0) was used as an auxiliary platform. Two-sided p-values <0.05 were considered statistically significant.

Among Chinese participants, 16,173 individuals met the inclusion criteria, and 2,032 were identified with MASLD (Table 1). Relative to those without MASLD, affected participants tended to be older, were more frequently men, and displayed greater adiposity and higher blood pressure, including both systolic and diastolic measures (all p < 0.001). The MASLD group also had a less favorable biochemical profile, with elevated TB, ALT, AST, GGT, LDL-C, TG, and CHG, alongside reduced Direct Bilirubin (DBIL) and HDL-C (all p < 0.001). By contrast, Blood Urea Nitrogen (BUN) showed no meaningful between-group differences (p > 0.05).

In the Japanese dataset, 11,981 participants were evaluated, of whom 945 had MASLD (Table 2). The same clinical pattern was observed. Individuals with MASLD were generally older, included a higher share of males, and had higher BMI, waist circumference, and both systolic and diastolic blood pressure compared with non-MASLD participants (all p < 0.001). Laboratory indices likewise indicated metabolic impairment, as MASLD cases exhibited higher GGT, ALT, AST, FBG, HbA1c, total cholesterol, triglycerides, LDL-C, and CHG, while HDL-C was lower (all p < 0.001).

Overall, both cohorts consistently indicate that MASLD is associated with poorer anthropometric and metabolic characteristics, and that CHG levels are higher in participants with MASLD in each population.

Across the Chinese and Japanese samples, increasing CHG levels tracked with progressively more prevalent MASLD cases (Table 3). Treating CHG as a continuous exposure, each 1-unit rise in CHG corresponded to a markedly higher MASLD risk after multivariable adjustment, with ORs of 1.82 (95% CI 1.67 to 1.98) in China and 1.60 (95% CI 1.42 to 1.80) in Japan. Categorizing CHG into quartiles yielded the same pattern of stepwise escalation. In the fully controlled Chinese model, relative to Q1, the odds increased to 1.47 (1.03 to 2.10) in Q2, 2.12 (1.51 to 2.97) in Q3, and 4.56 (3.29 to 6.33) in Q4, showing a strong dose gradient (P-trend <0.001). The Japanese cohort showed parallel increases versus Q1, with ORs of 2.21 (1.18 to 4.14), 3.95 (2.18 to 7.16), and 5.51 (3.03 to 10.03) for Q2 through Q4, respectively, again supporting a robust monotonic trend (P-trend <0.001).

Using the fully adjusted models, restricted cubic spline analyses were conducted to characterize the dose–response relationship between CHG and the odds of prevalent MASLD (Figure 1). In the Chinese cohort (Figure 1A), CHG was significantly associated with MASLD across the observed range (P for overall <0.001), with clear evidence of non-linearity (P for non-linearity <0.001). With CHG = 5.013 set as the reference (OR = 1.0), the odds of MASLD increased modestly at lower CHG values but rose more steeply once CHG exceeded approximately 5.0, with a progressively stronger increase at higher CHG levels. In the Japanese cohort (Figure 1B), a significant positive association was also observed (P for overall <0.001), again with evidence of non-linearity (P for non-linearity = 0.001). Using CHG = 5.035 as the reference point, the odds of MASLD increased sharply around the reference range and continued to rise thereafter, albeit with a more gradual slope at higher CHG levels. Overall, these spline curves support a monotonic increase in the odds of prevalent MASLD with increasing CHG in both populations.

ROC analysis compared the discriminative ability of CHG with TyG, HDL-C, FBG, and TC for MASLD (Figure 2). In the Chinese cohort (Figure 2A), CHG achieved the highest AUC of 0.804 (95% CI 0.794–0.814), followed by TyG 0.753 (0.742–0.764), HDL-C 0.681 (0.669–0.693), FBG 0.657 (0.644–0.669), and TC 0.590 (0.578–0.604). In the Japanese cohort (Figure 2B), CHG also showed the highest AUC of 0.832 (0.820–0.844), marginally exceeding TyG 0.826 (0.813–0.839), with lower AUCs for HDL-C 0.777 (0.762–0.791), FBG 0.714 (0.697–0.730), and TC 0.629 (0.611–0.647). These findings suggest that CHG provides stronger discrimination for MASLD than single traditional lipid or glucose indices in both populations.

To examine whether CHG improves discrimination beyond conventional risk factors, ROC curves were compared between the baseline model and the baseline + CHG model in both cohorts (Figure 3). In the Chinese cohort (Figure 3A), the AUC increased from 0.846 (95% CI 0.834–0.857) for the baseline model to 0.862 (95% CI 0.852–0.873) after inclusion of CHG (p < 0.001). In the Japanese cohort (Figure 3B), the AUC rose from 0.923 (95% CI 0.916–0.931) to 0.927 (95% CI 0.920–0.934) with CHG added to the baseline model (p < 0.001). Although the absolute gains in AUC were numerically small, the consistent improvement across both cohorts indicates that CHG provides additional discriminative information beyond traditional risk factors for MASLD.

To evaluate the incremental value of CHG beyond the baseline model, NRI and IDI were calculated (Table 4). In the Chinese cohort, adding CHG to conventional risk factors resulted in an NRI of 0.5319 (95% CI: 0.4679–0.5959) and an IDI of 0.0368 (95% CI: 0.0299–0.0437). In the Japanese cohort, the corresponding estimates were 0.3421 for NRI (95% CI: 0.2761–0.4081) and 0.0063 for IDI (95% CI: 0.0025–0.0100). Collectively, these findings indicate that inclusion of CHG significantly improves MASLD risk reclassification and overall discriminative performance in both populations.

Subgroup analyses stratified by age and sex (and additionally by BMI, SBP, and DBP) showed that the positive association between CHG and prevalent MASLD was directionally consistent across all strata in both cohorts (Figure 4). In the Chinese cohort, the association was similar across age, sex, BMI, and DBP strata (all P for interaction >0.05), whereas a significant interaction was observed for SBP (P for interaction <0.001), with a stronger association in participants with SBP < 140 mmHg. In the Japanese cohort, evidence of effect modification by sex was observed (P for interaction = 0.002), with a stronger association in females; interactions for BMI, SBP, and DBP were not statistically significant (all P for interaction >0.05), and age showed a borderline interaction (P for interaction = 0.052).