This work reanalyzed public, fully de-identified data. The cross-sectional component used a Dryad-hosted dataset from a single-center retrospective study by Yan et al. (DOI: https://doi.org/10.5061/dryad.7d7wm3809) (19). That parent study received approval from the Institutional Review Board of Tongji Medical College, Huazhong University of Science and Technology (IRB: S155) (19), and reported that informed consent had been obtained. The longitudinal component relied on an open-access follow-up cohort published by Sun et al. (20); the source study was approved by the Ethics Committee of Wenzhou People’s Hospital and likewise documented informed consent. Because the present analyses involved only anonymized records and required no new enrollment or data acquisition, re-consent was not required. Data use conditions were observed, and the study adhered to the Declaration of Helsinki.

The study proceeded in two phases. Phase 1 assessed the association of SPISE with baseline MASLD in a cross-sectional sample. Phase 2 used follow-up data to verify the SPISE–incident MASLD relationship and to construct risk prediction models.

For the cross-sectional stage, eligible participants were identified from the Dryad dataset of Yan F et al. Participants with missing key variables required for SPISE calculation or outcome definition were excluded, yielding 1,592 individuals for analysis (19).

For longitudinal validation, the follow-up cohort reported by Sun DQ et al. was used. From a health-check population, 16,173 non-obese individuals free of MASLD at enrollment were included and followed longitudinally (20). Participants were excluded at baseline for any of the following (20): missing/incomplete clinical data with unsuccessful follow-up; current use of oral antihypertensive, lipid-lowering, or glucose-lowering agents; heavy alcohol intake (men ≥140 g/day; women ≥70 g/day); pre-existing MASLD, autoimmune hepatitis, viral hepatitis, or other chronic liver disease of known etiology; LDL-C > 3.12 mmol/L; or BMI ≥ 25 kg/m². Based on the public dataset, after calculation of the exposure index and harmonization of covariates, participants with missing key variables were additionally excluded, resulting in 16,172 individuals included in longitudinal analyses. The endpoint was the first diagnosis of MASLD during follow-up.

The primary exposure was the SPISE, calculated using a prespecified formula based on BMI, TG, and HDL-C. SPISE was analyzed as a continuous variable and in tertiles (T1–T3). For comparative assessment, the METS-IR (21), TG/HDL-C ratio (22), and the TyG index (22) were computed according to published formulas with harmonized units.

The outcome was MASLD. Hepatic steatosis in both datasets was assessed by abdominal ultrasonography and defined by the presence of ≥2 of the following (23): diffusely increased hepatic echogenicity, increased hepatorenal contrast, and posterior beam attenuation with poor visualization of deep intrahepatic vessels. MASLD was operationalized as ultrasound-confirmed steatosis plus ≥1 cardiometabolic risk factor assessed at baseline. Cardiometabolic risk was considered present if at least one of the following criteria was met (24): BMI ≥ 23.0 kg/m²; abnormal glycemia/diabetes (fasting plasma glucose ≥5.6 mmol/L or self-reported diabetes); dyslipidemia (TG ≥ 1.7 mmol/L or reduced HDL-C: men <1.0 mmol/L, women <1.3 mmol/L); or elevated blood pressure (SBP ≥ 130 mmHg or DBP ≥ 85 mmHg). Prevalent MASLD was defined at baseline in cross-sectional analyses; incident MASLD was defined as first meeting the criteria during follow-up among those free of MASLD at baseline.

Prespecified covariates included age, sex, blood pressure, liver-related biomarkers (ALT, AST, GGT, ALP, total bilirubin [TB], direct bilirubin [Dbil], albumin [ALB], globulin [GLB]), kidney-related biomarkers (creatinine [Cr], blood urea nitrogen [BUN], uric acid [UA]), and other routinely measured laboratory variables available in the datasets (25–27).

All statistical procedures were performed in R (version 4.5.1), applying two-tailed tests with a significance threshold of p < 0.05. Depending on distribution assessed by the Shapiro–Wilk test, continuous variables were reported as mean ± SD or median (IQR) and compared using Student’s t test or the Mann–Whitney U test. Categorical variables were summarized as n (%) and evaluated using the χ² test or Fisher’s exact test.

For the cross-sectional analysis, adjusted logistic regression models were fitted to quantify associations between SPISE and prevalent MASLD, reporting ORs with 95% CIs for SPISE modeled continuously and as tertiles. Dose–response trends were evaluated by entering SPISE tertiles as an ordered variable. For longitudinal analyses, Kaplan–Meier estimates were generated by tertile and compared using log-rank tests. Associations with incident MASLD were then quantified using Cox proportional hazards regression, reporting HRs with 95% CIs. A prespecified sequential adjustment strategy was applied (unadjusted; age- and sex-adjusted; fully adjusted for available clinical and laboratory covariates). The proportional hazards assumption was examined using Schoenfeld residuals. Nonlinear exposure–response patterns were explored by fitting restricted cubic spline terms in the fully adjusted Cox model, with statistical tests reported for both the global association and the nonlinear component.

Incremental predictive value was examined by entering SPISE, TyG, METS-IR, and TG/HDL-C one at a time into the fully adjusted reference model; changes in prediction were quantified using NRI and IDI with 95% CIs.

Model development used the longitudinal cohort. Candidate predictors were prespecified from the fully adjusted model, and variable selection was performed using LASSO with cross-validated λ (28). Variables retained after selection were refitted in a multivariable Cox model to generate the final risk equation, which was then translated into a nomogram and a web-based dynamic tool for estimating 12- and 24-month incident MASLD risk. Discrimination was quantified using time-dependent ROC curves and AUC. At 24 months, decision curve analysis was used to examine net benefit across a range of risk thresholds, benchmarking the model against the “treat-all” and “treat-none” approaches.

As a sensitivity analysis, participants who developed MASLD within the first 12 months of follow-up were excluded, and the Cox regression and restricted cubic spline analyses were repeated to reduce potential reverse causation and possible misclassification related to occult baseline steatosis. In addition, to further assess the robustness and internal validity of the final prediction model, internal cross-validation was performed for the 12- and 24-month models, and model performance was evaluated using the AUC and the Brier score.

MASLD was identified in 744 of 1,592 participants (46.7%). Compared with non-MASLD individuals (n = 848), the MASLD subgroup showed male predominance and a uniformly less favorable cardiometabolic profile, including higher BMI and blood pressure (SBP/DBP), higher fasting plasma glucose and TG, and lower HDL-C (all p < 0.001). Hepatic transaminase activities (ALT and AST) were also increased (all p < 0.001). Consistently, SPISE was lower in the MASLD group (5.31 ± 1.13 vs. 6.89 ± 1.42, p < 0.001) (Table 1).

In the unadjusted model, SPISE was inversely associated with prevalent MASLD (continuous SPISE: OR = 0.37, 95% CI 0.33–0.41). This association remained unchanged after adjustment for age and sex (OR = 0.37, 95% CI 0.33–0.41). In the fully adjusted model, SPISE remained independently associated with lower odds of MASLD (OR = 0.43, 95% CI 0.38–0.48) (Table 2).

When analyzed by SPISE tertiles, and using the lowest tertile (T1) as the reference, the fully adjusted ORs were 0.30 (95% CI 0.22–0.40) for T2 and 0.12 (95% CI 0.08–0.17) for T3 (p for trend < 0.001), indicating a clear dose–response gradient between higher SPISE and lower odds of prevalent MASLD (Table 2).

The longitudinal cohort included 16,172 participants free of MASLD at baseline, among whom 2,000 incident MASLD cases occurred during follow-up. Compared with participants who did not develop MASLD, those with incident MASLD were older and more often male. Baseline BMI, SBP, and DBP were also higher in the incident MASLD group (all p < 0.001) (Table 3).

Enzymatic and metabolic profiles differed by outcome. Participants who later developed MASLD had higher baseline GGT and ALT (both p < 0.001), along with higher TG and lower HDL-C (both p < 0.001). Consistent with these patterns, SPISE was reduced, whereas TyG, METS-IR, and TG/HDL-C were elevated in the incident MASLD group (all p < 0.001) (Table 3).

A significant separation of Kaplan–Meier curves was observed across SPISE tertiles (log-rank p < 0.0001). Baseline group sizes were comparable (T1 = 5,391; T2 = 5,390; T3 = 5,391). Over time, the steepest decline in MASLD-free survival probability was observed in T1, whereas T3 maintained the highest event-free survival (Figure 1).

Cox proportional hazards analyses indicated that higher SPISE was associated with lower incident MASLD risk. For SPISE modeled continuously, the unadjusted HR was 0.39 (95% CI 0.38–0.40), and the estimate remained essentially unchanged after adjustment for age and sex (HR = 0.39, 95% CI 0.38–0.41). With additional control for liver and renal biomarkers, other laboratory indices, and blood pressure, SPISE retained an independent inverse association with incident MASLD (HR = 0.49, 95% CI 0.46–0.52) (Table 4).

Tertile analyses showed a clear gradient. Using T1 as the reference, the fully adjusted HRs were 0.23 (95% CI 0.19–0.27) for T2 and 0.04 (95% CI 0.02–0.07), with a significant trend across categories (p for trend < 0.001) (Table 4).

Restricted cubic spline analyses indicated a significant nonlinear association between SPISE and incident MASLD risk (overall p < 0.001; p for non-linear < 0.001). Using SPISE = 7.93 as the reference (HR = 1.0), risk increased more steeply at SPISE values below the reference point, whereas risk continued to decrease above the reference range with a progressively attenuated slope (Figure 2).

When added individually to the fully adjusted base model, SPISE yielded the largest improvements in risk reclassification and discrimination (NRI = 0.363, 95% CI 0.314–0.399; IDI = 0.093, 95% CI 0.077–0.109; both p < 0.001). By comparison, adding TyG resulted in NRI = 0.179 and IDI = 0.032 (both p < 0.001); adding METS-IR resulted in NRI = 0.214 and IDI = 0.089 (both p < 0.001). Adding TG/HDL-C did not significantly improve reclassification (NRI = 0.025, p = 0.391), although a small improvement in discrimination was observed (IDI = 0.010, p = 0.016) (Table 5).

Using LASSO applied to candidate variables from the fully adjusted model together with SPISE, seven variables with non-zero coefficients were identified at the cross-validated optimal λ (Figure 3). Subsequent Cox regression analyses retained six independent predictors in the multivariable model (Table 6): SPISE, GGT, ALT, total bilirubin (TB), direct bilirubin (Dbil), and DBP. SPISE was inversely associated with incident MASLD (HR = 0.50, 95% CI 0.47–0.53). ALT and DBP were positively associated with risk, whereas Dbil was inversely associated. Although BUN was associated with the outcome in univariable analysis, the association did not remain independent in multivariable models and was not included in the final prediction model (Table 6).

A nomogram was constructed using the six predictors in the final multivariable Cox model (SPISE, GGT, ALT, TB, Dbil, and DBP) (Supplementary Figure S1A), and an online dynamic prediction tool was developed to estimate individualized 12- and 24-month risks of incident MASLD (Supplementary Figure S2). Time-dependent ROC analysis demonstrated good discrimination for mid-term MASLD risk (AUC 0.859 at 12 months and 0.886 at 24 months) (Supplementary Figure S1B). At 24 months, decision curve analysis showed that the model achieved greater net benefit than both “treat-all” and “treat-none” over an approximate threshold probability range of 0.06–0.82. For threshold probabilities <0.06, the “treat-all” strategy yielded higher net benefit; for threshold probabilities >0.82, model net benefit approached zero and was not superior to the baseline strategies (Supplementary Figure S1C).

As a sensitivity analysis, participants who developed MASLD within the first 12 months of follow-up were excluded and the main analyses were repeated. The inverse association between SPISE and incident MASLD remained materially unchanged. In the fully adjusted Cox model, continuous SPISE remained significantly associated with a lower risk of incident MASLD (HR = 0.51, 95% CI 0.47–0.54, p < 0.001). In tertile analyses, compared with T1, the fully adjusted HRs were 0.26 (95% CI 0.21–0.32) for T2 and 0.05 (95% CI 0.03–0.08) for T3, with p for trend < 0.001 (Supplementary Table S1). Restricted cubic spline analysis also showed a persistent nonlinear association between SPISE and incident MASLD risk (p for overall < 0.001; p for non-linearity < 0.001), supporting the robustness of the primary findings (Supplementary Figure S3).

To further evaluate model robustness, internal cross-validation was performed for the 12- and 24-month prediction models. For the 12-month model, the mean cross-validated AUC was 0.852 and the mean Brier score was 0.038 (Supplementary Table S2 and Figure S4). For the 24-month model, the mean cross-validated AUC was 0.876 and the mean Brier score was 0.094 (Supplementary Table S2 and Figure S5). These values were close to the apparent model performance estimates, indicating stable discrimination and predictive accuracy with minimal evidence of overfitting.