Comparison of four insulin resistance surrogate indices for incident MASLD in Chinese and prevalent MASLD in Japanese nonobese adults

Insulin resistance (IR) is central to metabolic dysfunction-associated steatotic liver disease (MASLD), but head-to-head comparisons of IR surrogate indices in nonobese populations are scarce. We compared the triglyceride-glucose index (TyG), TyG-body mass index (TyG-BMI), metabolic score for insulin resistance (METS-IR) and single-point insulin sensitivity estimator (SPISE) in relation to incident MASLD in a longitudinal Chinese cohort and prevalent MASLD in a cross-sectional Japanese cohort of nonobese adults. We analyzed two Dryad health examination datasets with different designs. The Chinese cohort ( n = 16,172) was longitudinal: participants free of hepatic steatosis at baseline were followed for a median of 35.8 months (interquartile range [IQR] 23.4–47.3), the outcome was incident MASLD, and entry was restricted to low-density lipoprotein cholesterol (LDL-C) ≤ 3.12 mmol/L. The Japanese NAGALA cohort ( n = 12,932) was cross-sectional: prevalent MASLD was ascertained at the baseline examination, and entry was restricted to fasting blood glucose (FBG) < 6.1 mmol/L. The Japanese cohort was used for replication rather than harmonized external validation, and all participants had BMI < 25 kg/m². Steatosis was defined by ultrasound and MASLD by the 2023 multisociety criteria. Associations were examined with multivariable logistic regression and restricted cubic splines, together with a Cox proportional-hazards sensitivity analysis of the Chinese follow-up data; discrimination was assessed with the area under the receiver operating characteristic curve (AUC), DeLong tests, category-free net reclassification improvement (NRI), integrated discrimination improvement (IDI), decision curve analysis and bootstrap validation. Sensitivity analyses addressed the overlap between index components and the MASLD criteria by using steatosis alone as the outcome and excluding cases qualifying only through shared criteria. Incident MASLD developed in 2,114 (13.1%) Chinese participants during follow-up, and prevalent MASLD was present in 1,132 (8.8%) Japanese participants. Fully adjusted odds ratios per standard deviation increase in TyG, TyG-BMI, METS-IR and SPISE were 2.39, 5.09, 3.65 and 0.15 in the Chinese cohort and 2.19, 3.71, 2.90 and 0.18 in the Japanese cohort (all P < 0.001); the corresponding hazard ratios in the Cox sensitivity analysis were 1.83, 3.25, 2.10 and 0.24 (all P < 0.001), and associations were attenuated but persisted in sensitivity analyses. TyG-BMI had the highest AUC (0.8755 and 0.8865), and TyG-BMI, METS-IR and SPISE each outperformed BMI (0.8248 and 0.8404; DeLong P < 0.001), whereas TyG did not. Adding each index to a baseline model of total cholesterol, triglycerides, FBG and BMI improved NRI and IDI but increased the AUC by at most 0.0078, a gain that became negligible when HDL-C was added to the baseline model and when the components were modeled flexibly; decision curve analysis showed no clinically meaningful difference in net benefit, and NRI gains should therefore be interpreted cautiously. In nonobese adults, BMI-integrated IR indices, particularly TyG-BMI, were associated with incident MASLD in the longitudinal Chinese cohort and with prevalent MASLD in the cross-sectional Japanese cohort, and discriminated both outcomes better than single anthropometric or biochemical markers (BMI, fasting glucose and triglycerides), whereas TyG alone did not outperform BMI. Because the two cohorts differ in design and outcome, their estimates should not be compared numerically. Shared diagnostic variables may inflate these estimates, and the incremental gains were small, so prospective external validation is required before any clinical use.