TyG index boosts simple MASLD screening model for rural older adults

Metabolic dysfunction-associated steatotic liver disease, or MASLD, has become the most common chronic liver condition worldwide, affecting more than 30 percent of adults globally, and its burden is especially heavy in rural China, where prevalence estimates reach 38.0 percent compared with 34.4 percent in urban areas of Guangdong Province. Yet many rural primary care facilities lack ultrasound equipment and trained personnel for systematic screening, leaving a large share of cases undetected. A new cross-sectional study from Putian City in Fujian Province now reports that a screening model built entirely from routine health examination data, with the triglyceride-glucose (TyG) index and uric acid as key additions, can identify prevalent MASLD in rural older adults with moderate accuracy, offering a practical first-pass tool for resource-limited settings.

The study, published in Discover Public Health, analyzed health examination records collected from 1 January to 31 December 2024 from older adults in a rural area of Putian City. Of 3,258 initial records, 163 participants were younger than 60 years, and another 75 age-eligible individuals were excluded, most commonly for severe missingness or physiologically implausible values (38 people) or estimated chronic heavy alcohol consumption of at least 30 grams per day for men or 20 grams per day for women (37 people). The final analytic cohort comprised 3,020 adults aged 60 and older, of whom 792, or 26.2 percent, met the concurrent MASLD definition used in the study.

That definition followed the 2024 Chinese guideline and required three elements: hepatic steatosis detected on abdominal ultrasonography, at least one cardiometabolic criterion, and no history of chronic heavy drinking or other documented cause of hepatic steatosis. The cardiometabolic criteria incorporated documented diabetes as meeting the hyperglycemia criterion, and antihypertensive or lipid-lowering medication use counted toward the hypertension, hypertriglyceridemia, and low HDL-C criteria. The cohort included 639 participants with documented diabetes, 213 of whom had MASLD. Notably, viral hepatitis status was not available in the dataset and therefore could not be evaluated as an exclusion criterion, a limitation the authors acknowledge.

To avoid incorporation bias, the researchers prespecified a core model containing only variables that are not components of the MASLD definition: sex, age, alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transferase (GGT), and serum creatinine. Uric acid and the TyG index, calculated as the natural logarithm of triglycerides in milligrams per deciliter multiplied by fasting plasma glucose in milligrams per deciliter divided by two, were then added separately and jointly. No data-driven variable screening, stepwise selection, or LASSO selection was used in the primary model-development pathway. An exploratory two-marker model containing only TyG and uric acid was also fitted as a lower-resource option.

The results were striking in their internal hierarchy. The TyG index showed the strongest independent association with concurrent MASLD, with an odds ratio of 2.823 per 1-unit increase (95 percent confidence interval 2.422 to 3.290, P < 0.001). Uric acid was also independently associated, but far more weakly, at an odds ratio of 1.041 per 10 micromoles per liter (95 percent CI 1.029 to 1.054). The TyG index also revealed a clear dose-response pattern: MASLD prevalence rose monotonically from 7.7 percent (58 of 755) in the lowest TyG quartile to 18.0 percent in the second, 29.3 percent in the third, and 49.9 percent (377 of 755) in the highest quartile, with a trend test P value below 0.001.

In terms of discrimination, the prespecified core model alone achieved an AUC of 0.732 (95 percent CI 0.711 to 0.753). Adding uric acid raised the AUC to 0.750, a modest increment of 0.019, while adding TyG produced a larger improvement to 0.774, and the joint full model reached an apparent AUC of 0.786 (95 percent CI 0.768 to 0.804). After internal validation with 1,000 bootstrap resamples and model refitting, the mean optimism was only 0.0024, yielding an optimism-corrected AUC of 0.783 (95 percent CI 0.764 to 0.800). At a sample-derived probability cutoff of 0.261, sensitivity was 72.1 percent and specificity 72.2 percent, though the authors stress this cutoff is descriptive rather than a clinical decision threshold.

Calibration told a more nuanced story. The Hosmer-Lemeshow test was significant (chi-squared 39.780, P < 0.001), indicating detectable local miscalibration in portions of the risk range, and the smoothed calibration curve departed from the ideal line. Yet the calibration intercept (−0.001) and slope (1.002) were close to ideal values, and the model’s Brier score of 0.155 improved on the null-model score of 0.193, giving a Brier skill score of 0.197. The authors interpret this combination as improved average probabilistic accuracy with residual calibration deviations that warrant external reassessment and possible recalibration before any clinical application.

Standalone performance differed sharply between the two candidate markers. TyG alone achieved moderate discrimination with an AUC of 0.736 (95 percent CI 0.716 to 0.756), essentially matching the core liver-enzyme model, and a sample-specific Youden-derived cutoff of 8.675 that the authors explicitly decline to promote for clinical use without external validation. Uric acid alone, by contrast, showed limited discriminatory ability with an AUC of 0.590 (95 percent CI 0.567 to 0.613), and no clinical cutoff is proposed for it. The exploratory TyG plus uric acid model reached an AUC of 0.743 (95 percent CI 0.723 to 0.762), a small but statistically significant improvement over TyG alone (delta AUC 0.007, paired DeLong P = 0.020), with a Brier score of 0.167.

Reclassification analyses reinforced the same hierarchy. Adding uric acid to the core model yielded a continuous net reclassification improvement of 0.358 (95 percent CI 0.278 to 0.452) and an integrated discrimination improvement of 0.023; adding TyG produced larger values of 0.517 and 0.069, and the joint addition produced the largest, at 0.600 and 0.083. The authors caution, however, that continuous NRI counts any directional change in predicted probability, however small, and should be read as hypothesis-generating rather than definitive proof of clinical utility, particularly when predictors are correlated.

Several caveats frame the findings. Because the design is cross-sectional, the model identifies prevalent MASLD at the time of examination rather than predicting future incident disease, and the adjusted odds ratios, including inverse coefficients for AST and creatinine that the authors attribute to statistical suppression effects, must not be read as causal. HbA1c and oral glucose tolerance data were unavailable, medication records existed only as free text, ultrasound can miss mild steatosis, and comparison of the 75 excluded with the 3,020 included participants showed the excluded were more often male (73.3 percent versus 40.5 percent) with higher GGT and creatinine, raising possible selection bias. Internal benchmarking showed HSI and FLI achieved slightly higher AUCs of 0.808 and 0.809 in the same dataset, but both incorporate variables overlapping the MASLD definition, so their advantage may partly reflect incorporation bias. With no external validation yet performed, the authors estimate that roughly 800 participants, providing about 200 MASLD cases, would be needed for an informative validation study in a geographically distinct primary care setting, and they position the model strictly as a preliminary screening tool requiring prospective and external validation before clinical implementation.

Subject of Research: Development of a MASLD screening model using routine health indicators including the TyG index and uric acid in rural older adults

Article Title: A cross-sectional screening model using routine health indicators including the TyG index and uric acid for MASLD in rural elderly populations

Article References: Cai, D., Liang, L., Lin, S., Yao, L., & Chen, T. (2026). A cross-sectional screening model using routine health indicators including the TyG index and uric acid for MASLD in rural elderly populations. Discover Public Health, 23(1), Article 1721. https://doi.org/10.1186/s12982-026-03215-z

Image Credits: AI Generated

DOI: 10.1186/s12982-026-03215-z

Keywords: MASLD, TyG index, uric acid, screening model, rural elderly, cross-sectional study, primary care, bootstrap validation, calibration, China, fatty liver disease, health examination