MASLD Risk Score (MRS) and Random Forest model (RF Model): Novel tools for screening and severity assessment of MASLD

Metabolic Dysfunction-Associated Steatotic Liver Disease (MASLD) spans from simple steatosis to progressive forms like non-alcoholic steatohepatitis (NASH), fibrosis, and cirrhosis, making early diagnosis and grading crucial. This study aimed to develop and validate predictive models for diagnosing and assessing MASLD severity using routinely available biomarkers from both public and clinical datasets.

We developed a novel MASLD Risk Score (MRS) using data from the CDC NHANES (2000-2020) and validated it in a clinically profiled Indian cohort. Unlike existing indices, the predictors were derived through data-driven feature selection from large dataset, ensuring statistical robustness. It integrates novel (Uric Acid, HOMA-IR) and established (liver enzymes, triglycerides, waist circumference, BMI) biomarkers to improve metabolic profiling and predictive accuracy. The MRS also uniquely enables grading of MASLD severity, addressing a key limitation of previous models. The MRS achieved AUROCs of 0.91 (public) and 0.85 (clinical) with accuracies of 94 % and 82 %, respectively. A Random Forest (RF) model built on the same features provided AUROCs of 0.87 (public) and 0.94 (clinical), with accuracies of 83 % and 82 %. MRS parameters were optimized using a diverse population, improving generalizability across demographics. Both models showed strong correlation with ultrasonography results and outperformed existing indices.

The MRS offers a novel, interpretable, and cost-effective solution for MASLD screening. Its development from a large, demographically diverse population and incorporation of varied biomarkers supports generalizability. While results are promising, external validation in multi-center clinical settings is needed to confirm broad utility.