Predictive performance of noninvasive factors for liver fibrosis in severe obesity: a screening based on machine learning models.

Abstract

Objectives: Liver fibrosis resulting from nonalcoholic fatty liver disease (NAFLD) and metabolic disorders is highly prevalent in patients with severe obesity and poses a significant health challenge. However, there is a lack of data on the effectiveness of noninvasive factors in predicting liver fibrosis. Therefore, this study aimed to assess the relationship between these factors and liver fibrosis through a machine learning approach.

Methods: This study involved 512 patients who underwent bariatric surgery at an outpatient clinic in Mashhad, Iran, between December 2015 and September 2021. Patients were divided into fibrosis and non-fibrosis groups and demographic, clinical, and laboratory variables were applied to develop four machine learning models: Naive Bayes (NB), logistic regression (LR), Neural Network (NN) and Support Vector Machine (SVM).

Results: Among the 28 variables considered, six variables including (fasting blood sugar (FBS), skeletal muscle mass (SMM), hemoglobin, alanine transaminase (ALT), aspartate transaminase (AST) and triglycerides) showed high area under the curve (AUC) values for the diagnosis of liver fibrosis using 2D shear wave elastography (SWE) with LR (0.73, 95% CI: 0.65, 0.81) and SVM (0.72, 59% CI: 0.64, 0.80) models. Furthermore, the highest sensitivities were reported with SVM (0.83, 95% CI: 0.72, 0.91) and NB (0.66, 95% CI: 0.53, 0.77) models, respectively.

Conclusion: The predictive performance of six noninvasive factors of liver fibrosis was significantly superior to other factors, showing high application and accuracy in the diagnosis and prognosis of liver fibrosis.

Supplementary information: The online version contains supplementary material available at 10.1007/s40200-025-01564-1.