A Machine Learning based Approach for Liver Fibrosis Diagnosis in NAFLD Using Biomarkers and Demographics

Non-alcoholic fatty liver disease (NAFLD) has become one of the leading causes of chronic liver disorders worldwide, with advanced liver fibrosis posing significant risks of liver failure and other complications. Early and accurate detection of liver fibrosis is crucial for the effective management of NAFLD. General diagnosis of fibrosis is based on liver biopsies, but this is invasive and impractical for large-scale screening. So, finding a robust, simple and non-invasive way to diagnose advanced fibrosis in patients with NAFLD is important to monitor the disease and decide on treatment. In this work, a non-invasive diagnostic model was developed using biomarkers and demographics that can help differentiate between early and advanced fibrosis in patients with NAFLD from Southeast Asia. Further, the dataset consisted of 784 biopsy confirmed patients with NAFLD from China, Malaysia, and India, with 540 patients in a primary Discovery Set and 244 in two independent Validation Sets. The development of the model relies on eight key characteristics: age, BMI, FBG, DM/IFG, ALT, GGT, TG, and AST/PLT ratio. Overall, the proposed model achieved 94% recall, 98% NPV, 82% accuracy and 89% AUROC, with a 13% higher true positive rate than existing LiveFbr tool. Additionally, a web application has been developed to provide patients, doctors and healthcare practitioners with easy access to accurate liver fibrosis stage predictions. This work further highlights the potential of data-driven models to make significant contributions in healthcare.