Revolutionizing hepatic fibrosis staging: A machine learning approach combining clinical, biochemical, and microbiome...

Non-alcoholic Steatohepatitis (NASH) is a common disease that not only affects adults but has also been seen to affect all ages. This includes young adults, children and even babies. Non-alcoholic fatty liver disease (NAFLD), NASH and the progression of it to fibrosis have been the subject of extensive research, as there still remains a great deal we do not understand. There are multiple factors that will influence how quickly and aggressively the progression of the disease occurs and also the way the disease is diagnosed or assessed, such as medical history, blood results and ultrasound imaging. This study aims to look at the use of machine learning (ML) to integrate clinical, biochemical and microbiome data to create a model to allow for non-invasive staging of hepatic fibrosis for NASH patients. A total of 1834 patients with biopsy-confirmed NASH were included in the retrospective analysis. The cohort was comprised of patients from multiple healthcare systems with known biopsy-confirmed NASH and a stated fibrosis stage (F0, F1, F2, F3, F4). A range of clinical variables, including liver function tests, demographics and microbiome profiles (via 16S rRNA gene sequencing), were included to train the machine learning models (Random Forest & Extreme Gradient Boosting). The performance of these models were assessed using 10-fold cross-validation with the primary training cohort and external validation on an independent hospital database. The models demonstrated excellent classification accuracy, specifically a balanced accuracy of 99.1% for RF and an area under the curve (AUC) value of 1.0 for XGBoost. The addition of microbiome features (specifically, diversity indices and the relative abundance of certain taxa) enhanced the models' predictive capability, indicating that the gut-liver axis plays a significant role in the development of NASH. To interpret the machine learning models, we used SHapley Additive Explanations (SHAP) analysis to identify which of the clinical and microbiome features affected the models' predictions for the fibrosis stages. Advanced stages of fibrosis (F3 & F4) were found to have significant dysbiosis in the microbiome with increased relative abundance of pathogenic bacteria including Escherichia-Shigella and Enterococcus, as well as decreased Akkermansia and Ruminococcus. The study provides evidence for the accuracy of a non-invasive method of determining hepatic fibrosis stage in NASH and demonstrates its superiority compared to traditional scoring systems (i.e. APRI, FIB-4) for the purpose of guiding clinical decision making and risk assessment for patients in clinical practice.