A Machine Learning Model for Predicting Hepatocellular Carcinoma Development in Patients with Hepatitis B Virus-Related Cirrhosis Receiving Antiviral Therapy

Background There were few prediction models specifically designed to evaluate the hepatocellular carcinoma (HCC) risk in chronic hepatitis B patients with cirrhosis. This study aimed to develop a machine learning-based prediction model to assess the risk of HCC in patients with hepatitis B virus (HBV)-related cirrhosis undergoing nucleos(t)ide analogue (NA) therapy. Methods This study included 1,592 patients with HBV-related cirrhosis who had received entecavir, tenofovir disoproxil fumarate, or tenofovir alafenamide for at least one year. Patients developing HCC within 12 months of NA therapy were excluded. Patients were randomized in a 2:1 ratio into the derivation or validation group, and the prediction model was developed using the eXtreme Gradient Boosting (XGBoost) algorithm. The Shapley Additive exPlanations (SHAP) model was employed to assess each parameter's contribution to prediction accuracy. Results The cumulative HCC incidences in all patients at 5, 8 and 10 years were 15.2%, 22.7%, and 25.7%, respectively. Our ML-HCC model incorporated six parameters: serum albumin and platelet count at treatment initiation, and age, platelet count, aspartate transaminase, and AFP level at 1 year of NA therapy. In the validation group, AUROCs of the ML-HCC model ranged from 0.79 to 0.80 over 3 to 10 years, outperforming extant models including APA-B, PLAN-B, PAGE-B, mPAGE-B, REACH-B, and CU-HCC (AUROCs: 0.61–0.73, p < 0.05). Risk stratification showed 10-year HCC incidences for the low-, intermediate-, and high-risk groups were 9%, 26%, and 67%, respectively. Conclusion The proposed machine learning model for predicting HCC development exhibited good predictive performance in patients with HBV-related cirrhosis undergoing NA therapy.