Development and validation of machine learning models for predicting immunotherapy rechallenge efficacy in patients with hepatocellular carcinoma

Hepatocellular carcinoma (HCC) remains one of the leading causes of cancer-related mortality worldwide. Although immune checkpoint inhibitors have emerged as an important therapeutic option for advanced HCC, the clinical benefit of immunotherapy rechallenge following prior treatment failure remains highly heterogeneous. Therefore, accurate and non-invasive prediction of treatment response is crucial for optimizing patient selection and supporting individualized therapeutic decision-making. This retrospective cohort study included 157 patients with HCC who underwent immunotherapy rechallenge at a single tertiary care center between June 2020 and June 2025. Patients were randomly assigned to a training cohort ( n = 109) and a testing cohort ( n = 48) using a stratified 7:3 allocation scheme. An independent external validation cohort comprising 40 patients from another tertiary care center during the same period was also included. Demographic characteristics, tumor-related features, laboratory parameters, and treatment-related variables were systematically collected. Multiple machine learning (ML) algorithms were developed and systematically compared to predict treatment response to immunotherapy rechallenge. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, and the Brier score. Kaplan–Meier survival analyses were further performed to assess prognostic risk stratification based on model-derived predictions. The logistic regression model demonstrated stable predictive performance, with an AUC of 0.777 in the training cohort and 0.806 in the testing cohort. In the independent external validation cohort, the model maintained good discriminative performance, achieving an AUC of 0.805. AFP, alcohol consumption, the AST/ALT ratio, ECOG performance status, and ascites were identified as key predictors associated with non-response to immunotherapy rechallenge. Model-derived risk stratification revealed distinct survival outcomes, with patients in the low-risk group exhibiting significantly longer progression-free survival and overall survival than those in the high-risk group. Our ML-based model accurately identified patients who were likely or unlikely to benefit from immunotherapy rechallenge, thereby providing a clinically practical tool to support individualized treatment decisions in HCC.