Machine learning-based early survival prediction in early-onset hepatocellular carcinoma: a SEER-based multi-model comparative study.

Background: Early-onset hepatocellular carcinoma (eHCC) presents distinct clinical challenges; however, prognostic determinants remain inadequately characterized. This study aimed to develop and validate machine learning (ML) models to predict early survival in patients with eHCC and to identify key prognostic factors. Methods: Patients diagnosed with eHCC were identified from the Surveillance, Epidemiology, and End Results (SEER) database [2010-2021] based on predefined inclusion criteria. The cohort was randomly partitioned into training (70%) and validation (30%) sets. Five ML algorithms were implemented: multilayer perceptron (MLP), logistic regression (LR), support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost). Additionally, a conventional model based on the American Joint Committee on Cancer (AJCC) staging system was constructed as a benchmark for comparison with the ML-based models. Model discrimination was assessed by the area under the receiver operating characteristic curve (AUC). Calibration curves and complementary metrics quantified predictive accuracy. The SHapley Additive exPlanations (SHAP) framework was employed to rank variable importance and interpret model predictions. Results: A total of 1,922 patients with eHCC were included, of whom 1,345 were assigned to the training cohort and 577 to the validation cohort. Thirteen clinicopathological and treatment-related variables were incorporated into the analysis, including age, sex, race, pathological grade, surgical status, AJCC stage, SEER stage, tumor size, alpha-fetoprotein (AFP) level, marital status, radiotherapy, chemotherapy, and macrovascular invasion. The RF model demonstrated optimal performance in the validation cohort, achieving an AUC of 0.80, accuracy of 0.72, precision of 0.71, recall of 0.81, and F1 score of 0.75. This substantially outperformed the AJCC staging-based model with AUC of 0.66, representing a ΔAUC of 0.14 and 21.2% relative improvement. SHAP analysis identified surgical intervention as the dominant predictor in the RF model. Conclusions: The RF model demonstrated robust discriminative ability for predicting early survival in eHCC patients. This validated model can facilitate risk stratification and inform individualized therapeutic decisions for patients with eHCC.