Development and temporal validation of 1-, 2-, and 3-year survival prediction models for hepatocellular carcinoma using the SEER database.

Hepatocellular carcinoma (HCC) accounts for most primary liver cancers and displays substantial variability in survival outcomes. Existing staging systems, such as the American Joint Committee on Cancer (AJCC), Barcelona Clinic Liver Cancer (BCLC), and China Liver Cancer Staging (CNLC) systems, often fail to capture individual heterogeneity. Additionally, many published prognostic models rely on small cohorts, predict survival at only a single timepoint, or lack temporal or external validation. Using data from the Surveillance, Epidemiology, and End Results (SEER) Program, we developed and temporally validated multi-timepoint Cox models to predict 1-, 2-, and 3-year overall survival in HCC. A total of 3850 patients diagnosed between 2018 and 2022 were included for model development and internal testing, and 2268 patients diagnosed in 2016–2017 formed an independent temporal validation cohort. Age, alpha-fetoprotein (AFP) level, fibrosis status, AJCC stage, and surgical treatment were identified as independent prognostic factors and incorporated into a multi-timepoint nomogram. The model achieved C-indices of 0.762, 0.760, and 0.726 in the training, test, and temporal validation cohorts. Time-dependent AUCs at 1, 2, and 3 years were 0.814, 0.801, and 0.780 in the training cohort; 0.824, 0.809, and 0.807 in the test cohort; and 0.794, 0.793, and 0.779 in the temporal validation cohort, demonstrating consistently strong discrimination. Calibration and decision curve analyses confirmed accurate agreement and clinical benefit. These findings indicate that the proposed model provides a robust tool for individualized survival prediction and risk-stratified clinical decision-making in HCC.