Abstract:Background and Aims Clinical outcomes after targeted therapy vary substantially among patients with primary liver cancer (PLC). Early identification of patients at high risk of poor prognosis is essential for individualized treatment and prognostic assessment. This study aimed to identify risk factors associated with poor prognosis and develop a risk prediction model for PLC patients receiving targeted therapy.Methods Clinical data from 160 PLC patients who underwent targeted therapy between June 2016 and June 2019 were retrospectively collected and followed for 5 years. According to RECIST 1.1, patients were classified into a favorable prognosis group (n=112) and an unfavorable prognosis group (n=48). LASSO regression was used for variable selection, followed by multivariate Logistic regression to identify independent prognostic factors. A risk prediction model was established and evaluated using calibration analysis, the Hosmer-Lemeshow goodness-of-fit test, and receiver operating characteristic (ROC) curve analysis.Results Poor prognosis occurred in 48 patients (30.0%). Ten candidate variables were selected by LASSO regression. Multivariate Logistic regression identified tumor diameter >5 cm (OR=3.216, 95% CI=1.582-6.537), BCLC stage C (OR=2.779, 95% CI=1.367-5.649), Cheng's classification type Ⅱ-Ⅳ portal vein tumor thrombus (OR=4.573, 95% CI=2.191-9.545), incomplete tumor capsule (OR=3.099, 95% CI=1.524-6.300), multinodular confluent tumor margin (OR=4.121, 95% CI=2.027-8.377), and elevated AFP level (OR=2.380, 95% CI=1.171-4.838) as independent predictors of poor prognosis (all P<0.05). The model demonstrated satisfactory performance, with a C-index of 0.784 and good calibration (Hosmer-Lemeshow test, P=0.359). The AUC was 0.851 (95% CI=0.764-0.938), with a sensitivity of 0.846 and a specificity of 0.894.Conclusion Large tumor size, advanced BCLC stage, extensive portal vein tumor thrombus, incomplete capsule, multinodular confluent tumor margin, and elevated AFP level are independent risk factors for poor prognosis in PLC patients receiving targeted therapy. The LASSO-Logistic regression-based model shows good discrimination and calibration and may facilitate risk stratification and individualized clinical management.