Risk prediction of acute variceal bleeding in hepatocellular carcinoma patients undergoing systemic therapy based on immune checkpoint inhibitors

Figure 1 Retrospective selection process of patients. A total of 395 patients with hepatocellular carcinoma (HCC) who received systemic therapy between January 2021 and December 2022 were initially screened. Patients were excluded due to non–HCC (n = 6), concurrent malignant tumors of other systems (n = 10), severe heart, lung, or liver disease (Child-Pugh C > 9) or kidney dysfunction (n = 17), absence of cirrhosis (n = 15), history of transjugular intrahepatic portosystemic shunt (n = 4), history of previous systemic therapy (n = 21), incomplete data (n = 36). The final cohort included 286 patients for analysis. ICIs: Immune checkpoint inhibitors; TIPS: Transjugular intrahepatic portosystemic shunt.

Figure 2 Nomogram of logistic regression affecting acute variceal bleeding. A binary logistic regression model was constructed, which revealed that a maximum tumor diameter ≥ 5 cm, portal vein tumor thrombus, pretreatment bleeding history, and high FIB-4 grade were independent risk factors for bleeding (all P < 0.05). FIB-4: Fibrosis-4 index.

Figure 3 Receiver operating characteristic analysis of the model prediction for acute variceal bleeding. A: Training cohort; B: Validation cohort. The area under the curve in training cohort and validation cohort were 0861 (95%CI: 0.802-0.920) and 0.816 (95%CI: 0.714-0.919) respectively. Optimal cutoff: 0.132 (sensitivity 70.5%, specificity 89.5%). AUC: Area under the curve.

Figure 4 Calibration curve analysis of the model prediction for acute variceal bleeding. A: Training cohort; B: Validation cohort. Internal validation with 1000 bootstrap replicates showed good model fit (Hosmer-Lemeshow P = 0.067) and discrimination (C-statistic 0.861). External validation maintained calibration stability (P = 0.161), indicating excellent calibration performance and readiness for clinical implementation.

Figure 5 Decision curve analysis of the model prediction for acute variceal bleeding. A: Training cohort; B: Validation cohort. Decision curve analysis (DCA) was performed to evaluate clinical utility by quantifying net benefit across threshold probabilities (0%-100%). The 'None' strategy (all patients are classified as negative) and 'All' strategy (all patients are classified as positive) served as references. The DCA demonstrated clinical utility across threshold probabilities 0%-80%, with net benefit superiority over 'treat none' strategy (reference line at y = 0). The DCA results confirm the model's clinical utility, with superior net benefit compared to alternative strategies.