Abstract
Background: Microvascular invasion (MVI) is a crucial factor for early recurrence and poor outcomes in hepatocellular carcinoma (HCC). However, there are few studies on M2 classification. We aimed to build a predictive model for M2 in early-stage HCC, assisting clinical decision-making.
Methods: We retrospectively enrolled 451 patients with early-stage HCC and employed multiple machine learning algorithms to identify the risk factors influencing the robustness of M2. Model performance was evaluated using receiver operating characteristic (ROC) curve, calibration curve, decision curve analysis (DCA)，and clinical impact curve (CIC).
Results: There were 363 M0 and 88 M2 cases. Differences in recurrence-free survival (RFS) and overall survival(OS) between the M0 and M2 groups were statistically significant (P <0.0001). Complement C3, tumor size> 5cm, incomplete tumor capsule, and Edmondson-Steiner stage III-IV were independent risk factors for M2.The prediction model showed an area under the receiver operating characteristic curve(AUROC) of 0.765 and 0.807 in the training and validation groups, respectively. Calibration curves showed good agreement between actual and predicted M2 risks, and the DCA and CIC showed a significant clinical efficacy.
Conclusion: The nomogram-based model had a good predictive effect for M2 in patients with early-stage HCC ,providing guidance for treatment decisions.