Perioperative clinicopathologic model for predicting 12-month early recurrence after curative hepatectomy in...

Early recurrence (ER) within 12 months after curative hepatectomy remains a major determinant of poor survival in patients with hepatocellular carcinoma (HCC). Identifying high-risk patients using routinely available perioperative parameters may enable more tailored postoperative surveillance. This study aimed to develop and validate a machine-learning model based on clinical and laboratory variables to predict 12-month ER after resection.

This retrospective study included 100 consecutive patients who underwent curative-intent hepatectomy for pathologically confirmed HCC. Perioperative demographic, clinical, laboratory, and tumor-related variables were collected. A logistic regression model with L1 regularization was constructed. No missing values were observed among candidate predictors or outcome variables; therefore, all eligible patients were included in the final analysis without imputation. Continuous variables were standardized using z-score normalization, and categorical variables were transformed through one-hot encoding. Given the limited number of outcome events (n=40), penalized regression was applied to restrict model complexity and reduce the risk of overfitting. Model performance was evaluated using repeated stratified five-fold cross-validation (5×5 repetitions), and all performance metrics were calculated from out-of-fold predictions. Discrimination, calibration, and clinical utility were assessed using the area under the receiver operating characteristic curve (AUC), the area under the precision-recall curve (AUPRC), Brier score, calibration intercept and slope, and decision curve analysis (DCA).

Forty patients (40.0%) developed ER. Significant predictors of ER included diabetes mellitus (P=0.02), capsular invasion (P=0.02), multiple tumors (P=0.001), and lower total bilirubin levels (P=0.03). Edmondson-Steiner grade ≥3 was more frequent in the ER group but did not reach statistical significance (40.0% vs. 30.0%, P=0.23). The model demonstrated good discriminatory ability with an AUC of 0.862 and an AUPRC of 0.835. At the optimal cutoff (0.512), accuracy was 0.810, with a sensitivity of 0.800 and specificity of 0.817. Calibration showed good agreement between predicted and observed risk (Brier score =0.148). DCA indicated a meaningful net benefit across threshold probabilities of 0.20-0.60.

A machine-learning model based on routinely available perioperative clinicopathologic variables demonstrated favorable internally validated performance in predicting 12-month ER after curative hepatectomy for HCC. This accessible and practical tool may assist in postoperative risk stratification and individualized surveillance planning.