Development and validation of a nomogram based on tumor margin irregularity and alpha-fetoprotein for predicting microvascular invasion in hepatocellular carcinoma
Source: PubMed Central Open Access, NCBI / U.S. National Library of Medicine
Objective To develop and externally validate a preoperative nomogram based on tumor margin irregularity and alpha-fetoprotein (AFP) positivity for predicting microvascular invasion (MVI) in hepatocellular carcinoma (HCC). Methods This retrospective multicenter study included a training cohort of 487 patients and an external validation cohort of 256 patients with pathologically confirmed HCC after liver resection. Demographic, clinical, tumor-related, serological, and preoperative computed tomography (CT) variables were compared by MVI status. Two blinded radiologists independently assessed qualitative CT features, and interobserver agreement was evaluated using Cohen’s κ. Least Absolute Shrinkage and Selection Operator (LASSO) regression with 10-fold cross-validation was used for predictor selection, followed by multivariable logistic regression. Model performance was assessed by receiver operating characteristic (ROC) curves, area under the curve (AUC), DeLong tests, calibration, the Hosmer–Lemeshow test, and decision curve analysis (DCA). Tumor size-related sensitivity analyses were performed. Results MVI was present in 187 of 487 patients in the training cohort and 95 of 256 patients in the validation cohort. LASSO selected AFP positivity, irregular tumor margins, capsular interruption, and intratumoral hyperplastic vessels. Multivariable analysis identified irregular tumor margins (adjusted odds ratio [OR] = 5.275, 95% confidence interval [CI]: 3.165–8.791, p < 0.001) and
Abstract
Objective To develop and externally validate a preoperative nomogram based on tumor margin irregularity and alpha-fetoprotein (AFP) positivity for predicting microvascular invasion (MVI) in hepatocellular carcinoma (HCC). Methods This retrospective multicenter study included a training cohort of 487 patients and an external validation cohort of 256 patients with pathologically confirmed HCC after liver resection. Demographic, clinical, tumor-related, serological, and preoperative computed tomography (CT) variables were compared by MVI status. Two blinded radiologists independently assessed qualitative CT features, and interobserver agreement was evaluated using Cohen’s κ. Least Absolute Shrinkage and Selection Operator (LASSO) regression with 10-fold cross-validation was used for predictor selection, followed by multivariable logistic regression. Model performance was assessed by receiver operating characteristic (ROC) curves, area under the curve (AUC), DeLong tests, calibration, the Hosmer–Lemeshow test, and decision curve analysis (DCA). Tumor size-related sensitivity analyses were performed. Results MVI was present in 187 of 487 patients in the training cohort and 95 of 256 patients in the validation cohort. LASSO selected AFP positivity, irregular tumor margins, capsular interruption, and intratumoral hyperplastic vessels. Multivariable analysis identified irregular tumor margins (adjusted odds ratio [OR] = 5.275, 95% confidence interval [CI]: 3.165–8.791, p < 0.001) and AFP positivity (adjusted OR = 3.297, 95% CI: 1.983–5.481, p < 0.001) as independent predictors. The final two-variable model achieved AUCs of 0.740 and 0.781 in the training and validation cohorts. Tumor size did not materially improve discrimination, and model performance was generally consistent across size categories. Calibration was acceptable, and DCA suggested exploratory clinical utility. Conclusion The nomogram based on tumor margin irregularity and AFP positivity showed fair-to-acceptable discrimination and may support exploratory preoperative MVI risk stratification in surgically treated or resectable HCC patients with clinical characteristics similar to those in the study cohorts, pending prospective multicenter validation.