Comparison of preoperative CONUT score and PNI for predicting early severe complications after TACE in hepatocellular carcinoma patients and construction of a nomogram model

To compare the predictive value of the Controlling Nutritional Status (CONUT) score and the Prognostic Nutritional Index (PNI) for early severe complications after transarterial chemoembolization (TACE) in hepatocellular carcinoma (HCC) patients, and to develop and validate a nomogram for individualized risk prediction. This single-center retrospective cohort study enrolled 548 primary HCC patients undergoing TACE between January 2022 and June 2026, divided chronologically into training ( n = 365) and temporal validation ( n = 183) sets. The primary outcome was early severe complications (Clavien-Dindo grade ≥ III) during hospitalization or within 30 days after discharge. Variable selection was conducted using univariate analysis and least absolute shrinkage and selection operator regression, after which multivariate logistic regression was performed. A nomogram was then developed and assessed via receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). The DeLong test was applied to compare the area under the curve (AUC) of the PNI- and CONUT-based models. Among the 548 patients, 118 (21.53%) developed early severe complications. Multivariate analysis identified portal vein tumor thrombus (PVTT), C-reactive protein (CRP), prothrombin time (PT), aspartate aminotransferase-to-platelet ratio index (APRI), and PNI < 45 as independent risk factors (all P < 0.05). CONUT score ≥ 4 was excluded due to lack of statistical significance ( P = 0.053). The DeLong test showed no significant difference in AUC between the PNI-based and CONUT-based models (0.866 vs. 0.847, ΔAUC = 0.019, 95% CI: -0.003 to 0.042, P = 0.095). The nomogram incorporating the five predictors achieved good discrimination in both the training set (AUC = 0.866, 95% CI: 0.822–0.910) and the temporal validation set (AUC = 0.864, 95% CI: 0.794–0.935), with good calibration (Hosmer-Lemeshow test: P = 0.244 and P = 0.711, respectively). DCA confirmed its clinical utility. The PNI-based and CONUT-based models showed numerically comparable predictive performance. PNI retained independent significance in multivariate analysis, whereas CONUT did not. The nomogram incorporating PVTT, CRP, PT, APRI, and PNI showed good discrimination, calibration, and net benefit, serving as a practical tool for risk stratification and individualized management.