Preoperative Combined Prediction Models Have Superior Capability in Predicting Survival as the Child-Pugh Grade in Patients with HCC after Interventional Embolotherapy

Abstract

Background

It is of important clinical significance for hepatocellular carcinoma (HCC) patients to evaluate prognosis before interventional embolotherapy.

Methods

A total of 106 patients with HCC after interventional embolotherapy who had complete data with follow-up information until September 2019 were included in this study. These data were analyzed using SPSS Version 22.0 and R (version 3.6.1) statistical software.

Results

1) The diameter of the tumor, ascites, FIT, AFP, ALT, AST, GGT, and Child–Pugh score had the ability to predict the prognosis and survival of patients with HCC. Among these molecules, the predictive effectiveness (or the area under the receiver operating characteristic [ROC] curve) of GGT was the highest, although it was slightly lower than the predictive effectiveness of the Child–Pugh score, which is the gold standard for survival analysis. 2) Among survival analyses combining five molecular indicators, the predictive postoperative viability for combination 1 was the strongest with an area under the ROC curve (AUC) of 0.856 (0.779, 0.932), similar to the all-molecular combination (combination 16) with an AUC of 0.872 (0.798, 0.945), but much higher than that of the Child–Pugh score of 0.720 (0.616, 0.823) for HCC patients (all p<0.05). 3) Kaplan–Meier analyses showed that the 3-year cumulative survival rates were 55.3% for low-risk patients and 2.6% for high-risk patients.

Conclusion

A combined prediction model can determine the optimal combination of preoperative routine detection indices in patients with HCC intervention, and ROC curve analysis can quantify the efficacy of these indices in the survival and prognosis of HCC. Interestingly, combination 1 showed stronger predictive capability than the Child–Pugh score in predicting death risks for postoperative patients with HCC. When combination 1 has several missing clinical data, these combination prediction models (12, 3, 7, 13, 16) are also a replaceable choice. These findings may have important clinical significance in the formulation of individualized medical programs.