A novel prognostic model to predict outcome of artificial liver support system treatment

Abstract

The prognosis of Artificial liver support system (ALSS) for hepatitis B virus-related acute-on-chronic liver failure (HBV-ACLF) is hard to be expected, which results in multiple operations of ALSS and excessive consumption of plasma, increase in clinical cost. A total of 375 HBV-ACLF patients receiving ALSS treatment were randomly divided a train set and an independent test set. Logistic regression analysis was conducted and a decision tree was built based on 3-month survival as outcome. The ratio of total bilirubin before and after the first time of ALSS treatment was the most significant prognostic factor, we named it RPTB. Further, a decision tree based on the multivariate logistic regression model using CTP score and the RPTB was built, dividing patients into 3 main groups such as favorable prognosis group, moderate prognosis group and poor prognosis group. A clearly-presented and easily-understood decision tree was built with a good predictive value of prognosis in HBV-related ACLF patients after first-time ALSS treatment. It will help maximal the therapeutic value of ALSS treatment and may play an important role in organ allocation for liver transplantation in the future.

Figure 1

ROC (receiver-operation-characteristics) plots and AUC (area-under-curve) values of the train set and the test set for univariate analysis of significant factors. *Presented in log₁₀ scale.

Figure 2

ROC plots and AUC values of the train set and the test set for multivariate analysis of selected combinations of significant factors.

Figure 3

Conditional inference tree model for prediction of 3-month survival in ALSS-treated HBV-ACLF patients. (A) The decision tree; (B) ROC plot and AUC value for the tree model.