Models to predict the short-term survival of acute-on-chronic liver failure patients following liver transplantation

Abstract

Background: Acute-on-chronic liver failure (ACLF) is featured with rapid deterioration of chronic liver disease and poor short-term prognosis. Liver transplantation (LT) is recognized as the curative option for ACLF. However, there is no standard in the prediction of the short-term survival among ACLF patients following LT.

Method: Preoperative data of 132 ACLF patients receiving LT at our center were investigated retrospectively. Cox regression was performed to determine the risk factors for short-term survival among ACLF patients following LT. Five conventional score systems (the MELD score, ABIC, CLIF-C OFs, CLIF-SOFAs and CLIF-C ACLFs) in forecasting short-term survival were estimated through the receiver operating characteristic (ROC). Four machine-learning (ML) models, including support vector machine (SVM), logistic regression (LR), multi-layer perceptron (MLP) and random forest (RF), were also established for short-term survival prediction.

Results: Cox regression analysis demonstrated that creatinine (Cr) and international normalized ratio (INR) were the two independent predictors for short-term survival among ACLF patients following LT. The ROC curves showed that the area under the curve (AUC) ML models was much larger than that of conventional models in predicting short-term survival. Among conventional models the model for end stage liver disease (MELD) score had the highest AUC (0.704), while among ML models the RF model yielded the largest AUC (0.940).

Conclusion: Compared with the traditional methods, the ML models showed good performance in the prediction of short-term prognosis among ACLF patients following LT and the RF model perform the best. It is promising to optimize organ allocation and promote transplant survival based on the prediction of ML models.

Keywords: ACLF; Liver transplantation; MELD; Machine-learning models; Prognosis.

Fig. 1

Flowchart of the search strategy and selection of studies for inclusion

Fig. 2

The overall survival curves of patients with the Kaplan–Meier estimator, tested with a log-rank test. a The overall survival curves of patients with different levels of Cr, P = 0.008. b The overall survival curves of patients with different levels of INR, P = 0.04. c The overall survival curves of patients with different levels of TBiL, P = 0.49. d The overall survival curves of patients with different levels of Plt, P = 0.41. e The overall survival curves of patients with different levels of WBC, P = 0.71. Cr creatinine; INR International Normalized Ratio; TBiL total bilirubin; Plt platelet; WBC White Blood Cells

Fig. 3

Scatter dot plot diagrams of the groups with conventional models to predict prognosis of ACLF following LT by Student’s t-test or Mann–Whitney U test. a Child puge between the survival group and death group, P > 0.05. b MELD score between the survival group and death group, P < 0.05. c CLIF-OF between the survival group and death group, P > 0.05. d CLIF-C ACLFs between the survival group and death group, P < 0.05. e CLIF-SOFA between the survival group and death group, P > 0.05. f ABIC between the survival group and death group, P > 0.05. The lines in the diagrams represent mean with SD. NS no significance *P < 0.05, **P < 0.01. MELD Model for end-stage liver disease; CLIF-SOFA Chronic liver failure-Sequential organ failure assessment; CLIF-C OF Chronic liver failure consortium Organ Failure score; ABIC age-bilirubin-international normalized ratio-creatinine

Fig. 4

ROC curve comparison of conventional models to predict prognosis of ACLF following LT. MELD Model for end-stage liver disease; CLIF-SOFA Chronic liver failure-Sequential organ failure assessment; CLIF-C OF Chronic liver failure consortium Organ Failure score; ABIC age-bilirubin-international normalized ratio-creatinine

Fig. 5

The ROC curves and average AUC of the machine learning models. K-fold cross validation (k = 5) was used to estimate and compare the performance of different machine learning models. After five rounds of training/validation rotation, the average AUC was calculated. a The support vector machine (SVM) model. b The logistic regression (LR) model. c The multi-layer perceptron (MLP) model. d The random forest (RF) model. ROCcurve receiver operating characteristic curve. AUC area under the curve