Machine learning-based model used for predicting Portal vein thrombosis risk for patients with cirrhosis

Abstract

Abstract We aimed to assess the risk of portal vein thrombosis (PVT) in patients with hepatitis B-related cirrhosis (HBC) using artificial neural networks (ANN). PVT can exacerbate portal hypertension and lead to complications, increasing the risk of mortality. Unfortunately, accurate prediction models for PVT in hepatitis B cirrhosis patients are currently insufficient. To address this gap, we conducted a study at Beijing Ditan Hospital, affiliated with Capital Medical University, involving 986 hospitalized patients. The patients were randomly divided into a training set (685 cases) and a validation set (301 cases) using a 3:1 ratio. Through univariate analysis, we determined independent factors that influence the occurrence of PVT, which were then utilized to develop an ANN model. The performance of the ANN model was assessed using various indicators, such as the area under the receiver operating characteristic curve (AUC) and concordance index (C-index). In the training group, PVT developed within three years in 19.0% of patients, and within five years in 23.7% of patients. Similarly, in the validation group, PVT developed within three years in 16.7% of patients, and within five years in 24.0% of patients. The ANN model incorporated nine independent risk factors, including age, presence of ascites, manifestation of hepatic encephalopathy (HE), occurrence of gastrointestinal varices with bleeding, Child-Pugh classification, alanine transaminase (ALT) levels, albumin (ALB) levels, neutrophil-to-lymphocyte ratio (NLR), and platelet count (PLT). Importantly, the AUC of the ANN model was significantly higher at 0.9718 compared to existing models such as MELD and CTP (all p<0.001). Our ANN model effectively classified patients into high ,medium, and low risk groups for PVT development over a span of 3 and 5 years. These findings were further validated in an independent cohort.