Abstract
BACKGROUND: Liver cirrhosis complicated by portal vein thrombosis (PVT) is a fatal complication with no specific manifestations but often misdiagnosed, it crucially increases the mortality worldwide. This study aimed to identify risk factors and establish a predictive model for diagnosis of venous thrombosis clinical by routine blood tests and endoscopic characteristics. METHODS: Patients from Gansu Provincial Hospital from October 2019 to December 2023 were enrolled. The retrospective modelling cohort was screened by propensity score matching (PSM) at a 1:1 ratio from the baseline characteristics before endoscopic diagnosis. Variables were collected from blood test and endoscopic signs using machine learning method (ML). Logistic regression determined risk factors. The predictive performance was evaluated by receiver operation curve (ROC), calibration curve, clinical decision analysis (DCA) and influence curve (CIC). Furthermore, external cohort was used for validation, an online nomogram was established. RESULTS: A total of 1,058 patients were enrolled, and 470 patients were included after PSM 1: 1. The model identified 7 factors, including splenectomy, blood urea nitrogen (BUN), serum sodium, activated partial thromboplastin time (APTT), thrombin time (TT), D-dimer, and degree of oesophageal varices. The area under the curve (AUC) was 0.907 (95% CI, 0.877-0.931). The calibration curve, decision and clinical impact curves showed the model demonstrated a good predictive accuracy and clinical benefits. The validation got an AUC of 0.890 (95% CI, 0.831-0.934), A nomogram tool was finally established for application. CONCLUSION: Blood test combined endoscopic characters could preliminarily predict the liver cirrhosis with portal vein thrombosis for cirrhotic patients undergoing endoscopic examination.