LASSO-Based Machine Learning Model for Prediction of Liver Failure in Hepatocellular Carcinoma Patients Undergoing TACE

Abstract PURPOSE Transcatheter arterial chemoembolization (TACE) is a commonly used method for the nonsurgical treatment of hepatocellular carcinoma (HCC); however, it can cause liver failure with rapid progression and high mortality. METHODS We organized and analyzed the data of patients with HCC undergoing TACE at our hospital. Screening indicators related to liver failure were analyzed using least absolute shrinkage and selection operator (LASSO) regression to establish a predictive model. RESULTS Prothrombin activity (odds ratio [OR] [95% confidence interval (CI)], 0.965 [0.931–0.997]; p = 0.040), tumor number (OR [95% CI], 2.328 [1.044–5.394]; p = 0.042), and vascular invasion (OR [95% CI], 2.778 [1.006–7.164]; p = 0.039) are independent risk factors for liver failure after TACE. The prediction model established based on these results had areas under the curve of 0.821 and 0.813 for the training and validation groups, respectively. CONCLUSION The prediction model established using LASSO regression can predict the risk of liver failure after TACE and confirm whether patients with advanced HCC can benefit from TACE.