Lung ultrasound performed by primary care physicians as a predictive and diagnostic tool in COVID-19 patients.

Cirrhosis, a leading cause of chronic liver disease-related mortality, has seen rising incidence and death rates since 1990, accounting for 2.4% of global deaths in 2019 [,,]. Clinically, it progresses from a compensated to a decompensated phase, the latter presenting as hepatic encephalopathy (HE), ascites, or variceal hemorrhage [,]. Among these, HE stands out as it reflects cirrhosis progression and independently predicts poor survival, with a median post-diagnosis survival of only 0.92 years [,]. Approximately 40% of cirrhosis patients remain undiagnosed until a decompensating event occurs []; early detection could significantly improve both survival and quality of life [,]. HE presents a spectrum of symptoms ranging from mild cognitive impairment (minimal hepatic encephalopathy, MHE) to coma. Because its manifestations are nonspecific, the condition is often misdiagnosed or overlooked []. Current practice relies on neuropsychological tests, such as the Stroop test and Psychometric Hepatic Encephalopathy Score (PHES), for MHE screening, especially in high-risk patients with comorbid depression []. These tests are time-consuming and require specialized training, limiting their use in routine care []. Overlap between HE symptoms and other cirrhosis comorbidities (depression, uremic encephalopathy) compounds the diagnostic difficulty [,,]. Early detection could slow disease progression and improve outcomes of cirrhosis, and yet current screening methods fall short of this goal. Therefore, molecular biomarker-based early warning systems are needed for clinical deployment. In recent years, machine learning methods have been increasingly applied in chronic liver disease and hepatocellular carcinoma outcome prediction, representing a rapidly growing area of research. Numerous studies have focused on developing predictive models using routine clinical data to identify high-risk patients and guide early intervention. For instance, Grubert Van Iderstine et al. conducted a large-scale retrospective cohort study developing and validating an optimized XGBoost model to predict 1-, 3-, 5-, and 10-year hepatic decompensation risk at the time of the first outpatient visit in cirrhotic patients []. An et al. developed a multicenter model that integrated natural language processing for automated electronic health record data extraction with an XGBoost algorithm to predict prognosis in intermediate-stage HCC patients undergoing transarterial chemoembolization (TACE) []. Zhu et al. developed and validated an XGBoost-based machine learning model using large-scale population cohort data, demonstrating that the model could screen individuals with significant liver fibrosis risk in the general population, using easily accessible demographic and clinical indicators, with the risk score significantly correlated with all-cause mortality []. Collectively, these studies underscore the potential of machine learning, particularly XGBoost, for enabling early identification of high-risk individuals across the full spectrum of chronic liver disease, from fibrosis screening to decompensation prediction and post-treatment prognosis. However, despite these advances, an interpretable machine learning model specifically tailored for HE risk prediction in cirrhotic patients remains lacking.