Chronic liver disease (CLD) is highly prevalent in XXX and commonly managed in primary care and shared-care settings. Esophageal variceal bleeding (EVB) is a severe and potentially fatal complication of advanced CLD that often presents unpredictably and substantially affects quality of life and end-of-life care. For family physicians, early identification of patients at high risk of EVB is essential to enable timely referral, preventive interventions, and informed discussions with patients and families.This study aimed to develop an interpretable machine learning model to support EVB risk stratification in non-cancer CLD patients within routine clinical practice.We conducted a retrospective cohort study using data from the XXX Medical Research Database between 2010 and 2017. Patients with liver cancer, uncomplicated CLD, or incomplete data were excluded. Several machine learning models were developed to predict EVB risk in non-cancer CLD patients, and their predictive performances were compared. Model interpretability was assessed using SHapley Additive exPlanations (SHAP) to enhance clinical applicability in primary care settings.A total of 4,080 patients were included, of whom 1,004 experienced EVB. Among the evaluated models, the extreme gradient boosting (XGBoost) model achieved the best predictive performance. The most influential predictors were clinical parameters, including C-reactive protein, hemoglobin, platelet count, and serum albumin.An interpretable XGBoost-based model using commonly available laboratory data can effectively stratify EVB risk in non-cancer CLD patients.This approach has potential value in primary care to support early risk recognition, guide referral decisions, and facilitate shared decision-making regarding prevention, monitoring, and palliative care planning.