摘要Objective Acute respiratory distress syndrome(ARDS)is a frequent complication in patients with acute pancreatitis(AP),particularly those who develop sepsis,and is associated with a longer duration of hospitalization and mortality.Therefore,developing a predictive model to identify these high-risk patients is of significant clinical importance.Methods Retrospective data from 572 patients with AP complicated by sepsis identified using ICD-9 and ICD-10 codes were evaluated.Re-cursive feature elimination with cross-validation(RFECV)was used to identify significant predictors of ARDS.Four machine learning(mL)models-AdaBoost,RandomForest,LightGBM,and XGBoost-were developed and rigorously validated.SHapley Additive exPlanations(SHAP)analysis was used to interpret the influence of distinct features on model predictions.Results Of the 572 patients includ-ed,47.6％developed ARDS.The XGBoost model,achieving an Area Under the Curve(AUC)of 0.803(95％CI:0.737～0.863)on the validation dataset,outperformed the other models.SHAP a-nalysis identified the Oxford Acute Severity of Illness score(OASIS),the Logistic Organ Dysfunction System(LODS)score,temperature,and partial pressure of oxygen as significant predictors of ARDS.Conclusions The XGBoost model demonstrates significant potential for predicting ARDS in patients with AP with concurrent sepsis,highlighting the importance of specific clinical predictors.This study may help clinicians identify high-risk patients and optimize treatment strategies.