Abstract
BACKGROUND: To develop and externally validate a machine learning-based model for predicting the risk of acute respiratory distress syndrome (ARDS) in patients with intra-abdominal sepsis. METHODS: Data were obtained from the MIMIC-IV and the eICU-CRD database, including patients diagnosed with intra-abdominal sepsis. ARDS occurrence during intensive care unit (ICU) stay was defined as the primary outcome. Feature selection was performed using a combination of the Boruta algorithm, LASSO regression, and logistic regression. Ten base machine learning algorithms were trained and integrated into a stacked ensemble model. Model performance was systematically evaluated, and interpretability was assessed using SHapley Additive exPlanations (SHAP). External validation was conducted in an independent cohort of patients with intra-abdominal sepsis admitted to the First Affiliated Hospital of Xinjiang Medical University between 2016 and 2024. A web-based risk prediction calculator was subsequently developed to facilitate clinical decision support. RESULTS: Among 1,120 patients included from the MIMIC-IV and eICU-CRD databases, 554 (49.46%) developed ARDS during their ICU stay. Fourteen predictors were retained, including mechanical ventilation, use of vasoactive agents, history of chronic pulmonary disease, Sequential Organ Failure Assessment (SOFA) score, Glasgow Coma Scale (GCS) score, key vital signs, and routine laboratory indicators. The stacking model achieved areas under the receiver operating characteristic curve (AUC) of 0.811 in the development cohort, 0.794 in the internal validation cohort, and 0.756 in the external validation cohort. SHAP analysis identified mechanical ventilation as the most influential predictor, while early vasoactive agents use was associated with a reduced ARDS risk. CONCLUSION: A stacked ensemble model for predicting ARDS risk in patients with intra-abdominal sepsis demonstrated robust performance, stability, and interpretability. This model provides a practical tool for early risk stratification and informed clinical decision-making.