Prognostic Assessment of Sepsis-Induced Acute Respiratory Distress Syndrome Using an Early Warning Model

Background & Objective: Sepsis-induced acute respiratory distress syndrome (ARDS) is a critical condition with high mortality, yet effective tools for its early prediction are still lacking. This study aimed to identify early risk factors, develop a machine learning-based early warning model for sepsis-induced ARDS, and evaluate the prognostic value of the model-derived risk score for 28-day mortality. Methods: In this retrospective study of 188 sepsis patients, three models—logistic regression, random forest (RF), and support vector machine—were developed using variables selected via univariate and least absolute shrinkage and selection operator (LASSO) regression. Model performance was compared using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. The optimal model generated an “ARDS Risk Score,” which was evaluated for its association with 28-day mortality using Cox regression, restricted cubic splines, and Kaplan-Meier analysis. Results: The RF model showed the best predictive performance (AUC: 0.941, 95% confidence interval [CI]: 0.910–0.972), with procalcitonin, oxygenation index, lactate, and pulmonary infection as key predictors. The derived risk score was independently associated with 28-day mortality (hazard ratio [HR]: 4.635, 95% CI: 1.277–16.824, P = 0.02) and demonstrated a significant non-linear relationship with mortality risk. High-risk patients (defined by the median score) had significantly lower survival than low-risk patients. Conclusion: The RF model effectively predicts sepsis-induced ARDS. Its risk score serves as both an early warning tool and an independent prognostic predictor of 28-day mortality.