Clinical Use of Nomogram Based on Machine Learning for Diagnosis Prediction of Acute Respiratory Distress Syndrome in Patients With Acute Pancreatitis.

Background: This study focused on utilizing machine learning techniques to construct a predictive nomograph for early identification of acute pancreatitis (AP) patients at risk of developing acute respiratory distress syndrome (ARDS). Methods: We retrospectively analyzed 427 AP patients from Tongji Hospital (2010-2021) and externally validated the model using the MIMIC-IV database. Six machine learning algorithms were compared, with the support vector machine (SVM) selected for nomogram construction. Key predictors included age, sex, SOFA score, C-reactive protein (CRP), platelet count (PLT), total bilirubin (TBIL), and direct bilirubin (DBIL). Model performance was assessed via area under the curve (AUC), calibration curves, and decision curve analysis (DCA). Results: The SVM model exhibited the best performance among six machine-learning models assessed. Key predictors including age, sex, SOFA, CRP, PLT, TBIL, and DBIL levels were incorporated into the nomogram. The nomogram demonstrated good discriminatory ability and clinical applicability, with a C-index of 0.818 in the training cohort and 0.799 in the testing cohort. External validation using the MIMIC IV database further confirmed its accuracy, with a C-index of 0.759. Notably, calibration curves showed excellent agreement between predicted and observed outcomes, and DCA indicated a favorable net benefit, reinforcing the model's reliability. Conclusions: The prediction nomogram constructed based on the SVM model in this study can effectively predict the probability of AP complicated by ARDS.