Objective: To develop a multimodal deep learning-radiomics-clinical (DRC) model based on abdominal dual-phase contrast-enhanced CT (CECT) for the early prediction of acute pancreatitis-associated acute lung injury/acute respiratory distress syndrome (AP-ALI/ARDS). Methods: 320 AP patients were enrolled and randomly divided into a training set (n=189) and an internal test set (n=81) in a 7∶3 ratio, with an additional external test set (n=50). Clinical data and CECT images acquired within 48 hours after admission were collected. Independent clinical risk factors were selected using Logistic regression. The pancreatic parenchyma was manually delineated on both arterial and portal venous phase images, from which radiomics and deep learning features were subsequently extracted and screened. The random forest algorithm was used to construct the clinical model, radiomics model, deep learning model, and DRC model, respectively. Model performance was evaluated and compared using receiver operating characteristic (ROC) curves, the DeLong test, calibration curves, and decision curve analysis (DCA). Results: Among the 320 AP patients, 109 (34.06%) developed AP-ALI/ARDS. AP severity and blood glucose level were identified as independent risk factors. 6 radiomics features and 19 deep learning features were ultimately selected. The DRC model demonstrated superior predictive performance, achieving areas under the ROC curve (AUCs) of 0.973, 0.934, and 0.920 in the training, internal test, and external test sets, respectively. The DeLong test indicated that the AUC of the DRC model was significantly higher than those of the clinical and radiomics models (P<0.05). Calibration curves demonstrated that the DRC model exhibited the best performance, and DCA showed that it provided a greater overall net benefit across most threshold probabilities compared to other models. Conclusion: The DRC model exhibits excellent performance for the early prediction of AP-ALI/ARDS and shows potential as decision-support tool to guide clinical intervention.