Development and external validation of models to predict acute respiratory distress syndrome related to severe acute pancreatitis

BACKGROUND Acute respiratory distress syndrome (ARDS) is a major cause of death in patients with severe acute pancreatitis (SAP). Although a series of prediction models have been developed for early identification of such patients, the majority are complicated or lack validation. A simpler and more credible model is required for clinical practice. AIM To develop and validate a predictive model for SAP related ARDS. METHODS Patients diagnosed with AP from four hospitals located at different regions of China were retrospectively grouped into derivation and validation cohorts. Statistically significant variables were identified using the least absolute shrinkage and selection operator regression method. Predictive models with nomograms were further built using multiple logistic regression analysis with these picked predictors. The discriminatory power of new models was compared with some common models. The performance of calibration ability and clinical utility of the predictive models were evaluated. RESULTS Out of 597 patients with AP, 139 were diagnosed with SAP (80 in derivation cohort and 59 in validation cohort) and 99 with ARDS (62 in derivation cohort and 37 in validation cohort). Four identical variables were identified as independent risk factors for both SAP and ARDS: heart rate [odds ratio (OR) = 1.05; 95%CI: 1.04-1.07; P < 0.001; OR = 1.05, 95%CI: 1.03-1.07, P < 0.001], respiratory rate (OR = 1.08, 95%CI: 1.0-1.17, P = 0.047; OR = 1.10, 95%CI: 1.02-1.19, P = 0.014), serum calcium concentration (OR = 0.26, 95%CI: 0.09-0.73, P = 0.011; OR = 0.17, 95%CI: 0.06-0.48, P = 0.001) and blood urea nitrogen (OR = 1.15, 95%CI: 1.09-1.23, P < 0.001; OR = 1.12, 95%CI: 1.05-1.19, P < 0.001). The area under receiver operating characteristic curve was 0.879 (95%CI: 0.830-0.928) and 0.898 (95%CI: 0.848-0.949) for SAP prediction in derivation and validation cohorts, respectively. This value was 0.892 (95%CI: 0.843-0.941) and 0.833 (95%CI: 0.754-0.912) for ARDS prediction, respectively. The discriminatory power of our models was improved compared with that of other widely used models and the calibration ability and clinical utility of the prediction models performed adequately. CONCLUSION The present study constructed and validated a simple and accurate predictive model for SAP-related ARDS in patients with AP.