Early prediction of the severity of acute pancreatitis using radiologic and clinical scoring systems with classification tree analysis

OBJECTIVE. The objective of our study was to develop a decision tree model for the early prediction of the severity of acute pancreatitis (AP) using clinical and radiologic scoring systems. MATERIALS AND METHODS. For this retrospective study, 192 patients with AP who underwent CT 72 hours or less after symptom onset were divided into two cohorts: A training cohort (n = 115) and a validation cohort (n = 77). Univariate analysis was performed to identify significant parameters for the prediction of severe AP in the training cohort. For early prediction of disease severity, a classification tree analysis (CTA) model was constructed using significant scoring systems shown by univariate analysis. To assess the diagnostic performance of the model, we compared the area under the ROC curve (AUC) with each selected single parameter. We also evaluated the diagnostic performance in the validation cohort. RESULTS. The Acute Physiology and Chronic Health Evaluation (APACHE)-II score, bedside index for severity in acute pancreatitis (BISAP) score, extrapancreatic inflammation on CT (EPIC) score, and Balthazar grade were included in the CTA model. In the training cohort, our CTA model showed a trend of a higher AUC (0.853) than the AUC of each single parameter (APACHE-II score, 0.835; BISAP score, 0.842; EPIC score, 0.739; Balthazar grade, 0.700) (all, p > 0.0125) while achieving specificity (100%) higher than and accuracy (94.8%) comparable to each single parameter (both, p < 0.0125). In the validation cohort, the CTA model achieved diagnostic performance similar to the training cohort with an AUC of 0.833. CONCLUSION. Our CTA model consisted of clinical (i.e., APACHE-II and BISAP scores) and radiologic (i.e., Balthazar grade and EPIC score) scoring systems and may be useful for the early prediction of the severity of AP and identification of high-risk patients who require close surveillance.