Prediction of severe acute pancreatitis using classification and regression tree analysis.

Background: The available prognostic scoring systems for acute pancreatitis have limitations that restrict their clinical value.

Aims: To develop a decision model based on classification and regression tree (CART) analysis for the prediction of severe acute pancreatitis (SAP).

Methods: A total of 420 patients with acute pancreatitis were enrolled. Study participants were randomly assigned to the training sample and test sample in a 2:1 ratio. First, univariate analysis and logistic regression analysis were used to identify predictors associated with SAP in the training sample. Then, CART analysis was carried out to develop a simple tree model for the prediction of SAP. A receiver operating characteristic (ROC) curve was constructed in order to assess the performance of the model. The prediction model was then applied to the test sample.

Results: Four variables (systemic inflammatory response syndrome [SIRS], pleural effusion, serum calcium, and blood urea nitrogen [BUN]) were identified as important predictors of SAP by logistic regression analysis. A tree model (which consisted of pleural effusion, serum calcium, and BUN) that was developed by CART analysis was able to early identify among cohorts at high (79.03%) and low (7.80%) risk of developing SAP. The area under the ROC curve of the tree model was higher than that of the APACHE II score (0.84 vs. 0.68; P < 0.001). The predicted accuracy of the tree model was validated in the test sample with an area under the ROC curve of 0.86.

Conclusions: A decision tree model that consists of pleural effusion, serum calcium, and BUN may be useful for the prediction of SAP.