Predicting 6-week mortality after acute variceal bleeding: role of Classification and Regression Tree analysis.

Background: Available prognostic scores for mortality after acute variceal bleeding are mainly based on logistic regression analysis but may have some limitations that can restrict their clinical value.

Aims: To assess the efficacy of a novel prognostic approach based on Classification and Regression Tree -CART- analysis to common easy-to-use models (MELD and Child-Pugh) for predicting 6-week mortality in patients with variceal bleeding.

Methods: Sixty consecutive cirrhotic patients with acute variceal bleeding. CART analysis, MELD and Child-Pugh scores were performed to assess 6-week mortality. Receiver operating characteristic (ROC) curves were constructed to evaluate the predictive performance of the models.

Results: Six-week rebleeding and mortality were 30% and 22%, respectively. Child-Pugh and MELD scores were clinically relevant for predicting 6 weeks mortality. CART analysis provided a simple algorithm based on just three bedside-available variables (albumin, bilirubin and in-hospital rebleeding), allowing accurate discrimination of two distinct prognostic subgroups with 3% and 80% mortality rates. All MELD, Child-Pugh and CART models showed excellent and comparable predictive accuracy, with areas under the ROC curves (AUROC) of 0.88, 0.84 and 0.91, respectively.

Conclusions: A simple CART algorithm combining albumin, bilirubin and in-hospital rebleeding allows an accurate predictive assessment of 6-week mortality after acute variceal bleeding.