Predicting early mortality after acute variceal hemorrhage based on classification and regression tree analysis

Area of Science:

• Hepatology
• Medical Statistics
• Clinical Prognostication

Background:

• Existing prognostic models for acute variceal hemorrhage (AVH) mortality have clinical limitations.
• A novel prognostic approach using classification and regression tree (CART) analysis was evaluated.

Purpose of the Study:

• To assess the performance of a CART-based prognostic model for predicting 6-week mortality in patients with AVH.
• To compare the CART model's accuracy against logistic regression (LR) and established scoring systems.

Main Methods:

• A cohort of 267 patients with AVH underwent logistic regression (LR) and CART analyses.
• Models were developed using a training set and validated on a separate test set.
• Receiver operating characteristic (ROC) curves were used to evaluate model performance.

Main Results:

• The CART model identified three distinct prognostic subgroups (low, intermediate, high risk) with varying mortality rates (8%, 17%, 50-73%).
• CART model accuracy (AUC 0.81-0.83) was comparable to the best LR model (AUC 0.84) and superior to Child-Pugh (AUC 0.75) and MELD (AUC 0.74) scores.
• The CART algorithm utilized only Child-Pugh score, creatinine level, and bacterial infection for prediction.

Conclusions:

• A simple CART algorithm effectively predicts 6-week mortality after acute variceal bleeding.
• This approach offers accurate prognostic assessment using readily available clinical variables.
• The CART model enhances risk stratification for patients with AVH, aiding clinical decision-making.