Improving Risk Stratification for Upper Gastrointestinal Bleeding

A machine learning model outperformed standard risk scores in predicting which patients with UGIB could be safely managed as outpatients.

Risk scores in nonvariceal upper gastrointestinal bleeding (UGIB) have been developed to predict both high and low risk for outcomes such as rebleeding and mortality; for low-risk cases, no intervention is needed and outpatient management is safe. The three major pre-endoscopic risk scores are the Glasgow-Blatchford score (GBS), AIMS65 score, and pre-endoscopic Rockall score. A GBS of ≤1 has been recommended as predictive of low-risk cases.

Now, researchers have developed and validated a machine learning model to predict a composite outcome (hemostatic intervention, need for transfusion, or mortality) in patients with UGIB. They compared the model's accuracy with that of existing risk scores for identifying low-risk patients.

The machine lear…

Reviewing Author

John R. Saltzman, MD

Disclosures

Consultant/Advisory BoardIterative Scopes

Editorial BoardsGastrointestinal Endoscopy; UpToDate

Leadership Positions in Professional SocietiesAmerican College of Gastroenterology (Trustee)

OtherVice President of Medical Affairs, 1Globe Health Institute