Machine learning model surpasses clinical risk scores for GI bleeding

In a comparative analysis of an electronic health record-based machine learning model and established clinical risk scores for gastrointestinal bleeding, researchers found the model significantly outperformed the Glasgow-Blatchford Score and Oakland Score. With an area under the receiver-operating-characteristic curve (AUC) of 0.92 versus 0.89 (p<0.001), the machine learning approach identified a higher percentage of very-low-risk patients eligible for emergency department discharge, detecting 37.9% compared to 18.5% and 11.7%, respectively.

• Why it matters: Enhancing risk assessment improves patient safety in emergency departments.

Journal Article by Shung DL, Chan CE (…) Laine L et 11 al. in Gastroenterology

read the whole article in Gastroenterology

open it in PubMed