Validation of a Machine Learning Algorithm, EVendo, for Predicting Esophageal Varices in Hepatocellular Carcinoma

Area of Science:

• Hepatology
• Gastroenterology
• Oncology

Background:

• Standard HCC treatment (atezolizumab/bevacizumab) increases bleeding risk, necessitating esophageal varices (EV) screening via EGD.
• Pre-treatment EGD can delay care and incur costs. The EVendo score was developed to noninvasively identify low-risk patients.

Purpose of the Study:

• To validate the EVendo score's accuracy in predicting the presence of EV needing treatment (VNT) in HCC patients.

Main Methods:

• Retrospective multicenter cohort study of 112 HCC patients with 117 EGDs (9/2004-12/2021).
• Clinical parameters for EVendo score calculation were collected at EGD.
• EVendo predictions were compared to EGD findings for VNT detection.

Main Results:

• 33.3% of patients had VNT.
• The EVendo score demonstrated high accuracy with 97.4% sensitivity and 96.9% negative predictive value for VNT.

Conclusions:

• The EVendo score effectively rules out VNT in HCC patients.
• Applying the EVendo score could safely defer approximately 30% of EGDs, potentially reducing costs and treatment delays.
• Further validation is needed, but the EVendo score shows promise for optimizing HCC patient management.