# A novel dual elastography-based model for screening high-risk varices in hepatitis B virus-related cirrhosis

## TL;DR

A new non-invasive model called DELU improves the screening of high-risk esophageal varices in patients with hepatitis B-related cirrhosis, reducing the need for endoscopy.

## Contribution

The DELU model, developed using dual elastography and machine learning, offers superior accuracy and endoscopy-sparing potential compared to existing non-invasive tests.

## Key findings

- DELU achieved AUROCs of 0.822 in training and 0.779 in validation, outperforming other non-invasive tests like RESIST and Baveno VI criteria.
- DELU spared endoscopies in 20.7% of the training and 11.1% of the validation cohorts while maintaining high accuracy across subgroups.
- The model's performance was robust regardless of HCC status, sex, or BMI, and it maintained high accuracy in patients on antiviral therapy.

## Abstract

Variceal bleeding carries high mortality and recurrence rates, underscoring the urgent need for effective non-invasive tests (NITs) to assess the severity of esophageal varices (EVs) and predict bleeding risk. This study aimed to develop and validate a robust model for safely excluding high-risk varices (HRVs) in patients with hepatitis B virus (HBV)-related cirrhosis and to compare its diagnostic performance and clinical utility with established NITs.

Consecutive patients with HBV-related cirrhosis were prospectively recruited from five centers and underwent dual elastography (dual-elasto) within 1 week of esophagogastroduodenoscopy (EGD). A training cohort from four centers was used to identify predictive variables for HRVs, which were analyzed using machine learning to develop the most reliable model. The final model was externally validated in an independent cohort from the fifth center. Diagnostic efficacy and clinical utility were compared across multiple NITs, including the Baveno VI criteria (B6C), expanded Baveno VI criteria (EB6C), platelet–spleen ratio (PSR), and RESIST-HCV criteria (RESIST).

Among 703 screened patients, 328 were enrolled (training cohort n = 184; validation cohort n = 144). Using logistic regression, we developed the DELU model incorporating six variables: platelet count (PLT), alanine aminotransferase (ALT), ascites, portal vein thrombosis (PVT), inverse difference moment (IDM), and liver stiffness (liver Vs). DELU achieved areas under the receiver operating characteristic curve (AUROCs) of 0.822 (training) and 0.779 (validation), outperforming RESIST (0.600 and 0.626) and B6C (0.569 and 0.575). EB6C and PSR were excluded because they exceeded the 5% missed-HRV threshold. DELU demonstrated the highest spared-EGD rates (20.7% training; 11.1% validation) and maintained robust performance across subgroups, including Child-Pugh B/C (AUROC 80.3%), Child-Pugh A (79.7%), antiviral therapy (ART)-treated (82.5%), and virologically suppressed patients (84.6%). Performance was not influenced by hepatocellular carcinoma (HCC) status, sex, or body mass index (BMI).

DELU is a reliable and safe non-invasive tool for excluding HRVs in HBV-related cirrhosis, providing superior diagnostic accuracy and significant endoscopy-sparing potential across diverse clinical settings. RESIST may serve as a safe alternative when elastography is unavailable.

[ClinicalTrials.gov], identifier [NCT04640350].