AI-Enabled ECG for Detecting Undiagnosed Liver Cirrhosis: The DULCE Trial

Exploring AI-Enabled ECG for Early Detection of Advanced Chronic Liver Disease (ACLD)

An estimated 1–2% of the general population has undiagnosed advanced chronic liver disease (ACLD), emphasizing the need for innovative, non-invasive, and accessible screening tools in primary care.

Artificial intelligence (AI)-enabled models, including those utilizing electrocardiogram (ECG)-based convolutional neural networks, show potential but require validation in clinical trials.

This pragmatic clinical trial assesses the effectiveness of an AI-driven ECG model for early detection of ACLD in primary care settings.

AI-Driven ECG Screening for ACLD: Trial Overview

Study Design: A cluster-randomized trial involving 98 primary care teams from 45 clinics and hospitals.

Intervention: Notifications of DULCE-positive ECGs indicating higher ACLD risk were provided to 123 clinicians.

Control: Usual care was provided to 122 clinicians.

Exclusion Criteria: Patients with a prior ACLD diagnosis were excluded from the study.

Primary Endpoint: The occurrence of a new ACLD diagnosis within 180 days, assessed using non-invasive liver disease assessment (NILDA).

Analysis: Incidence differences were evaluated using a mixed-effects logit model.

DULCE Trial Results

Study Summary:

A total of 15,596 adults underwent 12-lead ECGs during routine care and met inclusion criteria (8,034 intervention; 7,562 control). Baseline demographics were similar, with 53.6% women, mean age 63.4 (SD 18.3), mean BMI 35.3 (SD 12.4), and 37% with diabetes.

ACLD Diagnosis:

Intervention group: 0.6% vs. control: 0.3% (OR 2.27; 95% CI 1.27–4.07; p = .006).

High-risk patients (positive DULCE score): 3.0% vs. 0.8% (OR 4.53; 95% CI 1.86–11.03; p < .001).

Liver Fibrosis Detection:

Overall cohort: 1.3% vs. 0.4% (OR 3.51; 95% CI 1.93–6.39; p < .001).

High-risk patients: 6.3% vs. 0.8% (OR 8.82; 95% CI 3.59–21.68; p < .001).

Procedure Completion:

NILDA or liver biopsy completion: 2.9% in the intervention group vs. 1.3% in the control group (p < .001).

Key Insight

This pragmatic trial demonstrates that a machine learning algorithm utilizing 12-lead ECG has the potential to facilitate early diagnosis of advanced chronic liver disease (ACLD) in routine primary care settings.

By - Eeshan Aggarwal

Reference: Hepatology. Volume 80, Issue S1. Abstract Supplement for The Liver Meeting by the American Association for the Study of Liver Diseases (AASLD), November 15-19, 2024, San Diego, CA.