AI-Enhanced ECGs Doubles Detection of Advanced chronic Liver Disease: Study Finds

Given that advanced CLD affects between 2% and 5% of the general population and often remains undiagnosed due to a lack of accessible screening, Douglas A. Simonetto and colleagues of the Mayo Clinic and other institutions conducted the pragmatic trial to determine if an ECG-based machine learning (ECG-ML) model could facilitate early case finding during routine medical visits.

Therefore, the pragmatic, cluster-randomized clinical trial involved 15,596 adult patients across 98 primary care teams, comprising 8,034 in the intervention group and 7,562 in the control arm, where 123 clinicians were given access to ECG-ML risk alerts while 122 provided standard care. The primary endpoint was the new diagnosis of advanced liver fibrosis within 180 days of a routine 12-lead ECG, while secondary outcomes tracked the detection of any hepatic fibrosis, excluding patients who did not meet standard clinical inclusion criteria for routine care.

Key Findings of the Study Include:

• Improved Disease Detection: The study showed that the intervention group experienced a twofold increase in the diagnosis of advanced chronic liver disease compared to the usual care cohort (1.0% versus 0.5%, P = 0.007).

• Targeted High-Risk Identification: Among those patients specifically flagged as positive by the machine learning algorithm, the research led to a significantly higher diagnostic rate of 4.4% compared to only 1.1% in the control group (OR 4.37, 95% CI 1.94–9.88).

• Secondary Fibrosis Yield: The trial data indicated that the intervention also tripled the detection of any stage of liver fibrosis across the entire cohort, identifying cases in 1.7% of patients versus 0.5% in those receiving standard care (OR 3.17, 95% CI 1.86–5.40).

• Enhanced Subgroup Performance: In the subgroup of patients with positive ECG-ML results, the investigation identified any fibrosis at a rate nearly eight times higher in the intervention arm at 8.4% compared to 1.1% in the control arm (OR 8.03, 95% CI 3.50–18.4).

• Effective Clinical Prompting: The analysis demonstrated that notifying clinicians of positive machine learning results was an effective way to trigger sequential liver disease assessments that might otherwise have been overlooked in routine primary care.

• The results suggest that integrating an ECG-based machine learning model into primary care workflows significantly improves the identification of advanced chronic liver disease, particularly among high-risk individuals, where the diagnostic yield reached 4.4% compared to 1.1% in traditional care settings.

The study concludes that clinicians can potentially utilize routine cardiac screening data to prompt earlier investigations into liver health, thereby addressing the current gap in early disease detection for asymptomatic patients.

While the study observed that overall diagnostic yields remained lower than epidemiological estimates due to varying levels of clinician adherence to the artificial intelligence alerts, further research into optimizing clinical integration could enhance the practical utility of these predictive models.

Reference

Simonetto, D. A., et al. (2026). Detection of undiagnosed liver cirrhosis via AI-enabled electrocardiogram: a pragmatic, cluster-randomized clinical trial. Nature Medicine, 32, 160–167.