Gastroenterology

AI-enabled ECG screening in primary care doubles detection of advanced chronic liver disease

In 98 randomized primary care teams, access to ECG-based machine learning alerts doubled new diagnoses of advanced chronic liver disease (1.0% vs 0.5%). Among ECG-ML–positive patients, advanced disease was detected more often with alerts than usual care (4.4% vs 1.1%). Detection of any fibrosis also increased in the intervention arm, both overall and within ECG-ML–positive patients. Diagnostic yield remained below epidemiologic expectations, suggesting incomplete follow-through on AI alerts. These data support integrating ECG-ML risk flags with standardized liver workup pathways to improve case finding in routine practice. 1

RIPK1-driven necroptosis in cholangiocytes emerges as a disease-modifying target in PSC

Human PSC biopsies showed necroptosis localized mainly to cholangiocytes, while apoptosis predominated in hepatocytes and nonbiliary cells. In vitro, necroptosis inhibition or RIPK1 deletion protected cholangiocytes from TNF-mediated cytotoxicity. In a murine PSC model, pharmacologic RIPK1 inhibition reduced cholestatic injury, hepatic inflammation, and biliary fibrosis. Multiomic profiling showed treated livers shifted toward wild-type-like molecular signatures, consistent with broad disease modification. These findings position RIPK1 inhibition and necroptosis blockade as rational therapeutic strategies for PSC, pending human trials. 2

New multi-society framework standardizes diagnosis and staging of AATD-associated liver disease

The consensus defines AATD-associated liver disease as liver enzyme elevation and/or fibrosis (≥F2) in adults with AATD, especially PiZZ genotype. Liver stiffness by elastography is preferred for staging, with vibration-controlled transient elastography ≥8 kPa indicating clinically significant fibrosis. APRI and FIB-4 are endorsed to identify low risk for advanced disease, while acknowledging limited sensitivity. A tiered longitudinal algorithm emphasizes serial noninvasive testing and timely transplant referral for progressive fibrosis. For phase 3 trials, inclusion of patients with F2–F4 fibrosis and a primary endpoint of ≥1-stage fibrosis improvement is recommended. 3

Nine-protein plasma model predicts Crohn’s disease up to 16 years before diagnosis

Among 39,634 UK Biobank participants, 44 plasma proteins were associated with incident Crohn’s disease. A nine-protein machine-learning model achieved an AUC of 0.76 in a UK Biobank test cohort and 0.73 in EPIC-Norfolk. The model discriminated Crohn’s disease in a Southern China cohort with an AUC of 0.79, supporting geographic generalizability. Combining protein markers with clinical data modestly improved prediction (AUC 0.78) up to 16 years before diagnosis. High-risk individuals by the protein model were over four times more likely to develop Crohn’s disease, supporting risk-stratified surveillance strategies. 4