FibroAgent: An Agentic AI Tool for Liver Fibrosis Screening and Clinical Decision Support in Metabolic Dysfunction-Associated Steatotic Liver Disease (MASLD)

Background Metabolic dysfunction-associated steatotic liver disease (MASLD) affects approximately one-quarter of adults worldwide, yet liver fibrosis remains markedly under-recognized in primary care. Existing screening approaches using static risk calculators yield single numeric outputs without explanation, actionable recommendations, or supporting documentation, limiting their adoption and clinical impact. Methods We developed FibroAgent, a conversational agentic AI framework that integrates a validated machine-learning XGBoost-based prediction model, patient-specific explainability via SHapley Additive exPlanations (SHAP), risk-stratified clinical pathways, an interactive educational knowledge base, and structured documentation within a single autonomous agent. FibroAgent was designed following established agentic AI principles: proactiveness, autonomy, reactivity, and transparency, and requires only seven routinely available clinical parameters (age, glycated hemoglobin (HbA1c), alanine aminotransferase (ALT), aspartate aminotransferase (AST), platelet count, body mass index (BMI), and glomerular filtration rate (GFR)). We demonstrated its capabilities across low-risk, intermediate-risk, and high-risk patient scenarios, batch screening of a six-patient cohort, and error-handling validation. Results FibroAgent correctly stratified patients with advanced liver fibrosis into rule-out (probability <=0.20), indeterminate (0.20-0.69), and rule-in (>=0.70) categories with corresponding risk-appropriate clinical recommendations. SHAP-based explanations identified platelet count, AST, and age as the top-three risk contributors in high-risk patients, providing a clinically interpretable rationale for each prediction. The agent demonstrated robust error handling, maintaining analytical continuity despite simulated tool failures. Conclusions FibroAgent represents a novel and promising approach to transitioning from static fibrosis calculators to interactive, explainable, and autonomous clinical decision support. By unifying prediction, explanation, recommendation, education, and documentation within a single conversational agent, FibroAgent offers a scalable, imaging-independent triage solution for primary care and resource-limited settings.