Using AI to Improve Risk Stratification, Disease Management in MASLD, With Basile Njei, MD, PhD, MPH

Use of a novel large language model-based tool, Fibro-GPT, may offer a promising approach to detecting advanced fibrosis and predicting hepatic decompensation in metabolic dysfunction-associated steatotic liver disease (MASLD).1

Data presented at the American College of Gastroenterology (ACG)’s 2025 Annual Scientific Meeting by Basile Njei, MD, PhD, MPH, an assistant professor at Yale School of Medicine, highlight Fibro-GPT’s superior performance, interpretability, and potential cost savings in the context of MASLD care.1

The most common cause of chronic liver disease, MASLD is projected to become the leading indication for liver transplant in the US. Findings from a decision analytical modeling study predict a steady increase in the prevalence of MASLD from 33.7%, or 86.3 million people, in 2020 to 41.4%, or 121.9 million people, by 2050.2

“In most of our clinics, the tool we have to predict advanced fibrosis is a simple clinical score called FIB-4,” Njei explained to HCPLive. “Unfortunately, it doesn't really do a good job in predicting liver disease, and a lot of people with advanced liver disease are either misclassified as not having advanced liver disease, or people without advanced liver disease are classified as having advanced liver disease, therefore getting tests that they do not need and liver biopsies that are not required.”

Recognizing recent advances in artificial intelligence tools like large language models and their potential to streamline MASLD care, Njei and colleagues developed Fibro-GPT using NHANES 2017–2020 data from adults with MASLD, including routinely collected variables like age, platelet count, HbA1c, eGFR, BMI, AST, and ALT. Inputs were formatted into standardized text prompts and analyzed via a secure Python interface, generating a numerical risk estimate and an explanatory summary. Fibrosis stage was determined using a 2-step method: blood-based (FIB-4) followed by vibration-controlled transient elastography (VCTE).1

Fibro-GPT was trained on 162 MASLD cases and externally validated on 105 biopsy-confirmed MASLD patients using liver histopathology as the reference standard. Predictive performance for 5-year incident hepatic decompensation was assessed in an additional 337 biopsy-confirmed MASLD patients.1

To further assess the tool’s economic impact, investigators modeled a US population-wide screening scenario comparing Fibro-GPT to FIB-4 as a first-line tool for detecting advanced fibrosis in MASLD.1

Results showed Fibro-GPT achieved an AUROC of 0.91 in the derivation cohort, with 84% sensitivity and 86% specificity, significantly outperforming FIB-4 (AUROC, 0.62; P <.001). Validation on biopsy-confirmed cases showed sustained performance (AUROC, 0.90 vs FIB-4 AUROC, 0.75; P < .001).1

For predicting 5-year incident hepatic decompensation, Fibro-GPT achieved AUROC 0.85, with 91% sensitivity and 67% specificity. Additional cost analysis revealed Fibro-GPT’s superior accuracy resulted in a 25% reduction in unnecessary VCTEs and an 18% reduction in liver biopsies compared to FIB-4, translating into estimated annual savings of $150-200 million in the US.1

Looking ahead at how Fibro-GPT may be implemented in clinical workflows, Njei emphasized that the tool does not require any new or additional testing not already being done in the clinic. As a result, he foresees it being widely utilized by a large number of clinicians.

“We are actually trying to develop what we call Fibro-GPT 2.0, and this would include using EKG data, so EKG notes, EKG wave forms, in addition to clinical laboratory tests to predict liver disease,” Njei added.

Editors’ note: Njei has no relevant disclosures.

References

1. Njei B, Njei N. Ilagan-Ying Y, et al. Fibro-GPT: A Large Language Model for Accurate Prediction of Advanced Fibrosis and Hepatic Decompensation in MASLD. Presented at the American College of Gastroenterology (ACG)’s 2025 Annual Scientific Meeting. Phoenix, Arizona. October 27-29, 2025.

2. Le P, Tatar M, Dasarathy S, et al. Estimated Burden of Metabolic Dysfunction–Associated Steatotic Liver Disease in US Adults, 2020 to 2050. JAMA Netw Open. doi:10.1001/jamanetworkopen.2024.54707