Improving Liver Fibrosis Diagnosis in Primary Care Using FibroX AI

Official title

Validation of an AI Tool for Improving MASLD Advanced Liver Fibrosis Diagnosis in Primary Care: A Provider-Level Crossover Randomized Controlled Trial Pilot

Overview

The goal of this clinical trial is to learn whether an artificial intelligence (AI) tool called FibroX can help primary care providers better diagnose significant liver fibrosis (≥F2) and clinically significant portal hypertension in adults with metabolic dysfunction-associated steatotic liver disease (MASLD). The main questions it aims to answer are: * Can FibroX improve the accuracy of diagnosing significant liver fibrosis (≥F2) and clinically significant portal hypertension compared to usual care? * Is FibroX easy to use and acceptable to primary care providers in simulated clinical settings? * Do providers trust FibroX as a decision-support tool? Researchers will compare FibroX-assisted care to usual care to see if FibroX improves diagnostic accuracy, provider trust, and supports better decision-making. Participants will: * Be primary care providers (MDs, DOs, NPs, PAs) from diverse clinics * Review simulated patient cases with MASLD risk factors * Use either usual care tools (standard labs and optional FIB-4 calculator) or FibroX (AI-generated risk score, triage band, and explainability panel) * Make diagnostic and referral decisions for each case * Complete surveys on usability, trust in AI, confidence, and cognitive workload This study will help determine whether FibroX can be integrated into real-world primary care workflows to support earlier and more accurate detection of liver fibrosis and portal hypertension, potentially reducing missed diagnoses, unnecessary referrals, and improving patient outcomes.

Detailed description

This study is a 12-month pilot clinical trial designed to evaluate the feasibility, usability, provider trust, and preliminary effectiveness of FibroX, an explainable artificial intelligence (AI) tool developed to improve the diagnosis of significant liver fibrosis (≥F2) and clinically significant portal hypertension in adults with metabolic dysfunction-associated steatotic liver disease (MASLD). MASLD is a common and progressive liver condition that can lead to cirrhosis, liver failure, and increased cardiovascular risk. Early detection of these conditions is critical because current guidelines recommend initiating therapy (e.g., resmetirom or semaglutide for ≥F2 fibrosis and beta-blockers for portal hypertension). However, existing tools like FIB-4 often lack accuracy and usability in routine primary care.

FibroX addresses these limitations by using routinely available clinical data-such as age, liver enzymes, platelet count, BMI, and kidney function-to estimate the probability of significant fibrosis and portal hypertension. It provides a triage band (rule-out, indeterminate, rule-in) and a one-line explanation of which clinical factors most influenced the prediction. This transparency is achieved using Shapley Additive Explanations (SHAP), which helps clinicians understand how the AI reached its conclusion.

In retrospective studies, FibroX demonstrated superior diagnostic performance compared to FIB-4 (AUROC 0.97 vs. 0.62) and was associated with long-term mortality risk, suggesting prognostic value beyond diagnostic utility.

This pilot trial will simulate real-world primary care workflows to test whether FibroX can be effectively used by clinicians. The study will recruit 30-40 primary care providers (MDs, DOs, NPs, PAs) from 4-6 diverse clinics. Each provider will participate in two simulation periods, each involving 16 synthetic or de-identified patient cases reflecting adults with MASLD risk factors. Ground truth for fibrosis stage and portal hypertension will be determined by biopsy or expert consensus using Vibration-Controlled Transient Elastography (VCTE) and guideline-based criteria.

Providers will be randomly assigned to review cases in one of two sequences:

* FibroX-Enabled Care: Providers will receive FibroX's risk probability, triage band, and explainability panel. * Usual Care: Providers will use standard labs and vitals, with optional access to the FIB-4 calculator.

After a one-week washout period, providers will switch to the other condition. For each case, providers will make a management decision (e.g., no action, order VCTE, refer to hepatology), record their confidence level, and complete surveys on usability, trust in AI, and cognitive workload.

Primary Outcomes

* Feasibility: Recruitment rate ≥70%, completion rate ≥85%, median decision time ≤3.5 minutes. * Usability and Acceptability: System Usability Scale (SUS) score ≥70. * Provider Trust: AI-Trust Scale score ≥6. * Effectiveness: Within-provider diagnostic accuracy for significant fibrosis (≥F2) and clinically significant portal hypertension.

Secondary Outcomes

* Appropriate referral rates * Net reclassification improvement (NRI) * Calibration metrics (intercept, slope) * Provider confidence and cognitive load (NASA-TLX) * Intended downstream testing burden * Adoption and fidelity to triage recommendations * Override rates and reasons * Fairness analysis across subgroups (age, sex, BMI, race/ethnicity)

All provider actions and decision times will be automatically logged. Post-period surveys and qualitative debriefs will explore barriers and facilitators to using FibroX.

Study Significance This pilot study will generate critical data to support a future multi-center trial and potential integration of FibroX into electronic health records. If successful, FibroX could enable scalable, guideline-concordant screening for significant liver fibrosis and portal hypertension in primary care, reducing missed diagnoses and unnecessary referrals. This aligns with national priorities for precision medicine and responsible AI implementation in healthcare.

Interventions

• Device FibroX
  FibroX is an explainable artificial intelligence (AI) tool designed to assist primary care providers in diagnosing significant liver fibrosis (≥F2) and clinically significant portal hypertension in patients with metabolic dysfunction-associated steatotic liver disease (MASLD). It uses routinely available clinical data (e.g., age, AST, ALT, platelets, BMI, HbA1c, creatinine) to generate a risk probability score, a triage band (rule-out, indeterminate, rule-in), and a one-line explainability panel
• Other Usual Care
  In the usual care condition, primary care providers assess simulated patient cases using standard clinical tools available in routine practice. These include laboratory results, vital signs, problem lists, medications, and prior imaging. Providers may optionally use the FIB-4 calculator to estimate liver fibrosis risk. No AI decision support is provided. This intervention serves as the comparator to evaluate whether FibroX improves diagnostic accuracy for significant liver fibrosis (≥F2) and cli

Primary outcome measures

• Diagnostic Accuracy for Significant Liver Fibrosis (≥F2) and Clinically Significant Portal Hypertension Using FibroX Compared to Usual Care [Time frame: Immediately after each simulation period, up to 24 weeks]
• System Usability Scale (SUS) Score for FibroX Integration [Time frame: Immediately after each simulation period, up to 24 weeks]
• Provider Trust in AI Tool (FibroX) [Time frame: Immediately after the FibroX-enabled simulation period, up to 24 weeks]
• Median Decision Time per Case [Time frame: Immediately after each simulation period, up to 24 weeks]

Secondary outcome measures (9)

• Appropriate Referral Rate [Time frame: Immediately after each simulation period, up to 24 weeks]
• Net Reclassification Improvement (NRI) [Time frame: Immediately after each simulation period, up to 24 weeks]
• Calibration of Risk Predictions [Time frame: Immediately after each simulation period, up to 24 weeks]
• Provider Confidence in Decision-Making [Time frame: Immediately after each simulation period, up to 24 weeks]
• Cognitive Load During Case Review [Time frame: Immediately after each simulation period, up to 24 weeks]
• Intended Downstream Testing Burden [Time frame: Immediately after each simulation period, up to 24 weeks]
• Adoption and Fidelity to Triage Recommendations [Time frame: Immediately after each simulation period, up to 24 weeks]
• Override Rate and Reasons [Time frame: Immediately after each simulation period, up to 24 weeks]
• Fairness Analysis Across Subgroups [Time frame: Immediately after each simulation period, up to 24 weeks]

Eligibility criteria

Inclusion criteria

• Licensed primary care providers (MD, DO, NP, or PA)
• Currently practicing in adult primary care (≥0.5 Full-Time Equivalent)
• Affiliated with one of the participating clinics (academic, community, or Federally Qualified Health Center)
• Willing and able to participate in simulated electronic health record (EHR)-based case reviews
• Able to provide informed consent

Exclusion criteria

• Providers not actively practicing in adult primary care
• Providers with less than 0.5 FTE in clinical practice
• Prior involvement in the development or validation of the FibroX tool
• Inability to complete both simulation periods due to scheduling or other constraints

Criteria are shown verbatim from the registry (in English). Final eligibility is always assessed by the study center.

Healthy volunteers: No

Study design

Allocation

Randomized

Model

Crossover

Masking

Open label

Primary purpose

Diagnostic

Study locations

Center list to be confirmed — check the primary protocol.

Publications

• Njei, B., et al., FIBROX: an explainable AI model for accurate prediction of advanced liver fibrosis and cardiovascular mortality in MASLD. Gastroenterology, 2024. 169(1): p. S-131-S-132.
• Njei B, Osta E, Njei N, Al-Ajlouni YA, Lim JK. An explainable machine learning model for prediction of high-risk nonalcoholic steatohepatitis. Sci Rep. 2024 Apr 13;14(1):8589. doi: 10.1038/s41598-024-59183-4. PMID 38615137
• Ratziu V, Charlotte F, Heurtier A, Gombert S, Giral P, Bruckert E, Grimaldi A, Capron F, Poynard T; LIDO Study Group. Sampling variability of liver biopsy in nonalcoholic fatty liver disease. Gastroenterology. 2005 Jun;128(7):1898-906. doi: 10.1053/j.gastro.2005.03.084. PMID 15940625
• Decharatanachart P, Chaiteerakij R, Tiyarattanachai T, Treeprasertsuk S. Application of artificial intelligence in non-alcoholic fatty liver disease and liver fibrosis: a systematic review and meta-analysis. Ther Adv Gastroenterol. 2021 Dec 21;14:17562848211062807. doi: 10.1177/17562848211062807. eCollection 2021. PMID 34987607
• Meng F, Zheng Y, Zhang Q, Mu X, Xu X, Zhang H, Ding L. Noninvasive evaluation of liver fibrosis using real-time tissue elastography and transient elastography (FibroScan). J Ultrasound Med. 2015 Mar;34(3):403-10. doi: 10.7863/ultra.34.3.403. PMID 25715361
• Boursier J, de Ledinghen V, Zarski JP, Fouchard-Hubert I, Gallois Y, Oberti F, Cales P; multicentric groups from SNIFF 32, VINDIAG 7, and ANRS/HC/EP23 FIBROSTAR studies. Comparison of eight diagnostic algorithms for liver fibrosis in hepatitis C: new algorithms are more precise and entirely noninvasive. Hepatology. 2012 Jan;55(1):58-67. doi: 10.1002/hep.24654. PMID 21898504
• Wong VW, Vergniol J, Wong GL, Foucher J, Chan HL, Le Bail B, Choi PC, Kowo M, Chan AW, Merrouche W, Sung JJ, de Ledinghen V. Diagnosis of fibrosis and cirrhosis using liver stiffness measurement in nonalcoholic fatty liver disease. Hepatology. 2010 Feb;51(2):454-62. doi: 10.1002/hep.23312. PMID 20101745
• Yoon JH, Lee JM, Joo I, Lee ES, Sohn JY, Jang SK, Lee KB, Han JK, Choi BI. Hepatic fibrosis: prospective comparison of MR elastography and US shear-wave elastography for evaluation. Radiology. 2014 Dec;273(3):772-82. doi: 10.1148/radiol.14132000. Epub 2014 Jul 7. PMID 25007047

Identifiers

NCT: NCT07305324 · 2000027433

Primary sources (government registries)

• ClinicalTrials.gov (US NIH)

View this study on ClinicalTrials.gov ↗