Related Experiment Videos
An Explainable AI Framework for Continuous Monitoring, Risk Stratification, and Clinical Decision Support in Primary
Basile Njei1, Ulrick Sidney Kanmounye2
1Engelhardt School of Global Health and Bioethics, Euclid University, Avenue de France, Campus ENAM, Bangui, BP 157, Central African Republic, +236 21 61 59 2.
Background:
Primary biliary cholangitis (PBC) management remains limited by reliance on static biochemical markers, fragmented assessment of symptom burden, and inadequate noninvasive risk stratification for clinically significant portal hypertension (CSPH). Existing tools fail to integrate longitudinal laboratory trends, elastography, and patient-reported outcomes, and translation of risk assessment into guideline-concordant clinical action remains inconsistent.
Objective:
This study aims to develop, validate, and pilot an explainable artificial intelligence (AI) framework-AIm-PBC-for continuous disease monitoring, early prediction of CSPH complications, and delivery of guideline-based clinical decision support in patients with PBC.
Methods:
We will conduct a multiphase study combining retrospective cohort analysis, prospective observational data collection, and implementation evaluation. At least 600 adults with confirmed PBC will be enrolled across academic hepatology centers. The AI framework will integrate longitudinal biochemical markers, noninvasive elastography metrics, and high-frequency patient-reported outcomes to generate an interpretable disease activity index and predict CSPH-related complications using gradient-boosted models with Shapley additive explanations. Outputs will be deployed via a SMART-on-FHIR (Substitutable Medical Applications, Reusable Technologies on Fast Healthcare Interoperability Resources)-enabled electronic health record clinical decision support tool. Implementation will be evaluated using a randomized crossover simulation followed by a pragmatic pilot assessing usability, cognitive load, clinician adherence, and feasibility.
Results:
The primary outcomes include calibration and responsiveness of the disease activity index; discriminatory performance of the CSPH prediction model compared with established criteria; and effectiveness of the decision support tool measured via improvements in guideline-concordant care, usability scores, and clinician cognitive workload. Secondary outcomes include fairness metrics and workflow efficiency.
Conclusions:
This protocol outlines a scalable, explainable AI framework designed to bridge gaps between disease monitoring, risk prediction, and clinical action in PBC. If successful, AIm-PBC may enhance early complication detection, improve symptom management, and support equitable, evidence-based care delivery in routine hepatology practice.
