Integrating temporal association rules into intelligent prediction system for metabolic dysfunction-associated fatty liver disease

Abstract

Healthcare big data provides trajectory data on chronic disease onset, progression, and outcomes, essential for understanding metabolic dysfunction-associated fatty liver disease (MAFLD) patient health dynamics. However, constructing explainable predictive models for MAFLD using longitudinal healthcare big data remains challenging due to its complexity. While several high-performance machine learning models have shown promise, their “black box” nature limits interpretability and trust among clinical healthcare professionals. Most studies also rely on cross-sectional data, which lacks the depth of longitudinal data, hindering accurate health status tracking. This paper proposes an intelligent MAFLD prediction system integrating temporal association rules (TARs) through a “human-in-the-loop” approach. By analyzing TARs that capture disease dynamics, the system incorporates high-quality domain knowledge into its predictive model. To enhance explainability, we use the SHapley Additive exPlanations framework alongside clinically significant TARs. The system’s effectiveness was validated on real-world data, showing improved MAFLD outcome prediction. Sensitivity analysis identified optimal TARs and robust model configurations. Finally, the online-deployed explainable prototype system demonstrates potential to boost trust and adoption among clinical healthcare professionals. Additionally, the system’s effectiveness and their willingness to use it were further evaluated through the “human-on-the-loop” method. These findings suggest the system could serve as a valuable tool for clinical applications and advance information systems design.