Stage-Aware Event-Based Modeling (SA-EBM) for Disease Progression.

Area of Science:

• Biomedical informatics
• Computational biology
• Disease modeling

Background:

• Disease progression impacts cognitive and biological factors.
• Event-Based Models (EBMs) use probabilistic models to discover disease progression for early diagnosis.
• Prior EBMs lack dynamic biomarker distribution estimation and explicit disease stage modeling.

Purpose of the Study:

• To improve Event-Based Models (EBMs) with dynamic biomarker distribution estimation and explicit disease stage modeling.
• To develop a novel stage-aware EBM (SA-EBM) for more accurate disease progression analysis.
• To evaluate the performance of SA-EBM against existing EBM approaches.

Main Methods:

• Developed a stage-aware Event-Based Model (SA-EBM).
• Incorporated dynamic estimation of healthy and pathological biomarker distributions.
• Explicitly modeled disease stage distribution.
• Tested SA-EBM on 9,000 synthetic datasets and real-world ADNI data.

Main Results:

• The stage-aware EBM (SA-EBM) significantly outperformed prior methods like GMM EBM, KDE EBM, and Discriminative EBM.
• SA-EBM demonstrated superior accuracy in recovering the order of disease events.
• SA-EBM showed improved performance in assigning individual disease stages.

Conclusions:

• The novel stage-aware EBM (SA-EBM) offers significant improvements over existing methods for disease progression modeling.
• SA-EBM enhances the potential for earlier disease diagnosis and more effective clinical interventions.
• The SA-EBM package is available for use (pip install pysaebm) with source code provided.