FastEBM: Fast, Scalable, and Uncertainty-Aware Event-Based Disease Progression Modeling

Area of Science:

• Computational Biology
• Biostatistics
• Machine Learning

Background:

• Event-based models (EBMs) infer biomarker alteration order in disease progression.
• Traditional EBMs face scalability and stability issues due to computational expense and independence assumptions.

Purpose of the Study:

• Introduce Fast Event-Based Model (FastEBM), a novel framework for scalable and robust disease progression inference.
• Reformulate disease progression as a subject-ordering problem on a diffusion manifold using Markov chains.

Main Methods:

• Developed FastEBM, a Markov-chain-based framework utilizing diffusion geometry.
• Quantified progression uncertainty via first-passage-time variability.
• Compared FastEBM against GMM-EBM, KDE-EBM, and DEBM using synthetic and real-world data.

Main Results:

• FastEBM demonstrated superior accuracy and runtime compared to existing EBM methods.
• Maintained event-order recovery in high-dimensional, low-subject settings and robust performance with correlated features.
• Successfully characterized biomarker progression in Alzheimer's disease using TAD-POLE and ADNI datasets, revealing lateralized trends.

Conclusions:

• Diffusion-based Markov geometry offers a scalable and robust alternative to conventional EBMs.
• FastEBM provides an effective tool for high-dimensional disease progression mapping and biomarker discovery.