DYNAMITE: Integrating Archetypal Analysis and Process Mining for Interpretable Disease Progression Modelling.

Area of Science:

• Computational biology
• Data science in healthcare
• Biostatistics

Background:

• Longitudinal clinical datasets offer rich information on disease progression.
• Modeling complex disease trajectories requires advanced analytical methods.
• Existing methods may lack interpretability or scalability for diverse patient data.

Purpose of the Study:

• To introduce DYNAMITE (DYNamic Archetypal analysis for MIning disease TrajEctories), a novel methodology for modeling disease progression.
• To leverage archetypal analysis and process mining for extracting and visualizing patient clinical trajectories.
• To demonstrate the utility of DYNAMITE in identifying disease states and progression patterns in patient cohorts.

Main Methods:

• Archetypal analysis is applied to longitudinal data to identify representative disease states (archetypes).
• Patient data is mapped to identified archetypes, creating an event log of disease state progression.
• Process mining visualizes archetype sequences, enabling the extraction of individual and population-level clinical trajectories.

Main Results:

• DYNAMITE successfully modeled disease progression in amyotrophic lateral sclerosis (ALS) patients using the ALSFRS-R questionnaire.
• The method identified six distinct archetypes representing different impairment types and severity without prior assumptions.
• Generated clinical trajectories were consistent with known ALS prognosis, validating the approach.

Conclusions:

• DYNAMITE provides a highly interpretable framework for analyzing complex disease trajectories from longitudinal data.
• The methodology is suitable for healthcare settings where explainability is critical.
• DYNAMITE enables comprehensive analysis of clinical pathways at both individual and population levels.