Decoding Parkinson's progression: A multi-modal SuStaIn ensemble approach validated on real-world PPMI data.

Area of Science:

• Computational biology and bioinformatics
• Neuroscience and neurology
• Biostatistics and data science

Background:

• Parkinson's disease (PD) presents significant heterogeneity in clinical symptoms and progression, challenging accurate prognosis and personalized treatment strategies.
• Existing Subtype and Stage Inference (SuStaIn) algorithms offer probabilistic modeling but require extensive comparative analysis and real-world validation for PD.

Purpose of the Study:

• To systematically compare various SuStaIn algorithm variants for modeling Parkinson's disease progression.
• To validate the stability and prognostic capabilities of these SuStaIn variants using simulated data and real-world Parkinson's Progression Markers Initiative (PPMI) cohorts.

Main Methods:

• A simulated PD cohort (n=400) with 12 motor/non-motor biomarkers and three predefined subtypes was created to mimic realistic data complexities.
• Six SuStaIn variants (Z-score, Ordinal, Event-based Mixture, Missing Data, s-SuStaIn, Temporal SuStaIn) were benchmarked on simulated and PPMI data (N=624).
• Model performance was assessed using cross-sectional subtype structure, test-retest stability (Cohen's κ), and prognostic prediction of clinical outcomes via Cox models.

Main Results:

• Simulation revealed six interpretable biomarker domains and three subtypes (slow, intermediate, fast); Temporal and s-SuStaIn variants showed optimal performance.
• PPMI data analysis identified three consistent subtypes: benign motor-predominant, intermediate mixed, and diffuse-malignant (PIGD-dominant).
• Ensemble SuStaIn subtypes significantly improved motor progression (Δc=+0.067) and cognitive decline prediction (Δc=+0.136), with enhanced stability (κ=0.81).

Conclusions:

• SuStaIn variants effectively capture distinct Parkinson's disease progression patterns, with an ensemble strategy providing stable and prognostically valuable subtypes.
• The identified diffuse-malignant subtype can aid clinical trial enrichment, potentially reducing sample size by 63%.
• This study provides a validated framework for applying algorithmic subtype discovery to real-world PD data, advancing precision medicine initiatives.