Early Prediction of Parkinson's Disease Progression by Integrating Research Cohort and Real-World Data Using...

Area of Science:

• Neuroscience
• Computational Biology
• Biomedical Informatics

Background:

• Parkinson's disease (PD) progression varies significantly among individuals.
• Early prediction of PD progression is essential for effective patient management.
• Existing research cohorts are often too small and lack generalizability to real-world patient populations.

Purpose of the Study:

• To develop a machine learning framework for early prediction of Parkinson's disease progression trajectories.
• To address challenges of small sample sizes and data heterogeneity in PD research.
• To enable robust prediction models applicable to both research and real-world data.

Main Methods:

• Introduced MedStitcher, a graph-based machine learning framework.
• Utilized a biomedical knowledge graph to integrate multimodal data from research cohorts and real-world data (RWD).
• Enabled predictive modeling despite missing data modalities and cross-dataset population differences.

Main Results:

• MedStitcher achieved an AUROC of 0.819 ± 0.040 in predicting rapid PD progressors on combined PPMI and PDBP data.
• Outperformed existing machine learning approaches in PD progression prediction.
• Identified key clinical and molecular drivers including cognitive vulnerability, alpha-synuclein biology, vesicle trafficking, and neuroinflammation.

Conclusions:

• MedStitcher effectively predicts rapid PD progression by integrating diverse data sources.
• Predicted rapid progressors in RWD cohorts showed higher risks for dementia, falls, and cognitive/gait impairments.
• The framework facilitates the identification of early indicators for rapid PD progression in real-world settings.