Interpretable machine learning for cognitive impairment prediction in Parkinson's disease: a multicenter validation...

Area of Science:

• Neurology
• Artificial Intelligence
• Biomarkers

Background:

• Parkinson's disease (PD) often leads to cognitive impairment (PD-CI), impacting quality of life.
• Current PD-CI detection methods may require specialized resources or lack broad applicability.
• There is a need for accessible and interpretable tools for PD-CI screening.

Purpose of the Study:

• To develop an interpretable machine learning framework for detecting PD-CI.
• To utilize only routine clinical data for PD-CI prediction.
• To ensure the framework is generalizable across different populations.

Main Methods:

• Analysis of 1,279 participants from the Parkinson's Progression Markers Initiative (PPMI) and 197 from an independent cohort.
• PD-CI defined by Montreal Cognitive Assessment (MoCA) and Unified Parkinson's Disease Rating Scale Part I (UPDRS-I) scores.
• Training and optimization of four machine learning models using 21 clinical features and cross-validation.

Main Results:

• Random Forest model achieved the highest AUC (0.83) in the discovery cohort.
• External validation demonstrated 71.57% accuracy.
• Key predictors identified include age, neutrophil-to-lymphocyte ratio (NLR), and serum uric acid, indicating roles for inflammation and oxidative stress.

Conclusions:

• The developed framework offers comparable accuracy to neuroimaging using accessible clinical data.
• It highlights neuroinflammation and oxidative stress as critical factors in PD-CI.
• Multicenter validation supports its use as a clinically actionable tool for PD-CI screening and monitoring.