Abstract
Parkinsons disease (PD) exhibits considerable variability in symptom onset, severity, and clinical trajectories, making prediction of individual disease progression a complex challenge. Leveraging large multimodal datasets from two independent cohorts of early PD patients, we examined predictive features for progression to mild cognitive impairment (MCI) and postural instability (PI) using Explainable Boosting Machines (EBM) - an interpretable machine learning (ML) approach.
We found comparable predictive performances of EBM models using a minimal clinico-demographic dataset compared to an extended multimodal dataset including volumetric brain imaging and fluid biomarkers. Baseline age and Montreal Cognitive Assessment total score were identified as most important predictive features for both MCI and PI using the minimal dataset, while additional motor, cognitive and imaging features were identified using the extended dataset. Our study highlights the potential of ML predictions of cognitive and motor endpoints of PD progression at the individual level bringing personalized clinical decision-making within reach.