Predicting motor, cognitive & functional impairment in Parkinson's.

Area of Science:

• Neurology
• Digital Health
• Machine Learning

Background:

• Previous research showed smartphone tests could differentiate Parkinson's disease (PD) from controls and other conditions.
• Early detection and prediction of PD progression are crucial for effective management.

Purpose of the Study:

• To investigate if smartphone-derived features can predict future clinical outcomes in early Parkinson's disease.
• To assess the predictive accuracy of these features for key disease manifestations.

Main Methods:

• 237 early PD participants underwent smartphone tests (voice, balance, gait, reaction time, dexterity, tremor) and clinical assessments.
• Machine learning models were trained using baseline smartphone data to predict outcomes at 18-month follow-up.
• Prediction accuracy was evaluated using cross-validation schemes.

Main Results:

• Smartphone tests successfully predicted the onset of falls, freezing, postural instability, cognitive, and functional impairment at 18 months.
• Area Under the Curve (AUC) values exceeded 0.90 for predicting outcomes using all features (10-fold cross-validation).
• Using the top 30 features, AUC values greater than 0.75 were achieved.

Conclusions:

• Simple smartphone tests can predict significant future clinical outcomes in early Parkinson's disease.
• This approach holds potential for personalized predictions in routine care to guide targeted interventions.
• Early prediction can help improve patient outcomes by enabling timely and appropriate management strategies.