A machine learning algorithm successfully screens for Parkinson's in web users.

Area of Science:

• Neurology
• Computer Science
• Public Health

Background:

• Parkinson disease (PD) diagnosis can be challenging and often occurs late.
• Early detection and monitoring are crucial for managing PD progression.
• Leveraging digital footprints for health screening is an emerging area.

Purpose of the Study:

• To develop and evaluate a novel web-based classifier for screening Parkinson disease (PD) in a large cohort of search engine users.
• To assess the classifier's ability to identify individuals at risk for PD based on their online behavior.
• To investigate disease progression in individuals screened positive for PD.

Main Methods:

• A supervised machine learning classifier was trained using search engine interaction data (mouse and keyboard movements) to distinguish users with self-reported PD from controls.
• The classifier was applied to a large cohort of web users over 40 years old.
• Web query content was used to categorize users into risk groups but not for classification; disease detection was unsolicited and anonymized.

Main Results:

• The classifier identified 1.2% (17,843/1,490,987) of users over 40 as potentially having PD.
• Screen-positive rates were significantly higher in at-risk groups, including those searching for PD information (64.4%) or with non-motor symptoms/affected relatives (5.3%).
• Individuals classified as having PD showed a higher mean rate of progression in disease-related features during longitudinal follow-up.

Conclusions:

• An automated classifier using search engine interactions can reliably identify individuals at high risk for Parkinson disease.
• The classifier demonstrated a higher rate of disease progression in positive cases, suggesting its utility for monitoring.
• The findings highlight the potential of digital tools for unsolicited disease screening and raise important ethical considerations.