Improving Medication Regimen Recommendation for Parkinson's Disease Using Sensor Technology.

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Parkinson's disease treatment planning relies on subjective clinical assessments.
• Wearable sensors like the Personal KinetiGraph™ (PKG) offer objective, continuous remote monitoring.
• Objective data can potentially refine current subjective treatment strategies.

Purpose of the Study:

• To cluster Parkinson's disease patients based on levodopa regimen and response using objective sensor data.
• To enhance Parkinson's disease treatment planning with data-driven initial estimates.
• To evaluate the feasibility of using machine learning for remote patient assessment and medication response clustering.

Main Methods:

• Applied k-means clustering to Parkinson's patient data, including MDS-Unified Parkinson's Disease Rating Scale-III (MDS-UPDRS-III) scores and PKG sensor data.
• Utilized a random forest classification model to predict patient cluster allocation based on demographics, MDS-UPDRS-III, and PKG time-series data.
• Validated clusters based on levodopa dosage, frequency, and total equivalent daily dose.

Main Results:

• Identified clinically relevant patient clusters based on levodopa treatment and response.
• The PKG sensor provided symptomatic assessments comparable to physician-administered MDS-UPDRS-III scores.
• A random forest model achieved 86.9% accuracy, 90.7% F1 score, and 0.871 AUC in classifying patients into similar clusters.
• A model using only demographic and PKG data achieved 83.8% accuracy, 88.5% F1 score, and 0.831 AUC, supporting remote assessment.

Conclusions:

• Objective sensor data and machine learning can effectively cluster Parkinson's disease patients by medication response.
• These computational methods demonstrate the potential to supplement physician assessments and improve treatment planning.
• The findings support the feasibility of fully remote patient assessments for Parkinson's disease management.