Detection of Medication Taking Using a Wrist-Worn Commercially Available Wearable Device

Area of Science:

• Digital health
• Biomedical engineering
• Behavioral science

Background:

• Medication nonadherence is a significant healthcare challenge with substantial costs.
• Current adherence monitoring methods (self-report, claims data, smart bottles) have limitations including recall bias, lack of real-time feedback, and high expense.

Purpose of the Study:

• To develop and evaluate a novel method for monitoring medication adherence.
• To leverage commercially available wearable devices for passive data collection and analysis.

Main Methods:

• Utilized passively collected motion data from wrist-worn wearable devices.
• Applied the Movelet algorithm, a dictionary learning framework, to analyze movement patterns.
• Adapted the Movelet method to create patient-specific models for predicting medication-taking behaviors.

Main Results:

• Demonstrated prediction of medication-taking behavior with a median accuracy of 85% in a controlled clinical setting.
• Analyzed 15 activity features from 10 breast cancer patients undergoing endocrine therapy.
• Validated the feasibility of using wearable device data for adherence monitoring.

Conclusions:

• Patient-specific models derived from wearable motion data show potential for real-time medication adherence measurement.
• Commercially available wearable devices offer a viable platform for innovative adherence monitoring solutions.