A Machine Learning Approach to Detecting Low Medication State with Wearable Technologies.

Area of Science:

• Digital Health
• Machine Learning in Healthcare
• Wearable Technology

Background:

• Medication adherence is crucial for patient outcomes and healthcare system efficiency.
• Non-adherence leads to significant financial and medical costs.
• Existing adherence monitoring methods are often insufficient to address complex patient needs.

Purpose of the Study:

• To develop and evaluate a novel method for detecting medication non-adherence using wearable sensor data.
• To leverage machine learning and cloud computing for accurate and scalable adherence monitoring.
• To explore the utility of smart devices in improving patient medication adherence.

Main Methods:

• Utilized sensor data from Apple Watch devices to monitor pill counts in prescription bottles.
• Employed distributed computing on a cloud platform for efficient processing of high-frequency sensor data.
• Trained a Gradient Boosted Tree machine learning model to classify adherence levels.

Main Results:

• The Gradient Boosted Tree model achieved an average cross-validated accuracy of 80.27%.
• The model demonstrated an average cross-validated F1 score of 80.22%.
• The system effectively detected instances of low medication adherence.

Conclusions:

• Wearable devices like the Apple Watch can be effectively utilized to monitor and improve medication adherence.
• This technology offers a scalable and data-driven approach to address a critical healthcare challenge.
• Future applications include personalized interventions to support patients with medication management.