Predicting Functional Improvement in Chronic Pain Using Machine Learning and Digital Health Data From the Manage My...

Area of Science:

• Digital Health
• Machine Learning
• Chronic Pain Management

Background:

• Chronic pain management is complex and challenging.
• Digital health tools offer potential for symptom monitoring and data collection.
• Predicting treatment outcomes is crucial for effective chronic pain care.

Purpose of the Study:

• To explore the utility of the Manage My Pain app for predicting significant user outcome improvements.
• To apply machine learning techniques to self-reported digital health data for outcome prediction.
• To assess the predictive performance of various machine learning models.

Main Methods:

• Extracted demographic, pain, and app usage features from 6,413 users over one month.
• Utilized temporal sequences of pain and function scores.
• Trained and validated multiple machine learning models, including Random Forest, CNN-MLP, and Time-Series Transformer-TabNet.

Main Results:

• Combining extracted and temporal features yielded superior predictive models.
• A CNN-MLP model achieved the highest balanced accuracy (0.79) and AUC (0.88).
• Machine learning integration with digital health data successfully predicted functional improvements.

Conclusions:

• Digital health tools like Manage My Pain can generate valuable data for outcome prediction.
• Machine learning models can forecast functional improvements in chronic pain patients.
• Predictive analytics hold transformative potential for chronic pain management and treatment strategies.