Predicting Pain Response to a Remote Musculoskeletal Care Program for Low Back Pain Management: Development of a...

Area of Science:

• Digital health interventions
• Precision medicine in musculoskeletal care
• Artificial intelligence in healthcare

Background:

• Low back pain (LBP) has diverse presentations requiring personalized treatment.
• Current AI prediction tools lack scalability and real-time capabilities.
• Digital care programs (DCPs) offer ideal infrastructure for AI integration.

Purpose of the Study:

• Develop an AI tool for continuous prediction of pain relief in LBP patients.
• Identify predictors of nonresponse to guide treatment adjustments.
• Enhance precision medicine for LBP through AI-driven insights.

Main Methods:

• Utilized cloud-stored data from 6125 patients in a remote digital musculoskeletal program.
• Employed recurrent neural networks (RNNs) and LightGBM for continuous analysis.
• Assessed model performance using ROC-AUC, precision-recall curves, specificity, and sensitivity.

Main Results:

• AI model predictions improved over time, with ROC-AUC reaching 0.71 by session 7.
• Models demonstrated high specificity, prioritizing precision.
• Key predictors included pain, exercise performance, motivation, and compliance.

Conclusions:

• AI predictive tools within DCPs can significantly enhance LBP management.
• The tool supports physical therapists in dynamically adjusting care pathways.
• This approach is vital for addressing heterogeneous LBP phenotypes.