JMIR Publications

ABSTRACT

Background:

Low back pain (LBP) appears with diverse presentations, requiring personalized treatment that recognize different phenotypes within the same diagnosis which could be attainable through precision medicine. While prediction strategies have been explored, even using artificial intelligence (AI), they still lack scalability and real-time capabilities. Digital care programs (DCP) enable seamless data collection through Internet of Things (IoT) and cloud storage, fostering the ideal condition to develop and introduce AI within a digital twin framework to dynamically optimize treatment.

Objective:

To develop an AI tool within a digital twin framework to assist physical therapists in predicting an individual's potential for achieving clinically significant pain relief by program-end. A secondary aim was to identify predictors of pain non-response to guide treatment adjustments.

Methods:

Data collected passively (e.g. demographic and clinical information, combined with socioeconomic data - from public data sources) and actively in real-time (e.g. range of motion, exercise performance) from 6,125 patients enrolled in a remote digital musculoskeletal intervention program, were stored in the cloud, allowing the virtual representation of the patient. Recurrent neural networks (RNN) and light gradient boosting machine (LightGBM) models continuously analyzed session updates, utilizing data from each patient up to session 7. Model performance was assessed using the area under the receiver operating characteristic curve (ROC-AUC), Precision-Recall curves, specificity, and sensitivity. Model explainability was assessed using SHapley Additive exPlanations (SHAP) values.

Results:

At each session, the model provided a prediction about the potential of being a pain responder, with performance improving over time. By the 7th session, the RNN achieved an ROC-AUC of 0.70 (95% CI: 0.65–0.71) and the LightGBM achieved an ROC-AUC of 0.71 (95% CI: 0.67–0.72). Both models demonstrated high specificity in scenarios prioritizing high precision. Key predictive features included pain-associated domains, exercise performance, motivation, and compliance, informing continuous treatment adjustments to maximize response rates.

Conclusions:

This study underscores the potential of a digital twin framework leveraged by ML models to enhance the management of LBP in a digital setting, redirecting care pathways early and along the treatment course, particularly important for the heterogeneous phenotypes observed in LBP. Clinical Trial: ClinicalTrials.gov (NCT04092946, NCT05417685)