Study overview
Remote MSK programs generate rich clinical and engagement data over time. This study evaluated whether machine learning could predict which members were likely to experience meaningful pain relief during a remote MSK care program.
Researchers analyzed data from 6,125 patients and tested machine learning models, including recurrent neural networks and LightGBM, using information gathered across the first seven sessions. The goal was to understand whether early signals could guide treatment adjustments and more personalized care.
Key findings
Prediction accuracy improved over time
The Sword summary reports that model performance improved as more session data became available, reaching an AUC of 0.70 to 0.71 by session 7.
Early program data helped inform response prediction
The study suggests that member interaction, clinical status, and early response patterns can help predict future pain outcomes in a remote care setting.
The study supports more personalized digital care
By identifying people who may be less likely to respond, predictive models could help clinical teams adjust care earlier rather than waiting until the end of a program.
AI was used to support clinical decision-making
This study is an example of AI applied to a practical care challenge: understanding response trajectory and helping clinicians personalize care.
Why this study matters
This study belongs at the center of Sword’s AI care evidence story. It shows how AI can move digital MSK care from standardized delivery toward more responsive, data-informed personalization.
The page should avoid implying that prediction alone improves outcomes. The strongest framing is that predictive modeling may help clinicians identify risk earlier, tailor support, and make remote care more proactive.
About the author
Fabiola Costa
Senior Director of Clinical Research, Sword Health
