89: Wearable and Self-Report Data Show Divergent Daily Associations With Pain among Veterans

Topic: Basic Science
Subtopic: Non-specific to any diagnosis (including studies in healthy animals), Developing methodology/technology, Artificial Intelligence/Machine Learning, Behavioral studies

Background and Aims: Chronic pain is highly prevalent among military veterans and is known to fluctuate daily, yet clinicians lack the ability to predict when pain will meaningfully worsen. Emerging research suggests data from wearable devices and daily self-report surveys may help predict pain increases. However, the nature of the predictive associations between these data sources and pain increases remains poorly understood. We used complementary predictive and interpretive methodological approaches to characterize links between both wearable device and daily self-report survey data and day-to-day pain increases among a sample of veterans with chronic pain.

Methods: Veterans (n=70) completed daily smartphone surveys and wore wearable devices for up to three months (~4,100 person-days). Pain increases were defined as ≥2 points above each veteran's baseline pain level. Mixed-effects random forests (MErf), which capture nonlinear and interactive associations, were trained on 50 multiply-imputed datasets . Combined, wearable-only, and self-report-only models were compared. Top predictors from the MErf models were carried forward for interpretation: Shapley Additive Explanation (SHAP) values visualized the direction and shape of each predictor's association with pain increases, and generalized linear mixed models (GLMM) tested whether those same predictors showed significant linear associations with pain increases.

Results: All three MErf models performed comparably: combined (AUC=0.80, sensitivity=76%, specificity=68%), wearable-only (AUC=0.79), and self-report-only (AUC=0.79). However, analyses revealed divergent patterns of associations between wearable and self-report predictors. Wearable predictors (activity, sleep) ranked among the most important in the combined MErf model. However, SHAP visualizations revealed that, while self-report predictors showed clear, directional associations with pain increases, wearable predictor associations were more variable and nonlinear. GLMMs confirmed this pattern: two self-report predictors showed significant linear associations with pain increases (depression variability OR=1.97; daily stress OR=1.15; p <.05); no wearable predictors reached significance.

Conclusions: Both wearable device and daily self-report data contribute meaningfully to predicting daily pain increases, but their associations with pain appear fundamentally different in nature. Self-report predictors, particularly depression variability and daily stress, show straightforward linear associations with pain that appear consistent across individuals. Wearable device predictors appear to contribute through more complex, interactive, and nonlinear patterns. These findings suggest these multimodal data streams are complementary, rather than redundant, for predicting pain increases. For just-in-time interventions, self-report-based triggers may be able to rely on simple threshold rules, whereas wearable-based triggers may require individualized, machine learning-based decision rules.

References: Gatchel, R. J., Peng, Y. B., Peters, M. L., Fuchs, P. N., & Turk, D. C. (2007). The biopsychosocial approach to chronic pain: scientific advances and future directions. Psychological bulletin, 133(4), 581.

Hajjem, A., Bellavance, F., & Larocque, D. (2014). Mixed-effects random forest for clustered data. Journal of Statistical Computation and Simulation, 84(6), 1313-1328.

Lundberg, S. M., & Lee, S. I. (2017). A unified approach to interpreting model predictions. Advances in neural information processing systems, 30.

Nahin, R. L. (2017). Severe pain in veterans: the effect of age and sex, and comparisons with the general population. The journal of pain, 18(3), 247-254.

Nahum-Shani, I., Smith, S. N., Spring, B. J., Collins, L. M., Witkiewitz, K., Tewari, A., & Murphy, S. A. (2016). Just-in-time adaptive interventions (JITAIs) in mobile health: key components and design principles for ongoing health behavior support. Annals of behavioral medicine, 1-17.

Presenting Author: Shaddy Saba
Submitting Author: Shaddy Saba

Senior Author: Jordan Davis

IASP 2026 Global Year theme of Neuropathic Pain: No

• Daniel Lightly: PhD - King's College London
• John Prindle: PhD - University of Southern California
• Aysha Allahverdiyeva: BsC - New York University
• Eric Pedersen: PhD - University of Southern California
• Lingzi Luo: MSW/MPH - New York University
• Jordan Davis: PhD - RAND