Smartwatch-derived digital biomarkers distinguish episodic pain phenotypes in chronic low back pain
- 1University of California San Francisco
Abstract
Chronic low back pain affects over half a billion people globally and is the leading cause of disability worldwide, yet objective tools for characterizing how individual patients experience pain over time remain lacking. Patients with chronic low back pain exhibit distinct pain trajectory patterns: some experience relatively stable pain while others experience episodic fluctuations with flare-ups. These trajectories have been found to be clinically meaningful but are currently captured only through subjective self-report. Consumer smartwatches offer an opportunity for passive, continuous, objective monitoring of physiological and behavioral signals that may reflect these fluctuations.
We evaluated whether temporal features extracted from six months of smartwatch-derived resting heart rate, heart rate variability, and step count data could discriminate between episodic and non-episodic pain phenotypes in 261 chronic low back pain patients from a longitudinal observational cohort. We applied stability-selected elastic-net logistic regression models to temporal features and evaluated model performance using nested cross-validation and SHAP-based feature importance analysis.
Summary statistics showed no significant differences between pain groups across any signal modality. Functional principal component analysis revealed high temporal heterogeneity in trajectories of resting heart rate, heart rate variability, and step count. Temporal trajectory models substantially outperformed summary-based approaches, with a models achieving areas under the receiver operating curve ranging from 0.66 to 0.82. Frequency-domain features and local variability measures drove classification performance, and combining heart rate and activity features provided complementary discriminative information, performing better than models trained on any subset of modalities.
These findings demonstrate the feasibility of passively collected smartwatch data as objective digital biomarkers for pain phenotyping in chronic low back pain, establishing a methodological approach that may permit personalized pain management through objective monitoring of patient pain and response to intervention.
AUTHOR SUMMARYChronic low back pain is the most common cause of disability worldwide, affecting over half a billion people. One of the biggest challenges in managing this condition is that patients experience it differently: some have relatively constant pain, while others experience unpredictable flare-ups. Understanding which pain trajectory a patient follows is important for guiding treatment but currently relies on patients recalling their experience, which is inherently subjective.
In this study, we asked whether data passively collected from consumer smartwatches could objectively distinguish between these two types of pain experience. We analyzed six months of heart rate and step count data from 261 chronic low back pain patients and found that simple averages of these signals looked identical between patient groups. However, when we examined how these signals changed and fluctuated over time, we were able to distinguish episodic from non-episodic pain patients with meaningful accuracy. Notably, combining heart rate and activity data performed better than either alone.
This work suggests that the smartwatch many people already wear on their wrist could one day help clinicians objectively monitor chronic pain in real time, reducing the burden of self-reporting and potentially enabling earlier detection of pain flare-ups.
Related
- Smartwatch-Derived Biomarkers Signal Pace of Biological Aging
- Explainable machine learning reveals threshold associations and distinct phenotypes linking migraine features to white matter hyperintensities in women with episodic migraine.
- Sepsis-associated endothelial dysfunction: clinical phenotypes, biomarkers, bedside evidence, and translational implications
- 1078 Mutational signatures as mechanistic biomarkers of tumor immunity: associations with TMB, PD-L1 and immune phenotypes across cancer types
- Plant‐Derived Senolytic Ingredient, Camellia sinensis Root Extract, Ameliorates Skin Aging‐Associated Phenotypes
BibTeX
@misc{khattab2026smartwatchderived,
title = {{Smartwatch-derived digital biomarkers distinguish episodic pain phenotypes in chronic low back pain}},
author = {K. Khattab and A. Pratapneni and G. Kurillo and T. Hue and P. Zheng and J. Lotz and A. Torres-Espin and J. F. Bailey},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.02.26362006},
url = {https://www.medrxiv.org/content/10.64898/2026.09.02.26362006v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Smartwatch-derived digital biomarkers distinguish episodic pain phenotypes in chronic low back pain
AU - K. Khattab
AU - A. Pratapneni
AU - G. Kurillo
AU - T. Hue
AU - P. Zheng
AU - J. Lotz
AU - A. Torres-Espin
AU - J. F. Bailey
PY - 2026
DA - 2026/09/02
PB - medRxiv
DO - 10.64898/2026.09.02.26362006
UR - https://www.medrxiv.org/content/10.64898/2026.09.02.26362006v1
AB - Chronic low back pain affects over half a billion people globally and is the leading cause of disability worldwide, yet objective tools for characterizing how individual patients experience pain over time remain lacking. Patients with chronic low back pain exhibit distinct pain trajectory patterns: some experience relatively stable pain while others experience episodic fluctuations with flare-ups. These trajectories have been found to be clinically meaningful but are currently captured only through subjective self-report. Consumer smartwatches offer an opportunity for passive, continuous, objective monitoring of physiological and behavioral signals that may reflect these fluctuations. We evaluated whether temporal features extracted from six months of smartwatch-derived resting heart rate, heart rate variability, and step count data could discriminate between episodic and non-episodic pain phenotypes in 261 chronic low back pain patients from a longitudinal observational cohort. We applied stability-selected elastic-net logistic regression models to temporal features and evaluated model performance using nested cross-validation and SHAP-based feature importance analysis. Summary statistics showed no significant differences between pain groups across any signal modality. Functional principal component analysis revealed high temporal heterogeneity in trajectories of resting heart rate, heart rate variability, and step count. Temporal trajectory models substantially outperformed summary-based approaches, with a models achieving areas under the receiver operating curve ranging from 0.66 to 0.82. Frequency-domain features and local variability measures drove classification performance, and combining heart rate and activity features provided complementary discriminative information, performing better than models trained on any subset of modalities. These findings demonstrate the feasibility of passively collected smartwatch data as objective digital biomarkers for pain phenotyping in chronic low back pain, establishing a methodological approach that may permit personalized pain management through objective monitoring of patient pain and response to intervention. AUTHOR SUMMARYChronic low back pain is the most common cause of disability worldwide, affecting over half a billion people. One of the biggest challenges in managing this condition is that patients experience it differently: some have relatively constant pain, while others experience unpredictable flare-ups. Understanding which pain trajectory a patient follows is important for guiding treatment but currently relies on patients recalling their experience, which is inherently subjective. In this study, we asked whether data passively collected from consumer smartwatches could objectively distinguish between these two types of pain experience. We analyzed six months of heart rate and step count data from 261 chronic low back pain patients and found that simple averages of these signals looked identical between patient groups. However, when we examined how these signals changed and fluctuated over time, we were able to distinguish episodic from non-episodic pain patients with meaningful accuracy. Notably, combining heart rate and activity data performed better than either alone. This work suggests that the smartwatch many people already wear on their wrist could one day help clinicians objectively monitor chronic pain in real time, reducing the burden of self-reporting and potentially enabling earlier detection of pain flare-ups.
ER -