Objective wearable measures correlate with self-reported chronic pain levels in people with spinal cord stimulation systems

Figures

• 

  Fig. 1. Improvement in PROs collected from the custom wearable application. a The weekly average of NRS scores across all participants during the study; 0 indicates the permanent implant date; the baseline period values (−1 to 0 month) are the weekly average of NRS pulled from the watch application; statistical testing indicates a comparison to pre-implant pain scores per subject and is calculated using the Wilcoxon signed-rank test on median values; *p ≤ 0.05, **p ≤ 0.001. b Monthly average of PGIC scores for all participants during the study; NRS numerical rating scale, PGIC patient global impression of change, CI confidence intervals.

• 

  Fig. 2. Important features from ML modeling for pain. a The average of feature importance for modeling pain levels using objective data across 10 runtimes. The standard deviation is shown with a line in the bar plot. b A sample of SHAP (SHapley Additive exPlanations) values for one run time. RMSSD: The root mean square of successive inter-beat intervals of heartbeats differences. SDANN: The standard deviation of the average inter-beat intervals without artifacts for each 5-minute interval over a 24-hour recording of heart rate variability. SDNNI: The average of the standard deviations of all inter-beat intervals without artifacts for each 5-minute interval over a 24-hour recording of heart rate variability.

• 

  Fig. 3. The average of heart rate median values correlates with the NRS score. The weekly average of heart rate median values across all participants during the study along with the weekly average of NRS scores; NRS numerical rating scale, CI confidence intervals.

• 

  Fig. 4. Feature engineering and predictive modeling pipeline. The patient-reported outcomes (PROs) of numerical rating scale (NRS), patient-reported outcomes measurement information system 29 (PROMIS-29), pain catastrophizing scale (PCS), patient health questionnaire-9 (PHQ-9), Oswestry disability index (ODI), patient global impression of change (PGIC) are outputs of machine learning models.