Core patient-reported outcome measures for chronic pain patients treated with spinal cord stimulation or dorsal root ganglia stimulation

Abstract

Background: Neurostimulation is a highly effective therapy for the treatment of chronic Intractable pain, however, due to the complexity of pain, measuring a subject's long-term response to the therapy remains difficult. Frequent measurement of patient-reported outcomes (PROs) to reflect multiple aspects of subjects' pain is a crucial step in determining therapy outcomes. However, collecting full-length PROs is burdensome for both patients and clinicians. The objective of this work is to identify the reduced set of questions from multiple validated PROs that can accurately characterize chronic pain patients' responses to neurostimulation therapies.

Methods: Validated PROs were used to capture pain, physical function and disability, as well as psychometric, satisfaction, and global health metrics. PROs were collected from 509 patients implanted with Spinal Cord Stimulation (SCS) or Dorsal Root Ganglia (DRG) neurostimulators enrolled in the prospective, international, post-market REALITY study (NCT03876054, Registration Date: March 15, 2019). A combination of linear regression, Pearson's correlation, and factor analysis were used to eliminate highly correlated questions and find the minimal meaningful set of questions within the predefined domains of each scale.

Results: The shortened versions of the questionnaires presented almost identical accuracy for classifying the therapy outcomes as compared to the validated full-length versions. In addition, principal component analysis was performed on all the PROs and showed a robust clustering of pain intensity, psychological factors, physical function, and sleep across multiple PROs. A selected set of questions captured from multiple PROs can provide adequate information for measuring neurostimulation therapy outcomes.

Conclusions: PROs are important subjective measures to evaluate the physiological and psychological aspects of pain. However, these measures are cumbersome to collect. These shorter and more targeted PROs could result in better patient engagement, and enhanced and more frequent data collection processes for digital health platforms that minimize patient burden while increasing therapeutic benefits for chronic pain patients.

Keywords: Chronic pain therapy; Confirmatory factor analysis; Digital health; Dimensionality reduction; Patient-reported outcomes; Questionnaire burden; Spinal cord stimulation.

Fig. 1

Analysis and validation flowchart. Statistical analyses and dimensionality reduction algorithms were performed to select the short-form questions. The left panel shows the long-form validated patient-reported outcomes (PROs) as inputs. The middle panel shows different algorithms used to generate the short-form questions. The right panel shows the ROC classifier used to differentiate model performance for both long-form and selected short-form questions

Fig. 2

The PCA loading plot of the first 3 principal components. PCS, PROMIS29, PGIC, and ODI scales and their sub-domains were selected for this analysis. PR is an abbreviated version of the PROMIS-29; Total stands for the total score for each scale of ODI and PCS. The black dashed line shows the manual grouping of the domains based on the loading weights in all three dimensions

Fig. 3

Correlation heatmap for PROMIS-29 domains. The colors are based on the correlation thresholds, yellow (strong positive correlation), light green (weak to moderate positive correlation), and dark green (strong negative correlation). The red solid line borders separate the PROMIS-29 domains to highlight the correlations within each PROMIS domain. The red dashed line shows the borders for the strongly correlated combined PROMIS-29’s Anxiety and Depression domains

Fig. 4

CFA flowchart for each domain of PROMIS-29. Short-form questions were selected based on the highest standardized factor loadings in each domain

Fig. 5

CFA flowchart for each domain of the PCS. Short-form questions were selected based on the highest standardized factor loadings in each domain