Hierarchical clustering by patient-reported pain distribution alone identifies distinct chronic pain subgroups differing by pain intensity, quality, and clinical outcomes

Abstract

Background: In clinical practice, the bodily distribution of chronic pain is often used in conjunction with other signs and symptoms to support a diagnosis or treatment plan. For example, the diagnosis of fibromyalgia involves tallying the areas of pain that a patient reports using a drawn body map. It remains unclear whether patterns of pain distribution independently inform aspects of the pain experience and influence patient outcomes. The objective of the current study was to evaluate the clinical relevance of patterns of pain distribution using an algorithmic approach agnostic to diagnosis or patient-reported facets of the pain experience.

Methods and findings: A large cohort of patients (N = 21,658) completed pain body maps and a multi-dimensional pain assessment. Using hierarchical clustering of patients by body map selection alone, nine distinct subgroups emerged with different patterns of body region selection. Clinician review of cluster body maps recapitulated some clinically-relevant patterns of pain distribution, such as low back pain with radiation below the knee and widespread pain, as well as some unique patterns. Demographic and medical characteristics, pain intensity, pain impact, and neuropathic pain quality all varied significantly across cluster subgroups. Multivariate modeling demonstrated that cluster membership independently predicted pain intensity and neuropathic pain quality. In a subset of patients who completed 3-month follow-up questionnaires (N = 7,138), cluster membership independently predicted the likelihood of improvement in pain, physical function, and a positive overall impression of change related to multidisciplinary pain care.

Conclusions: This study reports a novel method of grouping patients by pain distribution using an algorithmic approach. Pain distribution subgroup was significantly associated with differences in pain intensity, impact, and clinically relevant outcomes. In the future, algorithmic clustering by pain distribution may be an important facet in chronic pain biosignatures developed for the personalization of pain management.

Fig 1. Hierarchical clustering of patient-reported body map selections reveals 9 distinct clusters.

Each cluster is represented with a body map. The frequency of each region being selected within each cluster is depicted by heatmap (increasing frequency with white–yellow–red, scale at bottom).

Fig 2. Hierarchical clustering reveals distinct subgroups despite similar total number of painful body regions.

A. Heatmap with body map region on horizontal axis and each row on the vertical axis representing an individual patient out of the entire cohort (N = 21,658 unique patients) organized by cluster membership. B. Box and whisker plots representing median values (vertical line), interquartile range (boxed area), and total range (whiskers) of the simple sum of body regions by cluster membership.

Fig 3. Cluster membership is associated with different average pain intensity at baseline.

Histograms reflect subgroup means with error bars representing 95% confidence intervals. Total N = 21,658 unique patients. A-Axial LBP, B-Abdominal Pain, C-LBP Thigh, D-Upper and Lower Back Pain, E-Neck and Shoulder, F-LBP Below Knee, G-Neck Shoulder and LBP, H-Widespread—Light, I-Widespread—Heavy.

Fig 4. Patient-reported function, pain interference, mood, and sleep vary by cluster membership.

Histograms reflect subgroup means with error bars representing 95% confidence intervals. Total N = 21,658 unique patients for physical function, N = 21,507 for pain interference, and N = 21,571 for anxiety, depression, and sleep disturbance. A-Axial LBP, B-Abdominal Pain, C-LBP Thigh, D-Upper and Lower Back Pain, E-Neck and Shoulder, F-LBP Below Knee, G-Neck Shoulder and LBP, H-Widespread—Light, I-Widespread—Heavy.

Fig 5. Cluster membership is associated with different neuropathic pain quality.

Histograms reflect subgroup means with error bars representing 95% confidence intervals. Total N = 12,950 unique patients completed the PainDetect questionnaire, reflecting a neuropathic quality of pain. PainDetect scores 19 or greater (neuropathic),13–18 (unclear), and 12 or less (not neuropathic) are highlighted with dashed lines. A-Axial LBP, B-Abdominal Pain, C-LBP Thigh, D-Upper and Lower Back Pain, E-Neck and Shoulder, F-LBP Below Knee, G-Neck Shoulder and LBP, H-Widespread—Light, I-Widespread—Heavy.

Fig 6. Baseline cluster membership predicts improvement in chronic pain at follow-up.

Percentage of patients at 3-month follow-up who self-reported clinically significant improvements, as determined by a composite outcome consisting of improved pain, function, or a positive impression of change, are plotted by cluster membership. Proportions of patients classified as responders (n) over the total number of patients in the cluster group (N) are displayed at the base of the bars (n/N).