Pain qualities associated with body perception disturbances: insights from machine learning and SHapley additive exPlanations.

Abstract

To identify specific pain qualities associated with body perception disturbances in patients with musculoskeletal pain, using machine learning interpretability methods. A cross-sectional study utilizing self-reported questionnaires and SHapley Additive exPlanation (SHAP) analysis within a random forest model. Multicenter clinical settings where patients with musculoskeletal pain received physical therapy. A total of 179 patients with musculoskeletal pain, without restrictions on pain location or duration. Patients with cognitive impairment or difficulty completing questionnaires were excluded. Pain qualities assessed using the Short-Form McGill Pain Questionnaire-2 (SFMPQ-2), and body perception disturbances assessed using the Fremantle Body Awareness Questionnaire (FreBAQ). SHAP values were calculated to quantify the relationship between individual pain qualities and body perception disturbances. none. SHAP-based importance ranked "cramping pain" among the top contributors, whereas bivariate correlations with FreBAQ were strongest for "gnawing pain" (r = 0.90, p < 0.001), indicating convergence on kinesthesia-related descriptors. Multiple pain-quality descriptors were associated with body perception disturbances. While the analytic metrics differed in how they ranked individual descriptors, both SHAP-based importance and correlation analyses converged on kinesthesia-related qualities. These results indicate a prominent association between such descriptors and disturbed body perception, potentially informing sensorimotor-focused assessment and intervention strategies.