Chronic neck pain is among the most common types of musculoskeletal pain. Manual therapy has been shown to have positive effects on this type of pain, but there are not yet many predictive models for determining how best to apply manual therapy to the different subtypes of neck pain. The aim of this study is to develop a predictive learning approach to determine which basal outcome could give a prognostic value (Global Rating of Change, GRoC scale) for Mulligan's mobilization technique and to identify the most important predictive factors for recovery in chronic neck pain subjects in four key areas: the number of treatments, time of treatment, reduction of pain, and range of motion (ROM) increase. A prospective cohort dataset of 80 participants with chronic neck pain diagnosed by their family doctor was analyzed. Logistic regression and machine learning modeling techniques (Generalized Boosted Models, Support Vector Machine, Kernel, Classsification and Decision Trees, Random Forest and Neural Networks) were each used to form a prognostic model for each of the nine outcomes obtained before and after intervention: disability-neck disability index (NDI), patient satisfaction (GRoC), quality of life (12-Item Short Form Survey, SF-12), State-Trait Anxiety Inventory (STAI), Beck Depression Inventory (BDI II), pain catastrophizing scale (ECD), kinesiophobia-Tampa scale of kinesiophobia (TSK-11), Pain Intensity Visual Analogue Scale (VAS), and cervical ROM. Pain descriptions from the subjects and pain body diagrams guided the physical examination. The most important predictive factors for recovery in chronic neck pain patients indicated that the more anxiety and the lower the ROM of lateroflexion, the higher the probability of success with the Mulligan concept treatment.
Keywords: Mulligan concept; chronic neck pain; chronic pain; machine learning; manual therapy; musculoskeletal manipulations.