Abstract
OBJECTIVE: Musculoskeletal pain conditions are a leading cause of pain and disability internationally and a common reason to seek health care. Accurate prediction of recurrence of health care seeking due to musculoskeletal conditions could allow for better tailoring of treatment. The aim of this project was to characterize patterns of recurrent physical therapy seeking for musculoskeletal pain conditions and to develop a preliminary prediction model to identify those at increased risk of recurrent care seeking. DESIGN: Retrospective cohort. SETTING: Ambulatory care. SUBJECTS: Patients (n = 578,461) seeking outpatient physical therapy (United States). METHODS: Potential predictor variables were extracted from the electronic medical record, and patients were placed into three different recurrent care categories. Logistic regression models were used to identify individual predictors of recurrent care seeking, and the least absolute shrinkage and selection operator (LASSO) was used to develop multivariate prediction models. RESULTS: The accuracy of models for different definitions of recurrent care ranged from 0.59 to 0.64 (c-statistic), and individual predictors were identified from multivariate models. Predictors of increased risk of recurrent care included receiving workers' compensation and Medicare insurance, having comorbid arthritis, being postoperative at the time of the first episode, age range of 44-64 years, and reporting night sweats or night pain. Predictors of decreased risk of recurrent care included lumbar pain, chronic injury, neck pain, pregnancy, age range of 25-44 years, and smoking. CONCLUSION: This analysis identified a preliminary predictive model for recurrence of care seeking of physical therapy, but model accuracy needs to improve to better guide clinical decision-making.