Four Variables Were Sufficient for Low Back Pain: Determining Which Patient-Reported Tools Pain and Disability Improvements.

To predict 30- and 180-day improvements in disability and pain for patients seeking physical therapy care for low back pain (LBP). Longitudinal cohort. Baseline assessment was completed by 259 patients with chief complaint of LBP, and the assessment includes psychosocial measures (Keele STarT Back Screening [SBST] and the Optimal Screening for Prediction of Referral and Outcome Yellow Flag [OSPRO-YF] tools), the Optimal Screening for Prediction of Referral and Outcome Review of Symptoms (OSPRO-ROS) and the Review of Symptoms Plus (OSPRO-ROS+) tools, the Charlson Comorbidity Index (CCI), the Area Deprivation Index (ADI), and the National Institute of Health Chronic Pain Criteria (NIH-CP). Using the Modified Low Back Disability Questionnaire (MDQ) and the Numeric Pain Rating Scale (NPRS) as primary outcomes, statistical analysis determined multiple sets of predictor variables with similar model performance. The parsimonious "best model" for prediction of the 180-day MDQ change included 3 predictors (Admit MDQ, NIH-CP, and OSPRO ROS+) because it had the lowest penalized goodness-of-fit statistic (BIC = -35.21) and the highest explained variance (R2 = 0.295). The parsimonious "best model" for 180-day NPRS change included 2 variables (Admit NPRS and OSPRO-ROS+) with the lowest penalized goodness-of-fit statistic (BIC = -18.2) and the highest explained variance (R2 = 0.190). There were many model options with similar statistical performance when using established measures to predict MDQ and NPRS outcomes. A potential variable set for a standard predictive model that balances statistical performance with pragmatic considerations included the OSPRO-ROS+, OSPRO-YF, NIH-CP definition, and admit MDQ and NPRS scores. .