INTRODUCTION: This study utilized machine learning (ML) models to predict post-treatment outcomes in patients with chronic non-specific lumbar pain (CNSLP) undergoing lumbar extension traction therapy designed to increase the hypo-lordotic lumbar spine. The objective is to identify key factors that influence clinical outcomes, such as lumbar lordotic angle, pain scores, and the Oswestry Disability Index (ODI), with the aim of improving personalized treatment approaches and optimizing interventions.
METHODS: We utilized a retrospective consecutive case series data set from patient records from 2010 to 2023. 431 patients met the inclusion criteria including CNSLP with hypo-lumbar lordosis. Additional inclusion criteria included patients being treated with a multi-modal program including the primary intervention of lumbar extension traction and completing pre- and post-treatment data. Patients with a history of lumbar surgery, spinal deformities, or other specific diagnoses (e.g., herniated discs) were carefully excluded to maintain the study's focus and integrity. The dataset included a range of demographic variables as well as pre-treatment lumbar curvature (L1-L5 and sacral base angle), extension traction duration, NRS pain scores, and ODI. The fit of lordosis relative to sacral base angle was categorized. All variables were analyzed to predict post-treatment clinical outcomes. Three machine learning models—Random Forest (RF), XGBoost, and Multilayer Perceptron (MLP), were employed due to their capability to handle both continuous and categorical variables effectively. Model performance was evaluated in terms of predictive accuracy, and the most influential factors affecting treatment outcomes were identified using Shapley Additive Explanations (SHAP).
RESULTS: The XGBoost model outperformed the other algorithms in terms of predictive accuracy, particularly for the prediction of lumbar lordotic angle, pain, and ODI reduction. Feature importance analysis revealed that the pre-treatment lumbar curve (lumbar lordosis angle), traction duration, number of sessions, and curve fit type were the most significant predictors of post-treatment outcomes.
DISCUSSION: CNSLP remains one of the leading causes of disability worldwide, significantly impacting patients' function and quality of life. Traditional treatment modalities, including lumbar extension traction therapy, have been shown to be effective in addressing spinal misalignments and improving clinical outcomes. Despite its efficacy, predicting which patients will respond most favorably to such interventions remains a clinical challenge. This study aimed to fill this gap by developing a predictive machine learning (ML) model that incorporates key mechanical and demographic predictors to forecast treatment outcomes for lumbar extension traction therapy in patients with CNSLP. Specifically, we evaluated and compared the predictive performance of three widely used machine learning models—Forest (RF), XGBoost, and Multilayer Perceptron (MLP)—in predicting post-treatment outcomes, including the lumbar lordotic angle, pain score, and Oswestry Disability Index (ODI). The models were assessed using various performance metrics, including R², Root Mean Square Error (RMSE), Mean Absolute Error (MAE), as well as SHAP-based explainability and sensitivity analysis. These metrics provided valuable insights into each model’s ability to predict key treatment outcomes and their clinical relevance in personalizing lumbar extension traction therapy for individual patients.By identifying key predictors such as lumbar curvature and traction duration, the research provides a foundation for personalized treatment planning.