Prediction of Chronic Lower Back Pain Using the Hierarchical Neural Network: Comparison with Logistic Regression-A...

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Pain Management Research

Background:

• Chronic lower back pain (CLBP) is a prevalent condition with complex contributing factors.
• Conventional statistical models, like linear regression, struggle to capture nonlinear interactions influencing CLBP.
• There is a need for advanced predictive models to better understand and manage CLBP.

Purpose of the Study:

• To evaluate the predictive performance of a hierarchical neural network (HNN) against a conventional statistical model for CLBP.
• To determine if HNN offers superior accuracy in predicting CLBP compared to logistic regression (LR).

Main Methods:

• A cohort of 96 participants (30 men, 66 women, mean age ~72 years) was analyzed.
• Hierarchical neural network (HNN) and logistic regression (LR) models were developed for CLBP prediction.
• Model performance was compared using receiver operating characteristic (ROC) curves, root mean square error (RMSE), and -2 log likelihood.

Main Results:

• HNN achieved a higher area under the ROC curve (0.7650 vs. 0.7163) and lower RMSE (0.2483 vs. 0.2581) than LR.
• HNN demonstrated a lower -2 log likelihood (102.787 vs. 105.065) and a higher correct answer rate (73.3% vs. 70.8%).

Conclusions:

• The hierarchical neural network (HNN) model exhibits equal or superior performance in predicting CLBP probability compared to logistic regression (LR).
• HNN shows potential as a clinical tool for assessing individual patient risk for chronic lower back pain.