[Construction of a prognostic assessment model of acupuncture intervention for lumbar disc herniation based on interpretable machine learning algorithms].

OBJECTIVE

To construct and validate a prognostic assessment model of acupuncture intervention for lumbar disc herniation (LDH), and analyze the key factors of acupuncture efficacy, so as to provide the decision support for clinical individualized treatment.

METHODS

Clinical data of 478 LDH patients were retrospectively analyzed. The least absolute shrinkage and selection operator (LASSO) regression was used to select predictive variables, and 8 machine learning prediction models were constructed, including decision tree, random forest, extreme gradient boosting, support vector machine, multilayer perceptron, logistic regression, light gradient boosting machine, and K-nearest neighbor. The performance of each model was evaluated through five-fold cross-validation. SHapley additive exPlanations (SHAP) method was employed to analyze model interpretability, and an online interactive application based on Shiny was developed.

RESULTS

LASSO regression selected 15 predictive variables; the support vector machine performed the best in five-fold cross-validation, with an average area under the receiver operating characteristic curve (AUC) of 0.862 and an average Brier score of 0.157. Decision curve analysis indicated a good clinical application value for this model. SHAP analysis showed that the combined therapies of acupuncture with Fu's acupuncture, warm needling, electroacupuncture and acupuncture delivered once daily were associated with favorable prognosis; and the higher body mass indexes (BMI), engagement in heavy physical labor, and the use of glucocorticoids and hyperosmotic dehydration agents, as well as the disc herniation of different segments and spinal canal stenosis were associated with unfavorable prognosis.

CONCLUSION

The interpretable machine learning-based prognostic assessment model of acupuncture intervention for LDH demonstrates a good predictive performance, providing evidence for clinical individualized treatment. However, more external validations are required for its optimization.