Prediction of lumbar disc degeneration based on interpretable machine learning models: retrospective cohort study

Comprehensive Summary

In this study, Li et al. developed and validated a predictive machine learning model (MLM) to examine the impact of various paravertebral muscles on the risk of lumbar disc degeneration (LDD). The study set included 861 patients and was split into training, internal validation, and external validation groups. Four machine learning algorithms were tested in the development of the predictive model: Random Forest (RF), Extreme Gradient Boosting (XGBoost), Logistic Regression (LR), and Decision Tree (DT). The algorithm was developed and optimized using the training set, followed by internal and external validation using the respective test sets. XGBoost showed the strongest performance (AUC = 0.941, Accuracy = 0.871, Sensitivity = 0.817). The model also identified significant predictors for LDD, such as age, Psoas Muscle Index (PMI), and Multifidus Functional Cross-Sectional Area (MF fCSA). Overall, this MLM provides a novel approach to predicting LDD risk and may serve as a useful clinical tool after further refinement.

Outcomes and Implications

The use of MLMs in the prediction of LDD has positive clinical implications, benefitting both physicians and patients. This tool promotes earlier risk identification, which can lead to better planning and overall improved patient outcomes. Additionally, the use of this model can lead to more preventive strategies for LDD, allowing for the preservation of spinal stability and improved quality of life for patients. While this model shows promise for predicting LDD, further research and broader validation is needed to confirm its generalizability.