Development and Validation of Interpretable Machine Learning Models Incorporating Paraspinal Muscle Quality to Predict Cage Subsidence Risk Following Posterior Lumbar Interbody Fusion

Comprehensive Summary

The goal of this study was to identify the risk factors and develop a machine learning model to predict cage subsidence, a condition where the interbody implant sinks into the vertebral bones, after Posterior Lumbar Interbody Fusion (PLIF). Cage subsidence can cause a loss of disc height, spinal alignment issues, and nerve root compression. The study included 620 patients who dealt with degenerative lumbar diseases who underwent PLIF, each using the same interbody device (static polyetheretherketone cage). Important measurements taken were the intervertebral height (IH) and the segmental angle (SA) of the fusion segment, each at preoperative, postoperative, and follow-up time points. Additionally, data on bone quality, paraspinal muscle condition, surgery details, and basic demographics were collected. The study found that independent risk factors such as poor bone quality, weak paraspinal muscles, cage placement, and cage size all were linked to a higher chance of the cage sinking into the bone. With this information, eight machine models were evaluated: LightGBM, Logistic Regression, Random Forests, SVM, XGBoost, Gradient Boosting, Balanced Bagging, and MLP Classifier. LightGBM performed the best, boasting the highest area under the curve score (0.9752), accuracy (0.92), and f1 score (0.9216), which measures precision involving false positives and negatives. LightGBM also showed the lowest Brier score (0.0660), indicating that it performed the most accurate probabilistic predictions. An important note is that when muscle condition changed, the accuracy of all models decreased significantly.

Outcomes and Implications

The information learned in this study is important to surgeons when planning a PLIF surgery, as surgeons can predict the probability of any given patient suffering from cage subsidence after the surgery. This can help surgeons to adjust their surgical strategies, change the type of implant used, or even just to help patients understand the risk they agree to when undergoing the surgery. Another key implication is the significance of paraspinal muscle condition as a risk factor for cage subsidence.