Predicting the Risk of Distal Junctional Kyphosis After Surgery for Adolescent Idiopathic Scoliosis Using Preoperative Radiographic Indicators and Machine Learning.

Study DesignRetrospective cohort study.ObjectivesTo develop and internally validate an interpretable machine learning model using preoperative radiographic indicators to predict distal junctional kyphosis (DJK) after surgery for Lenke type 1 or 2 adolescent idiopathic scoliosis (AIS).MethodsWe included 279 patients with Lenke type 1 or 2 AIS undergoing posterior spinal fusion, including 47 with DJK and 232 without DJK. Thirty-three preoperative variables were evaluated. The cohort was stratified 7:3 into training and internal hold-out validation sets. Feature selection and hyperparameter tuning were restricted to the training set. Boruta and least absolute shrinkage and selection operator regression were used for feature selection, and nine algorithms were compared.ResultsIn the internal hold-out validation set, support vector machine (SVM) had the numerically highest C-index (0.930; 95% CI, 0.867-0.992) versus logistic regression (0.912; 95% CI, 0.838-0.987). A paired DeLong test showed no statistically significant difference (difference, 0.017; 95% CI, -0.015 to 0.050; P = 0.292). SVM sensitivity was 0.857, specificity 0.843, precision 0.522, F1 score 0.649, and Brier score 0.073. Ten-fold nested cross-validation yielded a pooled C-index of 0.891 (95% CI, 0.826-0.948). SHAP identified TLK, LIV-SSV, and TK as dominant contributors.ConclusionsPreoperative sagittal alignment and the LIV-SSV relationship may aid exploratory DJK risk stratification. The three-feature SVM showed promising internal performance, but its numerical advantage over logistic regression does not establish superiority. External validation is required before clinical implementation.