Construction and Validation of a Prediction Model for Surgical Outcome in Degenerative Lumbar Scoliosis

Abstract Background Degenerative lumbar scoliosis (DLS) is a complex spinal disorder that has attracted increasing attention in contemporary spine research and clinical practice. Predicting postoperative outcomes in patients with DLS remains challenging. This study aimed to identify independent predictors of postoperative outcomes and to develop and validate a prognostic model for patients with DLS. Methods This study included patients with DLS managed by posterior-only fusion surgery. Demographic, radiological and operating characteristics of patients were collected and organized from the hospital medical records. The surgical outcome was evaluated according to the Oswestry Disability Index (ODI), Physical Functioning (PF) and Mental Health (MH) in the Short Form- 36 (SF-36), complications and revision surgery. Satisfactory outcome (SO) was defined by ODI ≤ 22, PF and MH at or above normative values, no postoperative complications, and no revision surgery. Predictors of postoperative prognosis were identified through the least absolute shrinkage and selection operator (LASSO) regression analysis. The Shapley Additive Explanations (SHAP) method was applied to rank the importance of each variable. Four predictive models—Logistic Regression, KNearest Neighbors (KNN), Multi-layer Perceptron (MLP), Naive Bayes—were constructed using the training set and evaluated for performance, with validation conducted on the test set. Results In this retrospective case-control study, 233 patiens (mean age 61.4 years) were included. Ninety-nine patients (42.4%) patients achieved SO. LASSO regression identified six variable features, which, ranked in order of importance by SHAP analysis, were preoperative ODI, postoperative pelvic incidence minus lumbar lordosis (PI-LL), preoperative PF, preoperative Cobb angle, number of decompression segments, and postoperative LL. Among the four predictive models, the MLP model achieved the best performance, with an AUC of 0.869 in the training set and 0.851 in the test set. Clinical decision curve analysis indicated that the MLP model provides significant clinical utility. An MLP-based calculator was developed to support surgical prognostic assessment in patients with DLS. Conclusion Preoperative ODI, postoperative PI-LL, preoperative PF, preoperative Cobb angle, the number of decompression segments, and postoperative LL were defined as independent predictors of surgical prognosis in patients with DLS. The MLP-based models demonstrated robust performance, providing a valuable tool for prognosis prediction.