Machine Learning-Based Prediction of Quality of Life Improvement After Surgery for Spinal Metastases

Study Design. A prospective multicenter cohort study. Objective. To develop and validate machine learning models for predicting health-related quality of life (HRQoL) improvements in patients after one month and six months of surgery for spinal metastases. Summary of Background Data. The prediction of postoperative HRQoL of spinal metastases surgery remains understudied compared with studies of survival outcomes. Methods. We analyzed data from 413 patients who underwent surgery for spinal metastases at 40 participating institutions in Japan. The primary outcome was HRQoL improvement, defined as an increase in the EuroQol 5-Dimension 5-Level (EQ-5D) utility value of ≥0.32 from baseline. We developed two models for 1-month (n=360) and 6-month (n=189) outcomes using various machine learning algorithms. Missing values were imputed, and feature selection was performed using recursive feature elimination with cross-validation. We split the data into training (80%) and test (20%) sets for each model. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, precision, and F1-score. SHapley Additive exPlanations (SHAP) analysis was used to interpret feature importance. Results. The 6-month model outperformed the 1-month model across all metrics. For 1-month predictions, Logistic Regression achieved an AUC of 0.8136 and an accuracy of 0.7639 on the test set. For 6-month predictions, Naive Bayes demonstrated an AUC of 0.8928 and an accuracy of 0.8684. The 1-month model used 12 features, while the 6-month model required seven. SHAP analysis revealed that EQ-5D Mobility was the most influential feature in both models. Conclusions. Our models demonstrate high predictive accuracy for HRQoL improvements following spinal metastases surgery, with superior performance of the 6-month model. These models could enhance clinical decision-making and patient counseling by providing personalized predictions of postoperative QoL. Future research should focus on external validation and integration of these models into clinical practice.