Developmental and Validation of Machine Learning Model for Prediction Complication after Cervical Spine Metastases Surgery.

Abstract

Study design: This is a retrospective cohort study utilizing machine learning to predict postoperative complications in cervical spine metastases surgery.

Objectives: The main objective is to develop a machine learning model that accurately predicts complications following cervical spine metastases surgery.

Summary of background data: Cervical spine metastases surgery can enhance quality of life but carries a risk of complications influenced by various factors. Existing scoring systems may not include all predictive factors. Machine learning offers the potential for a more accurate predictive model by analyzing a broader range of variables.

Methods: Data from January 2012 to December 2020 were retrospectively collected from medical databases. Predictive models were developed using Gradient Boosting, Logistic Regression, and Decision Tree Classifier algorithms. Variables included patient demographics, disease characteristics, and laboratory investigations. SMOTE was used to balance the dataset, and the models were assessed using AUC, F1-score, precision, recall, and SHAP values.

Results: The study included 72 patients, with a 29.17% postoperative complication rate. The Gradient Boosting model had the best performance with an AUC of 0.94, indicating excellent predictive capability. Albumin level, platelet count, and tumor histology were identified as top predictors of complications.

Conclusions: The Gradient Boosting machine learning model showed superior performance in predicting postoperative complications in cervical spine metastases surgery. With continuous data updating and model training, machine learning can become a vital tool in clinical decision-making, potentially improving patient outcomes.

Level of evidence: Level III.