Researchers have developed a prediction model for spinal cord injury (SCI) in patients with spinal tuberculosis (STB) using multiple machine learning algorithms. The study, conducted across three different hospitals, aimed to create a practical tool to forecast the risk of SCI in STB patients.
A total of 373 patients with STB were analyzed at the First Affiliated Hospital of Guangxi Medical University over the past decade, with an additional 100 patients from other hospitals forming a prospective cohort for testing the prediction model. The research team collected clinical data, including patient demographics, laboratory results, and various scores related to spinal health. The patients were divided into two groups based on the severity of SCI – ASIA grades A, B, C, D were classified as the SCI group, while the rest were included in the No-SCI group, with further assessments conducted using ODI, JOA, and VAS scores.
The process of constructing the prediction model involved partitioning the dataset, assessing data imbalance, screening characteristic indicators through univariate analysis, and employing ten supervised machine learning classifiers. These classifiers were tuned using grid search to optimize their performance, evaluated using various metrics like ROC curves, AUC, PR curves, calibration curves, and DCA, with the most optimal model selected for further testing. Additional explanatory tools like the DALEX package and SHAP method were utilized to interpret the model and visualize its performance.
The results of the study demonstrated the successful development of an AI-powered prediction model for SCI in STB patients, providing valuable insights into risk assessment and patient outcomes. The model, deployed through an interactive web platform, shows promising potential to enhance clinical decision-making and improve patient care in the management of spinal tuberculosis.
The study received ethical approval from the participating hospitals and obtained informed consent from all participants, ensuring compliance with relevant guidelines and regulations. Moving forward, the research team aims to further validate and refine the prediction model to enhance its accuracy and practical utility in real-world clinical settings.
