Using multiple machine learning algorithms to predict spinal cord injury in patients with cervical spondylosis: a multicenter study

Abstract

Degenerative cervical spondylosis, a chronic and progressive condition, has a considerable impact on global health. Spinal cord injury, a severe sequela of this disease, can result from this disease. Machine learning (ML) has emerged as a valuable tool for medical data analysis, effectively predicting disease outcomes. A multicenter study involving retrospective analysis of data from 737 patients diagnosed with cervical spondylosis was performed. On the basis of clinical data obtained from three hospitals, a predictive model was developed and demonstrated using multiple ML algorithms. In accordance with the exclusion criteria, a training set consisting of 385 samples, a test set of 129 samples, and an external validation set of 149 samples were acquired. Through univariate analysis and LASSO regression, 11 core predictive factors were identified. Results: Among the 10 trained machine learning models, the random forest model exhibited superior performance, as evidenced by elevated AUC values and accuracy across both the training and testing sets. The incidence of cervical spondylosis is evidently high, with a rising trend noted among younger individuals. Early prediction of spinal cord injury in these patients is paramount. Machine learning was utilized in this study to ascertain key predictive factors and develop a model capable of supporting clinical decision-making. The random forest model, developed from extensive analysis of clinical and imaging features across multiple hospitals, was subjected to cross-validation for accuracy and stability. This model can assist surgeons in the development of precise, individualized treatment approaches, with the aim of enhancing therapeutic effectiveness and minimizing unnecessary medical procedures.

Ähnliche Arbeiten