Development and Application of a Machine Learning-Based Predictive Model for Carotid Restenosis After Interventional Surgery in Elderly Ischemic Stroke Patients.

Objective: This study aims to develop a machine learning-based risk prediction model for carotid restenosis in elderly ischemic stroke patients after interventional surgery.

Methods: We collected clinical data from 371 elderly ischemic stroke patients who underwent carotid interventional surgery between January 2020 and December 2023, as training dataset. Various machine learning methods were applied to analyze and compare the predictive performance of different models. In addition, data from 75 cases collected between January and June 2024 was as a validation set to assess model applicability.

Results: Six factors influencing carotid restenosis were identified: homocysteine (Hcy), platelet count (PLT), platelet distribution width (PDW), mean platelet volume (MPV), Interleukin-6 (IL-6), and C-reactive protein (CRP). Machine learning models were developed, with the Gradient Boosting Machine showing the best performance (AUROC=0.969). Other models included support vector machine (AUROC=0.962), logistic regression (AUROC=0.945), decision tree (AUROC=0.885), and extreme gradient boosting (AUROC=0.753). The GBM model's predictive variable ranking was Hcy, IL-6, CRP, PDW, PLT, and MPV. In the validation set, the GBM model demonstrated excellent performance, with an AUC 0.939, sensitivity 0.909, specificity 0.969, accuracy 0.960, negative predictive value 0.984, and positive predictive value 0.833.

Conclusion: Our research showed that compared with other machine learning algorithms, the GBM model demonstrates the best accuracy and stability in predicting the risk of carotid restenosis after interventional surgery in elderly ischemic stroke patients, and it has high clinical application value.