Abstract: Objective  To develop and compare three machine learning prediction models for predicting vascular calcification in patients with maintenance hemodialysis (MHD).  Methods  A total of 300 MHD patients were enrolled and randomly divided into a training set (n=210) and a validation set (n=90) in a 7:3 ratio. Based on the presence or absence of vascular calcification, the training set was further categorized into a calcification group (n=124) and a non-calcification group (n=86). Logistic Regression (LR), Random Forest (RF), and Support Vector Machine (SVM) models were developed to predict vascular calcification. The predictive performance of these models was evaluated using the validation set.  Results  In the training set, the area under curve (AUC) values for the LR, RF, and SVM models were 0.835, 0.886, and 0.872, respectively. In the validation set, the AUC values were 0.823, 0.879, and 0.866, respectively. DeLong's test showed significant differences in the AUC values among the three models (Z=2.663, 2.751; P=0.003, 0.001). All three models demonstrated good goodness-of-fit (χ²=4.018, 4.661, 3.892; P=0.642, 0.887, 0.739).  Conclusion  The LR, RF, and SVM machine learning models demonstrated good predictive performance for vascular calcification in MHD patients, with the Random Forest model showing superior performance.

Key words: Maintenance hemodialysis, Vascular calcification, Machine learning, Predictive model