关键词: 机采献血者, 血管迷走神经反应, 预测模型, LASSO回归, 机器学习

Abstract: Objective To develop a risk prediction model for vasovagal reaction (VVR) in apheresis blood donors, thereby supporting evidence-based strategies for donor safety as well as adverse event prevention and control. Methods We retrospectively enrolled 58 031 apheresis platelet donors from Ningbo Central Blood Station (2021—2025), including 155 VVR cases with a positive rate of 0.27%. Demographic characteristics, physical examination findings, hematological indices and apheresis parameters were collected. Least Absolute Shrinkage and Selection Operator (LASSO) regression was used for feature screening, and a multivariate logistic regression model was established. The dataset was divided into training and test sets at a 7:3 ratio by stratified sampling, with five-fold cross-validation for model verification. Model performance was assessed using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and other metrics. Results From 29 candidate variables, 21 core predictors were retained. The model achieved an AUC of 0.840 in the training set and 0.800 in the test set, with an accuracy of 72.60%, sensitivity of 68.09%, specificity of 72.61%, precision of 0.67%, and F1-score of 0.013 2. Logistic regression analysis identified apheresis volume grouping, hemoglobin level, and pulse rate as independent risk factors, whereas male sex, platelet count, history of multiple apheresis donations, and body weight were independent protective factors. Although extracorporeal circulation volume and anticoagulant dosage exhibited statistically protective effects, these were recognized as statistical artifacts arising from early termination of apheresis following VVR occurrence. Conclusion The proposed model displays good discriminative ability and stability. It could identify high-risk VVR candidates based on pre-collection baseline indicators before blood and facilitate individualized preventive interventions, supporting its clinical utility in blood donation settings.

Key words: Apheresis blood donors, Vasovagal reaction, Prediction model, LASSO regression, Machine learning