Abstract: Objective To investigate the factors influencing vasovagal reactions (VVR) in whole blood donors, and to develop and validate a risk prediction model. This aim is to provide blood bank medical staff with a quantitative tool for VVR prediction, enabling early identification and precise intervention for high-risk individuals, thereby ensuring blood donation safety. Methods A total of 101 187 donation records from whole blood donors at the Guizhou Blood Center between January and December 2024 were included, encompassing demographic characteristics (gender, age, weight, ethnicity, etc.) and donation records (donation history, pre-donation pulse, blood pressure, planned collection volume, etc.). The data were randomly split into a training set (n=70 698, 70%) and a validation set (n=30 489, 30%). Univariate analysis and logistic regression were used to identify independent risk factors for VVR, and a prediction model was developed and presented as a nomogram. Model performance was evaluated using ROC curves and calibration curves to assess discriminative ability and prediction accuracy respectively. Results The results of logistic regression analysis showed that gender, age, body weight, education level, pre-collection volume, and blood donation history were independent influencing factors for vasovagal reactions during whole blood donation (P<0.05). The model exhibited good discrimination in the training set, with an area under the curve (AUC) of 0.764 (95%CI: 0.750-0.777). The optimal risk threshold determined based on the Youden index was 0.014, at which the sensitivity was 76.5% and the specificity was 64.1%. In the independent validation set, the model maintained robust performance, with an AUC of 0.772 (95%CI: 0.751-0.792), and at the same threshold, the sensitivity and specificity were 78.4% and 65.1%, respectively, confirming its generalizability. Calibration: The calibration curve showed that the predicted probabilities of the model were highly consistent with the actual occurrence rate. Quantitative analysis further confirmed that the expected calibration errors (ECE) of the training set and validation set were as low as 0 and 0.000 1, respectively, indicating that the model has excellent calibration accuracy. The model demonstrated characteristics of high sensitivity (78.4%), high negative predictive value (99.6%), and low positive predictive value (2.9%). Conclusion The risk prediction model for vasovagal responses in whole blood donation constructed in this study can reliably identify low-risk individuals for safe diversion and provides a scientific basis for blood station medical staff to develop personalized preventive measures.

Key words: Adverse reaction to blood donation, Vasovagal response, Risk prediction model, Influencing factors, Nomogram