Centroid Regression for Preoperative Risk Assessment of Acute Type A Aortic Dissection Based on Multivariate Clinical Data.

Background/Objectives: Acute type A aortic dissection (ATAAD) has high preoperative mortality, and an interpretable multivariable model based on clinically accessible data is crucial for clinical risk stratification. Methods: The data for this study were obtained from West China Hospital, Sichuan University. Centroid regression was used to construct the predictive model, with logistic regression, classification and regression tree, explainable boosting machine and extreme gradient boosting as the reference. Variables were screened by iterative selection, the literature review and clinical experience. Model performance was evaluated by accuracy, sensitivity, precision, Youden's index, AUROC and AUPRC. Results: The vital signs and tests of 361 ATAAD patients during the first 24 h of their first admission were included in the final analysis. Centroid regression outperformed logistic regression, achieving accuracy (90.7% vs. 81.5%), sensitivity (0.813 vs. 0.741), specificity (0.983 vs. 0.900), Youden's index (0.796 vs. 0.641), AUROC (area under the receiver operating characteristic curve, 0.953 vs. 0.843) and AUPRC (area under the precision-recall curve, 0.978 vs. 0.863) in the test set. It revealed that the use of α-blocker (the weights w = -1.20) and hydrochlorothiazide (w = -1.20), clinical features like dyspnea (w = -0.94), chest pain (w = 0.91) and lactate dehydrogenase (w = -0.95) were variables that had the greatest impact on model prediction. Conclusions: The centroid regression model not only has relatively high predictive performance and interpretability but also can be easily implemented in hospital systems to provide a practical and cost-effective tool for ATAAD preoperative risk stratification.