A prediction model for the death risk of aortic dissection based on machine learning and preoperative indicators
- VernacularTitle:基于机器学习与术前指标的主动脉夹层死亡风险预测模型构建
- Author:
Jianke RONG
1
;
Yeao WANG
1
;
Zhili WEI
1
;
Zunhui WAN
2
;
Yundan DENG
2
Author Information
1. The First Clinical Medical College of Lanzhou University, Lanzhou, 730000, P. R. China
2. Department of Cardiovascular Surgery, The First Hospital of Lanzhou University, Lanzhou, 730000, P. R. China
- Publication Type:Journal Article
- Keywords:
Aortic dissection;
machine learning;
preoperative indicators;
mortality risk prediction;
SHAP theory;
XGBoost algorithm
- From:
Chinese Journal of Clinical Thoracic and Cardiovascular Surgery
2026;33(10):1640-1649
- CountryChina
- Language:Chinese
-
Abstract:
Objective To construct a preoperative objective index-based model for predicting the mortality risk of aortic dissection, aiming to provide a quick risk assessment tool for primary healthcare. Methods The patients with aortic dissection from the Medical Information Mart for Intensive Care (MIMIC-Ⅳ) database between 2008 and 2019 were included. These patients were randomly divided into a training set, a validation set, and a test set at a ratio of 7:2:1. Based on the Akaike information criterion (AIC), forward regression was used to select the risk factors for patients with post-dissection mortality, and the XGBoost algorithm was employed to establish the prediction model. The SHAP (SHapley Additive exPlanations) theory was used for interpretive analysis. Results Out of the 271 patients of aortic dissection, 158 were males and 113 were females, with a median age of 70.3 (58.8, 79.5) years. The training set, validation set, and test set consisted of 189, 54, and 28 patients, respectively. During the follow-up period, 99 (36.5%) deaths occurred. Forward stepwise regression based on the AIC, clinical relevance assessment and literature review, was employed to identify 18 preoperative independent predictors. An XGBoost prediction model was constructed accordingly. After grid search optimization, the model demonstrated good discrimination in both the validation set [area under the curve (AUC)=0.681] and the test set (AUC=0.735). The SHAP analysis indicated that age (mean |SHAP|=0.081), activated partial thromboplastin time (mean |SHAP|=0.065), and red cell distribution width (mean |SHAP|=0.038) were the top three predictive contributors. Conclusion The aortic dissection mortality risk prediction model constructed based on the XGBoost algorithm can effectively predict the incidence of mortality outcomes. Characteristic indicators such as age, activated partial thromboplastin time, and red cell distribution width can assist clinicians in identifying high-risk patients, making triage referral decisions, and optimizing preoperative interventions within the golden time window, ultimately aiming to reduce the mortality rate of patients with aortic dissection.
