| 摘 要: 为了及早发现重症监护室中的急性肾损伤高危患者,为其提供适当的护理,实现医疗资源的合理利用,研究建立因果贝叶斯网络模型进行急性肾损伤高危患者死亡风险预测。从重症监护医学信息市场(Medical Information Mart for Intensive Care Ⅲ, MIMIC-Ⅲ)数据库中筛选了25个研究变量和3 870条患者数据,使用因果发现算法进行特征降维。通过NO TEARS算法构建因果图并建立因果贝叶斯网络进行实验,通过机器学习算法验证重要特征的合理性,并对网络结构进行因果效应估计,模型具有最高的受试者工作特征曲线下面积(Area Under the Receiver Operating Characteristic, AUROC)分数,为81.7%,优于逻辑回归(Logistic Regression, LR)、随机森林(Random Forest, RF)和极端梯度提升树(eXtreme Gradient Boosting, XGBoost)。此外,模型的重要特征预测能力在各种建模中都很稳健,构建的因果贝叶斯网络具有更好的预测效果并具备良好的解释能力。 |
| 关键词: 急性肾损伤;因果贝叶斯网络;因果发现;死亡风险预测 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 南京医科大学附属江宁医院医工融合实验室资助项目(JNYYZXKY202107);国家自然科学基金项目:基于贝叶斯网络预测ICU术后患者死亡风险的方法研究(82072228);科技部国家重点研发计划“主动健康和老龄化科技应对”重点专项(2020YFC2008700) |
|
| Predicting Mortality Risk of AKI Patients Based on Causal Bayesian Network |
|
XU Naiyue1, ZHOU Liang2, LIU Kun1, ZHOU Mengyu1
|
(1.School of Health Science and Engineering, University of Shanghai f or Science and Technology, Shanghai 200093, China;
2.Biomedical Engineering Fusion Laboratory, Jiangning Hospital Af f iliated to Nanjing Medical University, Nanjing 211100, China)
xunaiyue21@163.com; wenzhou6@sjtu.edu.cn; lkun11111@163.com; zhou_meng_yu_66@163.com
|
| Abstract: In order to promptly identify high-risk patients for Acute Kidney Injury (AKI) in the Intensive Care Unit (ICU), provide appropriate care, and achieve rational utilization of medical resources, this paper proposes to establish a causal Bayesian network model for predicting mortality risk in high-risk AKI patients. 25 study variables and 3 870 patient records are selected from the MIMIC-Ⅲ (Medical Information Mart for Intensive Care Ⅲ) database, and causal discovery algorithm is used for feature dimension reduction. The NO TEARS algorithm is employed to construct a causal graph and establish a causal Bayesian network for experimentation. Machine learning algorithm is utilized to validate the rationality of important features, and causal effect estimation is performed on the network structure. The model achieves the highest Area Under the Receiver Operating Characteristic (AUROC) score of 81.7% , which is superior to Logistic Regression (LR), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost). Additionally, the predictive ability of important features in the model remains robust across various modeling scenarios. The proposed causal Bayesian network has better prediction performance and good interpretability. |
| Keywords: AKI; causal Bayesian network; causal discovery; mortality risk prediction |
