Construction of patient-level prediction model for in-hospital mortality in congenital heart disease surgery: regression and machine learning analysis
10.3760/cma.j.issn.1001-4497.2020.02.001
- VernacularTitle:基于回归分析和机器学习的先天性心脏病患儿术后住院死亡预测模型的建立—单中心12年大数据汇总分析
- Author:
Xiaoqi SONG
1
;
Xinwei DU
;
Shunmin WANG
;
Zhiwei XU
;
Zhaohui LU
Author Information
1. 上海交通大学医学院附属上海儿童医学中心心胸外科 200127
- From:
Chinese Journal of Thoracic and Cardiovascular Surgery
2020;36(2):65-73
- CountryChina
- Language:Chinese
-
Abstract:
Objective:Explore a predictive model for predicting postoperative hospital mortality in children with congenital heart disease.Methods:We retrospectively analyzed the characteristics of all children with congenital heart disease from January 1, 2006 to December 31, 2017 at Shanghai Children's Medical Center. Each procedure was assigned a complexity score based on Aristotle Score. In-hospital death prediction models including a procedure complexity score and patient-level risk factors were constructed using logistic regression analysis and machine learning methods. The predictive values of the models were tested by C-index. Results:A total of 24 693 patients underwent CHD operations were include in the study, there were 585 (2.4%) in-hospital deaths. In-hospital mortality for each procedure varies between 0 to 77.8%, with 32 procedures with 0 death record. The prediction model constructed using logistic regression found that in addition to the complexity score, other risk factors included age, height, operation history, echocardiography characteristics as well as certain laboratory test results (mainly coagulation factors) were significantly correlated with in-hospital death. Receiver operating curve analysis showed that prediction with only the complexity score resulted in an AUC of 0.654 (95% CI: 0.628-0, 681, P<0.01) while model containing patient-level risk factors had significant higher prediction value with AUC of 0.886 (95% CI: 0.868-0.904, P<0.01). Training with machine learning method resulted in a final prediction model with high prediction value ( AUC 0.889, with a sensitivity value for death prediction of 0.817). The key risk factors in machine learning model are in general agree with the logistic regression model however with subtle differences. Conclusion:Through combination of procedure complexity score with pre-operative patient-level factors, predictive model constructed using regression or machine learning method had high accuracy in in-hospital mortality prediction.
