Premium
Using machine learning to predict stroke‐associated pneumonia in Chinese acute ischaemic stroke patients
Author(s) -
Li X.,
Wu M.,
Sun C.,
Zhao Z.,
Wang F.,
Zheng X.,
Ge W.,
Zhou J.,
Zou J.
Publication year - 2020
Publication title -
european journal of neurology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.881
H-Index - 124
eISSN - 1468-1331
pISSN - 1351-5101
DOI - 10.1111/ene.14295
Subject(s) - medicine , receiver operating characteristic , logistic regression , stroke (engine) , modified rankin scale , random forest , ischaemic stroke , pneumonia , area under the curve , machine learning , artificial intelligence , ischemic stroke , computer science , ischemia , mechanical engineering , engineering
Background and purpose Stroke‐associated pneumonia (SAP) is a common, severe but preventable complication after acute ischaemic stroke (AIS). Early identification of patients at high risk of SAP is especially necessary. However, previous prediction models have not been widely used in clinical practice. Thus, we aimed to develop a model to predict SAP in Chinese AIS patients using machine learning (ML) methods. Methods Acute ischaemic stroke patients were prospectively collected at the National Advanced Stroke Center of Nanjing First Hospital (China) between September 2016 and November 2019, and the data were randomly subdivided into a training set and a testing set. With the training set, five ML models (logistic regression with regulation, support vector machine, random forest classifier, extreme gradient boosting (XGBoost) and fully connected deep neural network) were developed. These models were assessed by the area under the curve of receiver operating characteristic on the testing set. Our models were also compared with pre‐stroke Independence (modified Rankin Scale), Sex, Age, National Institutes of Health Stroke Scale (ISAN) and Pneumonia Prediction (PNA) scores. Results A total of 3160 AIS patients were eventually included in this retrospective study. Among the five ML models, the XGBoost model performed best. The area under the curve of the XGBoost model on the testing set was 0.841 (sensitivity, 81.0%; specificity, 73.3%). It also achieved significantly better performance than ISAN and PNA scores. Conclusions Our study demonstrated that the XGBoost model with six common variables can predict SAP in Chinese AIS patients more optimally than ISAN and PNA scores.