Survival Analysis of COVID-19 Patients With Symptoms Information by Machine Learning Algorithms

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 42
  • Download : 0
In this study, a survival analysis of the time to death caused by coronavirus disease 2019 is presented. The analysis of a dataset from the East Asian region with a focus on data from the Philippines revealed that the hazard of time to death was associated with the symptoms and background variables of patients. Machine learning algorithms, i.e., dimensionality reduction and boosting, were used along with conventional Cox regression. Machine learning algorithms solved the diverging problem observed when using traditional Cox regression and improved performance by maximizing the concordance index (C-index). Logistic principal component analysis for dimensionality reduction was significantly efficient in addressing the collinearity problem. In addition, to address the nonlinear pattern, a higher C-index was achieved using extreme gradient boosting (XGBoost). The results of the analysis showed that the symptoms were statistically significant for the hazard rate. Among the symptoms, respiratory and pneumonia symptoms resulted in the highest hazard level, which can help in the preliminary identification of high-risk patients. Among various background variables, the influence of age, chronic disease, and their interaction were identified as significant. The use of XGBoost revealed that the hazards were minimized during middle age and increased for younger and older people without any chronic diseases, with only the elderly having a higher risk of chronic disease. These results imply that patients with respiratory and pneumonia symptoms or older patients should be given medical attention.
Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
Issue Date
2022
Language
English
Article Type
Article
Citation

IEEE ACCESS, v.10, pp.62282 - 62291

ISSN
2169-3536
DOI
10.1109/access.2022.3182350
URI
http://hdl.handle.net/10203/297079
Appears in Collection
EE-Journal Papers(저널논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0