Early Detection of Mortality in COVID-19 Patients Through Laboratory Findings with Factor Analysis and Artificial Neural Networks


ÖĞÜTCÜ S., İNAL M. M., çelikhası c., YILDIZ U., DOĞAN N. Ö., PEKDEMİR M.

ROMANIAN JOURNAL OF INFORMATION SCIENCE AND TECHNOLOGY, vol.25, no.3-4, pp.290-302, 2022 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 25 Issue: 3-4
  • Publication Date: 2022
  • Journal Name: ROMANIAN JOURNAL OF INFORMATION SCIENCE AND TECHNOLOGY
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.290-302
  • Keywords: Artificial neural networks, COVID-19, factor analysis, triage, RISK-FACTORS, RECEPTOR
  • Kocaeli University Affiliated: Yes

Abstract

In this paper, some biochemical findings of patients who applied to Kocaeli University Faculty of Medicine Emergency Service with suspicion of COVID-19 are examined. The common characteristics of the cases regarding mortality status are analyzed via factor analysis (FA). Following the FA, blood parameters related to the severity of the cases are determined. Finally, a multi-layered artificial neural network (ANN) is trained with these parameters. This paper proposes a method that helps early detection of severe cases and determination of non-risk group vaccination priority. Thus, the main contribution is the creation of a decision-support system to start advanced medical support as soon as possible. The data set consists of 105 patients with 19 different input parameters. After FA, 7 parameters are found relevant to one-month mortality. These are HB, AST, BUN, LDH, pH, HCO3 and LAC. The chi-square value was 1252.9552, the p value for the significance level of 0.05 was close to zero (7.3696x10(-156)). An ANN is accurately trained based on this subset of the data. The most successful model of ANN's training and testing errors as a root sum squared estimate of error (RSSE) are 0.1958 and 0.2402, respectively. This ANN model can be queried for patient data with determined parameters. This paper shows that the early detection of patients who can have the severe or fatal disease can be determined regarding COVID-19. The proposed method can be used to determine vaccination priority, for early intervention to expected severe course of treatment, and medical analysis and analytics of unknown diseases via their outcomes, enriched with numerical laboratory test results.