Probability Threshold Optimization for Classification of COVID-19 Patients with Higher Mortality Risk: A Case Study from the North-Eastern Region in North Macedonia

Mogilevska Gruevska, Dragana and Gruevski, Ilija and Boshevska, Golubinka and Gaber, Stevan (2024) Probability Threshold Optimization for Classification of COVID-19 Patients with Higher Mortality Risk: A Case Study from the North-Eastern Region in North Macedonia. European Journal of Medical and Health Research, 2 (5). ISSN 2786-8524

[thumbnail of 300-Article Text-508-1-10-20241002.pdf] Text
300-Article Text-508-1-10-20241002.pdf

Download (588kB)
[thumbnail of journal] Text
journal

Download (20kB)

Abstract

Research goal: The goal of this research is to identify, compare and
demonstrate some available methodological approaches which are
sufficient to draw the optimal probability threshold, particularly in the case
of classification of infected patients with increased risk of dying from COVID19. The presented methodologies generate identical results if the purpose of
classification is to maximize the prognosis accuracy from the point of
sensitivity. Sample: As part of the whole population, the sample counts 1013
patients from the north-eastern region of the Republic of North Macedonia.
Methodology: The general methodological frame used to calculate and
forecast the probabilities of death outcome from COVID-19 is the binary
logistic regression. In extension, we applied the rules of maximum sum and
maximum product as well as the so-called Youden Index for the purpose of
optimization of the probability threshold. The principals of the ROC curve in
addition with the Index of Union was also helpful for the same purpose.
Prognosis accuracy was evaluated through the status of patient according to
the rules of “golden standard” in which sensitivity, specificity and the general
accuracy of prognosis play a crucial role. Results: Accordingly, the results
from the research indicate that the optimal probability threshold or “cut-off”
point that provides maximal accuracy, particularly from the perspective of
sensitivity in the prognosis is 0,1. In that point, the coefficient of sensitivity
(the percentage of true positively predicted death cases in respect to all
death cases from the sample) is measured 85,71%. Conclusion: The applied
methodological approaches offer scientifically sound foundations in the
context of mortality risk evaluation and classification of COVID-19 patients.
Then targeted patients will be subject of precaution with strict measures,
protocols and more aggressive treatment in order to minimize the chances
of death outcome

Item Type: Article
Subjects: Medical and Health Sciences > Clinical medicine
Divisions: Faculty of Medical Science
Depositing User: Ilija Gruevski
Date Deposited: 03 Oct 2024 08:24
Last Modified: 03 Oct 2024 08:24
URI: https://eprints.ugd.edu.mk/id/eprint/34769

Actions (login required)

View Item View Item