Prediction of Mortality in Inpatients of Covid-19 Using Statistical and Artificial Intelligence Approaches: A Case Study in Sakarya

Abstract

The virus SARS-Cov-2 speared rapidly as pandemic disease and cause serious respiratory distress and death. The parameters such as D-dimer, fibrinogen, C-reactive protein, and serum ferritin become higher indicators for acute inflammation. It is vital to understand whether the disease will progress severely or result in death by using laboratory findings. In this way, a doctor concentrating on the patient can make faster and more accurate decisions about the prognosis of the disease and survey patients using artificial intelligence and statistical methods. Therefore, this study aims to determine the attributes associated with deteriorating prognosis by analyzing the laboratory findings of patients who are followed in Sakarya U. Training and Res. Hospital with Covid-19 by using artificial intelligence and statistical methods. More precisely the general aim of this study is to leverage artificial intelligence and statistical methods to identify specific laboratory markers associated with Covid-19 patient deterioration and mortality, enabling more accurate prognosis and clinical decision-making. Our results demonstrated that high fibrinogen, troponin, albumin, d-dimer, ferritin, intubation, and uric acid level were directly related to mortality in Covid-19 patients. The Random Forest technique yielded the highest average accuracy, precision, F1 values, sensitivity, and specificity (all above 0.99), making it the most successful. Additionally, the training-based models give higher accuracy results than statistical methods. Thus, these laboratory findings of Covid-19 patients and the training-based models may potentially provide doctors make easier and more accurate decisions about mortality and survey predicting.