key: cord-1020703-ll2nfs4a
authors: Khayat Kashani, Hamid Reza; Hajijafari, Mohamad; Khayat Kashani, Fereshte; Salimi, Sohrab
title: Predictive value of the preliminary findings in the severity of COVID‐19 disease and the effect on therapeutic approaches
date: 2021-02-17
journal: Dermatol Ther
DOI: 10.1111/dth.14828
sha: 7957706da52cead9fe5f99ab5d0071e98a015b8f
doc_id: 1020703
cord_uid: ll2nfs4a

In this retrospective multicenter case series study, the predictive value of initial findings of confirm COVID‐19 cases in determining outcome of the disease was assessed. Patients were divided into two groups based on the outcome: low risk (hospitalization in the infectious disease ward and discharge) and high risk (hospitalization in ICU or death). A total of 164 patients with positive PCR‐RT were enrolled in this study. About 36 patients (22%) were in the high‐risk group and 128 (78%) were in the low‐risk group. Results of statistical analysis revealed a significant relationship between age, fatigue, history of cerebrovascular disease, organ failure, white blood cells (WBC), neutrophil‐to‐lymphocyte ratio (NLR), and derived neutrophil‐to‐lymphocyte ratio (dNLR) with increased risk of disease. The artificial neural network (ANN) could predict the high‐risk group with an accuracy of 87.2%. Preliminary findings of COVID‐19 patients can be used in predicting their outcome and ANN can determine the outcome of patients with appropriate accuracy (87.2%). Most treatment in Covid‐19 are supportive and depend on the severity of the disease and its complications. The first step in treatment is to determine the severity of the disease. This study can improve the treatment of patients by predicting the severity of the disease using the initial finding of patients and improve the management of disease with differentiating high‐risk from low‐risk groups.

The novel viral pneumonia, known by the World Health Organization (WHO) as Coronavirus 2019 , began in late December 2019 as an epidemic in Wuhan, China. The virus causes fever, dry cough, dyspnea, weakness, and lymphopenia. It can also lead to pneumonia, SARS, and death in severe cases. 1, 2 COVID-19 is the third zoonotic virus that leads to the involvement of the respiratory system. Like its counterparts, SARS-CoV and MERS-CoV, it uses the angiotensin-converting enzyme 2 (ACE2) receptor to enter the cell, creating pneumonia and bilateral pulmonary infiltrates. The virus is very similar to the bat coronavirus, and although its exact origin is unknown, an influential theory suggests that it is transmitted from wild animals. 3 According to the experiences from SARS and MERS, its complete virus genome was identified in less than a month. 4, 5 COVID-19 is a single positive-stranded RNA virus coated with two lipid layers. The layer that covers the virus contains S protein. Sprotein is fused to cell membrane receptors called ACE2. RNA is injected into the cell to make virus proteins. Once the virus multiplies, the cell explodes, and the viruses are released and penetrate other cells. ACE2 receptors are more commonly found in tissues of the lungs, heart, kidneys, and adipose tissue; these organs consequently are more at higher risk of infection. 6, 7 COVID-19 virus is 10-20 times more likely to bind ACE2 than is SARS-CoV. These two viruses have up to 40% similarity of amino acids in S protein. 8, 9 The disease is asymptomatic in most cases but can be accompanied by mild to moderate or severe clinical symptoms. COVID-19 can cause acute respiratory syndrome and possibly death. [10] [11] [12] The virus has spread rapidly to other countries and caused a pandemic.

According to WHO, on May 31, 2020, more than 5 934 000 people were infected, and 367 000 people died in 225 countries and territories.

Many studies have been performed on clinical and laboratory symptoms and chest CT scans of COVID-19 patients. Given the importance of severe and critical types of the disease, the present study investigates not only the clinical, laboratory symptoms, and underlying diseases, but also the relationship between these findings and the course of the disease and patients' outcome. The studied variables are statistically analyzed, and then, the ANN efficiency in predicting the patients' outcome was explicitly investigated. The present study aimed to determine the predictive effect of demographic characteristics, clinical and laboratory symptoms, and underlying diseases upon the first referral, on the clinical outcome of the patients. Patients were divided into two groups. The first group included patients who were hospitalized in the infectious disease ward and were discharged without serious complications (low-risk group). The second group included patients who were transferred to the ICU upon referral or during hospitalization or death (high-risk group). In the case of the normal distribution, the Pearson correlation coefficient was applied, and the Spearman statistical correlation coefficient test was used in the case of the abnormal distribution. ROC curve and AUC were applied to assess the diagnostic value of the preliminary findings of the disease outcome. In the present research, both-sided P-value <.05 was considered statistically positive.

An artificial neural network (ANN) can be used in cases to predict the disease outcome by multiple variables. One of the most widely used ANNs is a multilayer perceptron (MLP). MLP involves a three-layer structure, including the input, the middle, and the output layers. The middle layer consists of one or more layers. Besides the input layer nodes, each node in the other layers uses a nonlinear activation function. MLP uses a supervised learning technique where a set of inputs and outputs is sent to the system, and the system tries to learn a function to convert the input to the output. Backpropagation is the training method in MLP. In this method, first, the network error is determined, and the weights are then corrected so that the mean squares of the difference between the actual output and desired output of the network are reduced.

In the present research, the training set for each case includes demographic characteristics and laboratory results of patients with positive COVID-19 test as the input and disease outcome as the output. Each person has one of the following two outcomes.

(a) Infectious disease ward admission and discharge (Class 0); and (b) ICU admission or death (Class 1). The MLP network is trained using the available data, and its classification accuracy of patients in two groups of Class 0 or Class 1 is evaluated.

The evaluation was performed using the k-fold cross-validation method. In this method, the whole data set is divided into k parts, and each part is considered as a fold. In this study, k = 10 was used. The 

A total of 164 patients with positive PCR-RT results were included in the study. As presented in Table 1 , 79 patients were female and 85 were male. The mean age of patients was 58.8 years (range of 20-86 years). The average prereferral duration of symptoms was 6.5 days. Of the total patients, 128 were hospitalized in the infectious disease ward and were discharged from the same ward, thus did not require ICU admission and intensive care (low-risk patients); 36 patients required ICU admission from the beginning or during hospitalization or died due to the disease complications (high-risk patients). The mortality rate was 10 individuals (6.1%).

The most common symptoms upon admission were cough and dyspnea (65.2%), followed by fever (55.5%) and anorexia (53.7%). Diarrhea (7.3%) was the rarest symptom in patients. A total of 52.3% of patients had comorbidities, being higher in the high-risk patient group. The most common comorbidity was hypertension (25%), followed by cardiovascular (13.4%) and respiratory disorders (11%).

Intergroup comparison concerning demographic characteristics showed that age was significantly higher in high-risk patients (P < .001). A significant relationship between the patient's sex and the disease severity was observed so that it was more severe in men (P < .004). Comparison of clinical symptoms upon referral in the two low-risk and high-risk patient groups showed a significant relationship between fatigue (P < .03) and the disease severity; however, other clinical symptoms were not significantly different in the two groups. Concerning comorbidities, cerebrovascular disease (.007) and organ failure (.002) were significantly related to increased risk of being in the high-risk group, and there was no significant relationship between the two groups concerning other types of underlying diseases. Further, results showed no relationship between the prereferral duration of the symptoms and the disease severity. 

The laboratory findings of the patients are shown in Table 2 . In the present study, in addition to the patient's tests, the blood markers were also calculated to examine their relationship with the degree of risk of the disease. Among blood cells, the mean leukocyte, neutrophil, and monocyte counts are higher, and platelet and lymphocyte counts are lower in the high-risk group.

Hemoglobin levels are slightly lower in high-risk patients, and the mean erythrocyte sedimentation rate (ESR) and C-reactive protein (CRP) levels are higher in high-risk patients. In this study, there were significant differences between high-risk and low-risk groups of patients in terms of WBC (P < .001), neutrophil (P < .001), NLR (P < .001), and dNLR (P < .001). Pearson correlation test was used to examine the relationship between blood markers and disease severity; the results revealed a positive relationship between NLR (r = .21, P < .007), dNLR (r = .22, P < .005), WBC (r = .31, P < .001), and neutrophil count (r = .31, P < .005) with the disease severity. Univariate logistic regression showed a significant difference between the two groups in terms of WBC (P < .004), neutrophil (P < .004), and CRP (P < .003). Multivariate logistic regression analysis of the relationship between laboratory findings and disease severity showed no significant relationship between blood markers as an independent predictor and disease severity. Figure 1 shows the ROC curve of the predictive value of neutrophil, WBC, NLR, and dNLR for the disease severity. Among these blood markers, WBC was the best predictor of disease severity (AUC = 0.646), followed by neutrophil and dNLR, respectively.

In the present work, the data set consisted of blood markers, age, and sex of 164 COVID-19 patients. The blood markers used included WBC, neutrophil, lymphocyte, monocyte, platelet, NLR, LMR, PLR, and dNLR1.

The outcome of each individual was classified into two categories: low-risk (class 0) or high-risk (class 1). A total of 128 patients were placed in Class 0 and 36 patients in Class 1. The patient data set was used for training and testing of MLP.

As shown in the confusion matrix of the artificial neural network (Figure 2) , the classification accuracy of the ANN is 87.2%, so that 143 cases are categorized correctly, and there are 21 classification errors.

As given in Table 3 

Despite its high contamination, COVID-19 is currently untreatable. Most cases are asymptomatic but can lead to clinical symptoms of varying degrees and acute respiratory failure and death. 6, 13, 14 In epidemic crises, a lack of medical equipment and ventilators can lead to an increase in the mortality rate. Clinical symptoms, laboratory findings, and underlying diseases have been identified as risk factors for the disease in various studies.

Accordingly, predicting the severity of the disease using the above indicators can be effective in admission indication and patient care levels. [15] [16] [17] The present study investigates the predictive value of disease severity using clinical symptoms, underlying diseases, and laboratory findings. The predictive value of the variables is examined in two sections: statistical analysis and artificial neural network. The statistical analysis revealed that there was no significant difference in clinical symptoms between two groups except fatigue. Among the underlying diseases, there was a difference between the two groups in terms of organ failure and cerebrovascular diseases. There was also a difference between low-risk and high-risk patient groups in terms of laboratory findings, that is, WBC, neutrophil, NLR, and dNLR and these lab tests were significantly higher in the high-risk patient group. The pre- and PLR levels as effective factors in disease severity; however, in the present research, this difference was not statistically significant. 19 The advantage of the present study is the use of ANN in predicting the disease severity using the patient admission data. This research results showed that MLP could determine the disease sever- 

Most treatment in COVID-19 are supportive and depend on the severity of the disease and its complications. The first step in treatment is to determine the severity of the disease. This study can improve the treatment of patients by predicting the severity of the disease using the initial finding of patients and improve the management of disease with differentiating high-risk from low-risk groups.

The higher WBC, neutrophil, NLR, and dNLR in the first lab test of hospitalized COVID-19 patients is associated with an increased risk of the disease. The artificial neural network could accurately predict the risk of disease (87.2%), and it is reliable (AUC = 0.736).

Clinical course and outcomes of patients with severe acute respiratory syndrome coronavirus 2 infection: a preliminary report of the first 28 patients from the Korean cohort study on COVID-19

Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus-infected pneumonia in Wuhan

COVID-19: an update on the epidemiological, clinical, preventive and therapeutic evidence and guidelines of integrative Chinese-Western medicine for the management of 2019 novel coronavirus disease

A SARS-like cluster of circulating bat coronaviruses shows potential for human emergence

A new coronavirus associated with human respiratory disease in China

ACE2: from vasopeptidase to SARS virus receptor

Epidemiological and clinical characteristics of 99 cases of 2019 novel coronavirus pneumonia in Wuhan, China: a descriptive study

Evolution of the novel coronavirus from the ongoing Wuhan outbreak and modeling of its spike protein for risk of human transmission

Clinical characteristics of 140 patients infected with SARS-CoV-2 in Wuhan

Clinical characteristics of coronavirus disease 2019 in China

The epidemiological characteristics of an outbreak of 2019 novel coronavirus diseases (COVID-19) in China

A multiscale and comparative model for receptor binding of 2019 novel coronavirus and the implications of its life cycle in host cells

Epidemiology and clinical characteristics of COVID-19

Review of the clinical characteristics of coronavirus disease 2019 (COVID-19)

Can we predict the severity of COVID-19 with a routine blood test?

Clinical and immunologic features in severe and moderate coronavirus disease 2019

Predictive value of the preliminary findings in the severity of COVID-19 disease and the effect on therapeutic approaches

The authors declare no conflicts of interest.

All authors have contributed significantly in different area of the research and drafting the manuscript. In addition, none of the authors listed on the manuscript are employed by a government agency of Iran that has a primary function other than research and/or education.

The data that support the findings of this study are available from the corresponding author upon reasonable request.

https://orcid.org/0000-0002-9464-2703