Introduction of a Radiologic Severity Index for the 2019 Novel Corona Virus (COVID-19)

Background: Given the limited number of beds in intensive care units, establishing a system that can predict the outcome in COVID19 positive patients based on imaging plays an important role in using resources e�ciently. Therefor this study was conducted to design an optimal scoring system related to the severity of COVID19 cases for distinguishing severe from non-severe patients. Materials and Methods: In this cross-sectional retrospective study, 82 patients with a de�nite diagnosis of COVID-19 infection, who had at least one chest CT scan in hospital course were enrolled. To assess the severity of pulmonary parenchymal involvement, we semi-quantitatively evaluated the extent and nature of abnormalities. The area of lung involvement was scored in three levels based on a 0-4 grading scale. Also, we established a 4-point scoring system for de�ning the nature of lung abnormalities. The two scores were multiplied by each other. A �nal radiologic severity score was determined after adding together the scores of all levels. Result: Of all cases, fty-three (64.6%) were male with an average age of age 53.75. Among the patients in our study, 7 (8.5%) had severe disease and the mortality rate was 7.2%. The mean (±standard deviation) of the radiologic severity score was 34.3(±18.4) in the severe group and 11.3(±11.4) in the non-sever group. (P-value <0.05). Also, we found a signi�cant reverse relationship between our severity score and O 2 saturation (P-value <0.05). Conclusion: The radiologic severity score demonstrated a signi�cant correlation with the patients' mortality and severity of illness in COVID-19 patients.


Introduction
On 11th March 2020, the World Health Organization (WHO) announced Corona-virus disease (COVID- 19) as a pandemic (1).Due to the fast global spread, until April 2nd, 2020, COVID-19 is affecting 203 countries and territories all over the world with over 944,000 con rmed cases and over 47,000 deaths (2).
Our knowledge about COVID-19 is limited.Chen.et al reported an incubation period of up to 14 days (3).The clinical manifestations vary from mild symptoms including fever, cough, sore throat and myalgia to more severe presentations such as pulmonary edema, acute respiratory distress syndrome (ARDS) and multiple organ failure (4)(5)(6)(7).
COVID-19 has become a huge burden and a global health emergency with over 20% critical patients and mortality around 3% (8).Given the limited number of beds in intensive care units (9,10), establishing a system that can predict the outcome in COVID19 positive patients based on para clinical data can play an important role in using resources e ciently.
Like previous corona-viruses, including Middle East respiratory syndrome coronavirus (MERS) in 2012 and severe acute respiratory syndrome caused by corona-virus (SARS-COV) in 2002, COVID-19 can also lead to ARDS as a main cause of death (11,12).Due to the short interval between onset of symptoms and the development of ARDS in COVID-19 pneumonia, early diagnosis is essential for better management (13).
Chest CT plays a key role in the early diagnosis of infected patients.Typical ndings including groundglass opacity, consolidation, and linear opacities are the most common ndings in chest CT of COVID-19 positive patients (14,15).Also, one of the main methods for distinguishing sever and critical cases from milder cases is based upon the evaluation of disease progression by chest computed tomography (CT) in patients.
This study aims to compare the clinical condition and pulmonary involvement in CT scan of con rmed COVID-19 patients to develop a scoring system related to the severity of this disease for distinguishing severe from non-severe patients.

Method Study design and participants
In this cross-sectional retrospective study, patients with a de nite diagnosis of COVID-19 infection were recruited at the two main referral hospitals for COVID-19 from February through March 2020.
Inclusion criteria were as follows: A) having a proper history; B) Positive real-time PCR of SARS-CoV-2; C) having at least one non-contrast chest CT scan We studied the demographic, clinical, laboratory and radiologic ndings of the patients.The data from the patients were kept con dential through codes.

Computerized Tomography Scans
Non-contrast CT scan was performed in a supine position during full inspiration.Scanning was extending from thoracic inlet to the upper abdomen.The technical parameter included: 8 slice scanner (GE Medical Systems, Milwaukee, WI, and USA), 120Kvp, thickness of 1.25-2mm, 1.25 mm interval.

Imaging evaluation
The primary chest CT scan was reviewed independently by four radiologists', three of whom (SS, PI, SJ) were board-certi ed in general diagnostic radiology with approximately eight to fteen years of practice experience and one in training (YE with four years of practice experience).The radiologists were blind to the primary impression, clinical symptoms, and the patient's outcome and undertook the classi cation and categorization of scans.The CT ndings were arbitrarily considered to be one of these four categories as ground-glass opacity (GGO), consolidation, crazy paving, and nodular opacities.
The de nitions of these ndings were determined based on the Fleischer Society guidelines (16).Consolidation and GGO were de ned as hazy areas of increased attenuation with and without obscuration of the underlying vasculature, respectively.GGO with intra-lobular lines were de ned as crazy paving and nodular opacities were identi ed when focal round opacities either solid or GGOs with a diameter less than 3 cm.
To assess the severity of pulmonary parenchymal involvement, we quanti ed the extent and nature of abnormalities.For this purpose, the area of lung involvement was scored in the axial CT images based on the method described previously by Ooi et al in 2004 for the severe acute respiratory syndrome (SARS) (17).In this method each lung was assessed in three levels; upper (above the carina), middle (below the carina up to the upper limit of the inferior pulmonary vein), and lower (below the inferior pulmonary vein).Also, the right and left lung were evaluated separately and summed up to conclude the nal score of each level.
The nature of abnormalities can also determine the severity of lung involvement (18,19).In this regard, we arbitrarily established a 4-point scoring system for de ning the pattern of lung abnormalities in CT scans that summarized in Table 2.
Then, the two scores (extent and nature of abnormality) were multiplied by each other (Figure1).After adding together the scores of all 6 levels (3 levels in each side), a nal radiologic severity score (RSS) for parenchymal involvement with values ranging from 0 to 96 was determined for each patient (Figure 2-4).
Eventually, the nal score was determined by consensus between all four radiologists.
Eventually, this severity score was compared between sever and non-sever patients according to American Thoracic Society guidelines for community-acquired pneumonia (20).

Statistical analysis
The collected data was summarized as means (±SD), and categorical data are presented as the count (percentage).Unpaired Student's t-test, chi-square test, or Fisher's exact test was used to compare the RSS of COVID-19 patients as appropriate.A P. value of less than 0.05 considered indicating statistical signi cance.All the statistical analyses were performed by the Statistical Package for Social Sciences (SPSS Inc., Chicago, Illinois, USA) version 26.0.

Result
During the study period, 113 patients with con rmed COVID-19 PCR were registered at our clinical sites, which 82 cases with at least one chest CT scan were enrolled for the purpose of our study.Among the selected patients, fty-three (64.6%) were male and twenty-nine (35.4%) were female.The mean ±SD of age was 53.75± 1.56 [Min = 20, Max = 89].Only 20% of cases reported a clear history of contact with infected COVID-19 cases.
Seven patients were admitted in ICU with a mean age of 59.29±8.5 years and 6 patients died during hospitalization with a mean age of 57±1.9 years.Among the patients in our study, 7 (8.5%) had severe disease and the mortality rate was 7.2% (6 out of 82 patients).There was no signi cant difference between the severe and non-severe groups with respect to the patients' age.(P.value = 0.77) Underlying comorbidity diseases were collected from the patients' history, including hypertension, cardiovascular disease, diabetes, and chronic obstructive pulmonary disease.Based on our results 36% of the non-severe patients and 57.1% of the severe patients had underlying comorbidity diseases, however, no signi cant differences were observed between the two groups (P=0.417).
The mean interval between onset of symptoms and rst chest CT scans was 6.7 days (range: 1-22 days).
The mean (±standard deviation) of RSS was 34.3 (±18.4) in the severe group and 11.3 (±11.4) in the nonsevere group (P-value <0.05).There was also a signi cant correlation between RSS of the rst CT scan and both the clinically severe group (<0.05,Z=-3.5) and the mortality group (<0.05,Z=-3.5).
Receiver operating characteristic (ROC) analysis was used to access the relation between RSS and the patients' disease severity and mortality.The area under the curve stated 0.900 (95% con dence interval: 0.819, 0.983) for the prediction of severe cases and 0.784 (interval: 0.527, 1.000) for predication of death (Figure 5).Also, considering area under ROC curve, we suggested 13.5 as optimal cut-point in radiologic severity score with speci city of 72% and sensitivity of 100% for the prediction of sever group (95% con dence interval= 0.819-0.983;P-value=0.00) and 30.5 with speci city of 93% and sensitivity 0f 66% for the prediction of mortality group (95% con dence interval= 0.527-1.000;P-value: 0.021).
Statistically, we found a signi cant reverse relationship between our severity score and O 2 saturation (O 2 sat), which was measured by pulse oximeter during the rst examination (P-value <0.05) (Figure 6).However, no signi cant relation was found between the RSS and CRP or lymphocyte count (Table 3).

Discussion
A variety of clinical manifestations can be caused by coronaviruses family ranging from asymptomatic to severe illness.Previously, SARS and MERS which come from the same family as COVID-19 had caused epidemics throughout several countries in the last decades.However, the novel corona 2019 virus has caused pandemic unlike others and infected more people due to its high contagious characteristics (13,14,(21)(22)(23).
Given the highly contagious nature of the disease and limited resources, COVID19 infection poses a huge impact on the health care systems.As a result, early diagnosis is critical for disease treatment and control.
In the current study, the mean age of patients was 53.7 years and the majority of them were males (64%).Among them, 8.5% were severely ill and the mortality rate among hospitalized patients was 7.3%.Among the initial CT scan ndings, the most frequents was GGO that is consistent with other studies (24,25).
Our RSS demonstrated a signi cant correlation with the patients' mortality and severity of illness.Consistent with prior studies, we suggest that CT could be used to evaluate the severity and prognosis of the disease (26).We also achieved a cut-off point of 30.5 with sensitivity and speci city of 66% and 93% respectively for mortality and also a cut-off of 13.5 with sensitivity and speci city 100% and 72% respectively for the severity of illness.However, due to the small sample size of our study and the skepticism of the outcome of the patients, further studies on larger samples are needed for application in clinical practice.One of the advantages of our index is in the consideration of both the nature and extent of lung involvement in a more feasible manner.
In addition, we found a signi cant reverse relation between our score and O 2 -saturation that can be a helpful point in practice.29) also reported a scoring system in their studies which included only the extension of abnormality in each lung lobes to assess changes in pulmonary involvement during the course of COVID-19 pneumonia, however, the correlation between score and clinical condition was not evaluated.
Another study used the mean pulmonary in ammation index (PII) to assess the severity of involvement, focusing on the distribution and size of abnormality.They reported a signi cant relationship between PII and clinical symptoms with lymphocyte counts and CRP (26).
Yuan et al. (30) described the extension of abnormality using a modi ed version of Feng et al. (31) that was published in 2014 for avian in uenza H7N9 pneumonia.They used a 3-point scale: 3 as consolidation, 2 as ground-glass attenuation, and 1 as normal attenuation.The results showed that the median CT score of the mortality group was higher compared to their survival group.However, assigned cut-off points of their study achieved a higher sensitivity but lower speci city (85.6% and 84.5% versus 66% and 93%, respectively) for evaluating the patients' mortality.As mentioned earlier, in many cases consolidation and GGO cannot be easily differentiated which we resolved this issue with an intermediate score in our indexing score.
Yang et al. also suggested a 3-point scoring scale but in 20 regions of the lung achieved a sensitivity and speci city of 83.3% and 94% in predicting the severity of the illness (32).There were some limitations in our study.While highly accurate in our study population, our cut-offs were achieved based on a small sample size, with a smaller number of severe/critical patients.For more reliable accurate cut points, further studies with larger sample sizes are warranted.Also, the fact that most of our patients were already hospitalized at the time of the CT scan makes the results less applicable to the unadmitted COVID-19 patients.In the end, none of the patients had a lung biopsy or autopsy to re ect the histopathological changes.

Conclusion
This study provided a straightforward semi-quantitative method for evaluating and managing COVID-19 patients based on the initial chest CT scan.Such approaches could have importance in triage and assessing the patients in high-volume centers, especially during outbreaks when resources and other diagnostic tests such as PCR are limited.Although a signi cant correlation with the prognosis and severity of the disease was achieved, further studies in higher volume settings and comparing with clinical features are needed for applying this index in clinical practice.

Declarations Acknowledgments
The study was the subject of the MD dissertation of Yasaman Emami.

Statement of data access and integrity
The authors declare that they had full access to all of the data in this study and the authors take complete responsibility for the integrity of the data and the accuracy of the data analysis.

Authors Contribution
Mehrzad Lot , and Mohsen Moghadami designed the study.Pouya Iranpour, Yasaman Emami, and Seyed Hamed Jafari reviewed and interpreted the radiological images.Alireza Mirahmadizadeh carried out the statistical analysis.Keivan Ranjbar, Amirhossein Erfani, Mehrdad Emami, and Reza Shahriarirad carried out the manuscript preparation.All authors proofread the nal version of the manuscript.

Ethical statement
Tables Table 1.system based on the extent of lung involvement in each region of the patients CT-scan percentage of lung involvement Score 0 0 Li et al. (27) designed a scoring system that ranged from 0 to 25 based on the percentage of each lung lobe involvement and reported the sensitivity and speci city of 80.0% and 82.8% respectively based on a cut-off of 7. Similarly, Bernheim et al. (28) and Pan et al. (

Figure 1 Schematic 2 A 3 A
Figure 1

Figure 4 A
Figure 4

Table 2 .
Scoring system based on the nature of involvement in each region of the patients CT-scan

Table 3 .
Pearson Correlation between radiological severity score and CRP, O 2 saturation (O 2_ Sat) and lymphocyte count.