Can initial chest CT scan predict status and clinical outcomes of COVID-19 infection? A retrospective cohort study

Background We aimed to investigate the association of initial chest CT scan findings with status and adverse outcomes of COVID-19 (including ICU admission, mortality, and disease severity). This retrospective cohort study was performed in three hospitals in Babol, northern Iran, between February and March 2020. Cases were confirmed by real-time polymerase chain reaction (RT-PCR). Clinical and paraclinical data of the patients were collected from their medical records. CT severity score (CSS) was calculated by a senior radiologist. Disease severity was determined based on the World Health Organization criteria. Results In total, 742 patients were included, of whom 451 (60.8%) were males and 291 (39.2%) were females. The mean age was 56.59 ± 14.88 years old. Also, 523 (70.5%) were RT-PCR-positive. Ground glass opacity was directly associated with RT-PCR positivity (odds ratio [OR] = 2.07). Also, RT-PCR-positive cases had significantly a higher CSS than RT-PCR-negative cases (p = 0.037). In patients confirmed with COVID-19, peribronchovascular distribution of lesions, number of zones involved, and CSS were associated with increased risk of ICU admission (OR = 2.93, OR = 2.10, and OR = 1.14, respectively), mortality (OR = 2.30, OR = 1.35, and OR=1.08, respectively), severe disease (OR = 2.06, OR = 1.68, and OR = 1.10, respectively), and critical disease (OR = 4.62, OR = 3.21, and OR = 1.23, respectively). Also, patients who had consolidation were at a higher risk of severe disease compared with those who did not (OR = 4.94). Conclusion Initial chest CT scan can predict COVID-19 positivity, ICU admission, mortality, and disease severity, specifically through CSS.

A number of studies have stated that findings of chest CT scan are potentially predictive for the clinical outcomes of the patients, suggesting CT scan as a helpful tool in diagnosis and management of the patients with COVID-19 [8,9]. However, a notable limitation of those studies is a small sample size, potentially leading to decreased reliability of evidence. Also, to our knowledge, the number of studies that used the scoring system to quantify lung involvement is limited. Moreover, not enough information exists on the predictive ability of CT scan results for status of COVID-19 infection. To overcome these issues, we aimed to conduct a study to investigate the association of CT scan findings with status and adverse outcomes of COVID-19 (including ICU admission, mortality, and disease severity).

Locations and patients
This retrospective cohort study was performed in three hospitals affiliated to Babol University of Medical Sciences, including Rohani, Shahid Beheshti, and Yahyanejad hospitals, between February and March 2020. Babol was one of the first cities in north of Iran with confirmed cases. We initially included the individuals presenting with suspicious symptoms (fever, respiratory symptoms, such as cough, dyspnea, and sputum) who were referred to the emergency unit and underwent necessary clinical and paraclinical assessments for COVID-19. Chest CT scan was performed on all suspected individuals after triage. For RT-PCR testing, trained technicians collected nasopharyngeal swab specimens and sent them to the laboratories of Babol health center. RT-PCR was done for all triaged patients. The time gap between the nasopharyngeal swab sampling and CT scan was less than 12 h.
Patients with chest CT scans suggestive of COVID-19 and at least one of the following criteria were admitted to the hospitals, according to the national COVID-19 protocol: hypoxia (O2 saturation ≤ 92%), tachypnea (respiratory rate ≥ 22), tachycardia (pulse rate > 100), or hypotension (systolic blood pressure 100 mmHg or less). The following data were extracted from the patients' medical records by a trained research team: demographic information (such as sex and age), comorbidities (such as cardiovascular diseases [CVDs], asthma, chronic obstructive pulmonary disease [COPD], and diabetes). Cases with incomplete information about comorbidities and/ or RT-PCR results, as well as those declined to participate in the study, were excluded from further investigations. We categorized clinically the COVID-19 pneumonia into different disease severities (moderate, severe, critical) based on the definition by World Health Organization (WHO) [10]. Patients with mild disease were not admitted as per the national COVID-19 protocol.

Imaging collection and analysis
The patients underwent non-enhanced 16-detectorrow CT scan during deep inspiration breath-hold in the supine position (Siemens SOMATOM Emotion 16, Siemens Healthcare, Med Image Systems, Germany). The scanning parameters were as follows: tube voltage, 100 KV for patients with BMI ≤ 30 were not necessary. The CT scans were evaluated by a single radiologist with an experience of more than 15 years (R.M.), who was blinded to the patients' status. The following imaging characteristics were recorded: ground glass opacity, consolidation, reticular pattern, lesions distribution (peribronchovascular or peripheral), side of lung involvement, crazy paving, pleural effusion, number of lung zones involved, cavity, and tree-in-bud pattern.
The CT severity score (CSS) for each patient was calculated based on the percentage of lung zones involvement [11]. In this regard, right and left lungs were divided into three (upper, middle, and lower) and two (upper and lower) zones, respectively. The scoring system was as follows: score 0 representing no involvement, score 1 representing < 5% involvement, score 2 representing 5-25% involvement, score 3 representing 26-50% involvement, score 4 representing 51-75% involvement, score 5 representing > 75% involvement. Finally, sum of the scores yielded the total CSS, ranging from 0 to 25.

Data analysis
The statistical analyses were performed by SPSS software. The obtained data initially underwent descriptive analyses. To assess normality of the data, Kolmogorov-Smirnov test was used. For comparing parametric and non-parametric continuous data between the groups, independent t test and Mann-Whitney test were used, respectively. We performed chi-squared test and logistic regression analysis to investigate the association of baseline information and imaging findings of the patients with study outcomes (COVID-19 status and adverse disease outcomes). The factors with significant association were entered into the multivariable analysis. The results were presented as odds ratio (OR) as well as 95% confidence interval (CI). We also calculated the area under the curve (AUC) to estimate the predictive ability of CT scan features for the study outcomes.
A p value less than 0.05 was considered statistically significant.

Basic information
Initially, 829 patients were admitted to the hospitals, of whom 87 cases were excluded from the study due to meeting exclusion criteria. Totally, 742 patients were included for further assessments, of whom 451 (60.8%) were males and 291 (39.2%) were females. Imaging findings and COVID-19 status Table 1 shows distribution of the baseline information of the patients by COVID-19 status. Also, the findings of chest CT scan according to COVID-19 status are represented in Table 2. Due to the small number of cavity (n = 4) and tree-in-bud pattern (n = 3) found in CT scan, they were excluded from further analyses. Analyses demonstrated that ground glass and consolidation were significantly higher in COVID-19-positive patients than in COVID-19 negative patients (OR = 2.92 and OR = 2.40, respectively). After adjustment for imaging findings, ground

Imaging findings and outcomes of COVID-19 patients
Of 523 patients with COVID-19, 317 (60.6%) were males and others were females. The mean age was 56.55 ± 14.75 years old. In Table 3, the characteristics of the RT-PCR-positive patients are represented according to ICU admission, mortality, and disease severity. The most frequent abnormal imaging findings observed in COVID-19 patients were ground glass (94.1%) and consolidation (91.0%). Table 4 exhibits the distribution of different CT scan findings by outcomes of COVID-19 patients. In this regard, ground glass opacity, consolidation, bilateral involvement, and peribronchovascular distribution were associated with ICU admission and disease severity. Also, consolidation and peribronchovascular distribution were significantly higher in nonsurvivors compared with survivors. The median number of zones involved and CSS were significantly higher in patients with the unfavorable outcomes than those without. Figures 1, 2, and 3 depict the CT scans of the patients with moderate, severe, and critical COVID-19 pneumonia, respectively. According to logistic regression model, peribronchovascular distribution of lesions, number of zones involved, and CSS were associated with increased risk of ICU admission (OR = 2.93, OR = 2.10, and OR = 1.14, respectively; AUC = 60.5%, AUC = 66.6%, and AUC = 71.6%, respectively), mortality (OR = 2.30, OR = 1.35, and OR = 1.08, respectively; Fig. 1 A 77-year-old male patient presented with fever, chills, and dry cough, who was admitted to the regular ward and was discharged on day 10 of admission. Peripheral multilobar ground glass opacities (red arrow), parenchymal bands (black arrow), and crazy paving (blue arrow) are observed in the CT scan. The total CT severity score was calculated as 15, and the disease severity was moderate

Discussion
In the present study, we investigated the potential predictive ability of initial chest CT scan findings for COVID-19 status and adverse clinical outcomes. It was found that cases with ground glass opacity had a two-fold increased likelihood of COVID-19 infection compared with those without. This CT finding had also acceptable sensitivity and accuracy. In the study by Chen et al. [12], consolidation was reported to be predictive for COVID-19 infection, which was inconsistent with our results. A superiority of the present study to the mentioned survey is a larger sample size. Concerning the diagnostic performance of CT scan, a recent meta-analysis reported sensitivity and specificity of 87% and 46%, respectively. Despite a good diagnostic sensitivity, this metaanalysis recommended to use RT-PCR besides CT scan to reach the most accurate result [7]. We found that CSS could be predictive for COVID-19 infection, that is, a higher total CSS is directly correlated to increased odds of the disease. Similar to our results, the study by Al-Mosawe et al. [13] showed that the probability of RT-PCR positivity increases with increase in CT score. To the best of our knowledge, limited number of studies have explored the association between CSS and COVID-19 status. Our results propose that CSS can be used in combination with clinical examination for the Fig. 2 A 34-year-old male patient presented with fever, chills, myalgia, and dyspnea, who was admitted to ICU for 6 days and was discharged on day 21 of admission. Peripheral and peribronchovascular multilobar ground glass opacities (black arrow), and parenchymal bands (red arrow) are observed in the CT scan. The total CT severity score was calculated as 23, and the disease severity was severe initial management of the suspected COVID-19 cases while waiting for RT-PCR results. It would be useful for clinicians and radiologists to reach a consensus on a threshold for CSS to better identify the COVID-19 cases. With respect to the adverse outcomes of COVID-19, we assessed whether the initial CT scan findings inform ICU admission, mortality, and disease severity. In this regard, we found that peribronchovascular distribution of lesions, number of lung zones involved, and total CSS were associated with increased risk of the unfavorable outcomes. Also, consolidation was demonstrated to predict severe COVID-19 disease. Lei et al. [14] showed that a higher CT score was associated with an increased odds of mortality, which was in agreement with our results. On the other hand, number of lung zones involved did not predict mortality, which was not consistent with our findings. In the study by Liu et al. [8], which used the same criteria as the present study used for the disease severity (WHO), number of lung lobes involved and total CT score were directly correlated to disease severity. In other study, it was stated that the odds of adverse outcome (need for mechanical ventilation or mortality) is four times higher in patients with more than four lung zones involved than in those without [15]. Auger et al. [16] reported that ground glass, crazy paving, and consolidation did not have a significant association either with invasive endotracheal ventilation or mortality. On the other hand, number of lung zones involved was associated with invasive endotracheal ventilation, but not with death.
As observed, there are conflicting results between studies on the predictive ability of CT scan findings for clinical outcomes of COVID-19. However, CSS is apparently able to predict the prognosis of the patients with COVID-19. It is suggested to approve a threshold for CSS to discriminate high-risk from low-risk patients. In the present study, we considered a threshold for the unfavorable outcomes and witnessed relatively strong associations between these two study outcomes and CSS classification.
A limitation of this study was that we only used the initial CT scan without repetition. Thus, it is suggested to perform longitudinal studies to Fig. 3 A 62-year-old male patient presented with fever, chills, myalgia, headache, and sputum discharge, who was admitted to ICU for 2 days and died. Extensive multilobar ground glass opacity and consolidation (red arrow), and parenchymal bands (black arrow) are observed in the CT scan, suggesting acute respiratory distress syndrome. The total CT severity score was calculated as 25, and the disease severity was critical prospectively re-evaluate the patients with more details for testing generalizability. Also, considering that a single radiologist reviewed the CT scans, it is suggested that at least two senior radiologists contribute in reviewing the CT findings in the further studies.

Conclusion
According to the results, initial CT severity scores could predict positive COVID-19 status, ICU admission, mortality, and disease severity. Moreover, peribronchovascular distribution of the lesions and number of lung zones involved predicted adverse outcomes of COVID-19 infection. Having consolidation was also directly associated with severe disease.

Availability of data and materials
The datasets during and/or analyzed during the current study are available from the corresponding author on a reasonable request.

Declarations
Ethics approval and consent to participate The details of this study were initially explained to the patients, and then, the written informed consents were taken from all of them. The study protocol was approved by the ethics committee of Babol University of Medical Sciences (code: IR.MUBABOL.REC.1399.239). The patients' information was kept confidential.

Consent for publication
The written informed consent was obtained from all research participants after a full explanation of the study.