Prognostic utility of the chest computed tomography severity score for the requirement of mechanical ventilation and mortality in hospitalized patients with COVID-19

Purpose To correlate the chest computed tomography severity score (CT-SS) with the need for mechanical ventilation and mortality in hospitalized patients with COVID-19. Materials and methods The chest CT images of 224 inpatients with COVID-19, confirmed by reverse transcriptase-polymerase chain reaction (RT-PCR), were retrospectively reviewed from April 1 to 25, 2020, in a tertiary health care center. We calculated the CT-SS (dividing each lung into 20 segments and assigning scores of 0, 1, and 2 due to opacification involving 0%, <50%, and ≥50% of each region for a global range of 0–40 points, including both lungs), and collected clinical data. The receiver operating characteristic curve and Youden Index analysis was performed to calculate the CT-SS threshold and accuracy for classification for risk of mortality or MV requirement. Results 136 men and 88 women were recruited, with an age range of 23–91 years and a mean of 50.17 years; 79 met the MV criteria, and 53 were nonsurvivors. The optimal threshold was >27.5 points for mortality (area under ROC curve >0.96), with a sensitivity of 93% and specificity of 87%, and >25.5 points for the need for MV (area under ROC curve >0.94), with a sensitivity of 90% and specificity of 89%. The Kaplan-Meier curves show a significant difference in mortality by the CT-SS threshold (Log Rank p < 0.001). Conclusions In our cohort of hospitalized patients with COVID-19, the CT-SS accurately discriminates the need for MV and mortality risk. In conjunction with clinical status and laboratory data, the CT-SS may be a useful imaging tool that could be included in establishing a prognosis for this population.


Introduction
The current SARS-CoV-2 pandemic has caused global uncertainty at an unprecedented scale, having infected more people and taken more lives compared with the combined effects of SARS and MERS [1][2][3]. It has spread over a relatively short time and has caused healthcare services saturation, with exceeded hospital and intensive care unit bed capacity in some countries worldwide [4].
The definitive diagnosis of COVID-19 relies on real-time reverse transcriptase-polymerase chain reaction (RT-PCR) [5,6,7], but this has a high false-negative rate [8]. In several situations, the availability and processing time of the test makes the diagnosis inopportune. In these situations, radiology plays an important role. Noncontrast chest computed tomography (CT) has a higher pooled sensitivity (93%) [5] for COVID-19 compared with that of RT-PCR (71%-98%) [5] but lower specificity [9]. It is a valuable additional tool for identifying patients with moderate to severe COVID-19 and those with worsening respiratory status [10,11].
Compared with chest radiography, thin-section chest CT has a higher sensitivity and can detect subtle lung parenchyma changes in the early stages of the disease [7]. According to the Fleischer Society consensus statement [10], thorax imaging is indicated in patients with mild features of COVID-19 and have risk factors for progressive or worsening respiratory status, as well as for medical triage of patients with moderate to severe clinical features in resource-constrained environments. In patients in such clinical scenarios, chest CT allows the assessment of the virus' pulmonary damage [5].
To quickly and objectively identify patients with severe diseases, especially in institutions with limited healthcare resources, Yang et al. [7] developed a practical and easy semiquantitative scoring method. The CT severity score (CT-SS) is based on the amount of lung Fig. 1. Representative axial non-contrast chest CT images from different patients to show the calculation of the CT-SS in the right posterobasal segment (S10) (A-C) and left superior segment (S6) (D-F). In images A and D, the absence of ground glass opacities or consolidation correspond to 0 points. In images B and E, less than 50% compromised segment corresponds to 1 point. In images C and F, diffuse ground glass opacities in more than 50% of the studied segments correspond to 2 points. opacification in 20 regions. A higher CT-SS has been observed in patients with severe disease than in mild cases. A CT-SS threshold of 19.5 was reported to have 83.3% sensitivity and 94% specificity in identifying patients with severe COVID-19 [7].
However, no objective imaging measurement enables the prediction of the need for mechanical ventilation (MV) and mortality among hospitalized patients with COVID-19 among different populations (i.e., Hispanic). This study intended to evaluate the discriminating accuracy of a practical CT-SS score for the need for MV and mortality in hospitalized COVID-19 patients in a tertiary referral center in Mexico City.

Patients and groups
This observational retrospective study protocol followed the Declaration of Helsinki's ethical principles. The research and ethics committees of the National Institute of Nutrition and Medical Sciences approved this study (Reference number: 3414). Due to the retrolective methodology, the Informed consent was waived, although all data and imaging information was anonymized.
We retrospectively reviewed the chest CT examinations of 224 consecutive inpatients (in order of hospitalization) with COVID-19 confirmed by RT-PCR from April 1 to 25, 2020. According to our hospital protocol at that time, all hospitalized patients with suspected COVID-19 pneumonia underwent non-contrast chest CT. Hospitalized patients with clinical and radiologic manifestations and a negative RT-PCR result (43 patients) were excluded. The patients were classified according to outcomes as survivors or nonsurvivors and whether they met MV's criteria. The criteria for MV were: 1) no improvement in respiratory distress despite the use of a nonrebreather mask with high-flow oxygen (10-15 L/min), with a respiratory rate higher than 30 per minute, the partial pressure of oxygen (PaO 2 ) < 60 mmHg, and a PaO 2 /fraction of inspired oxygen (FiO 2 ) < 150 with respiratory distress while on noninvasive respiratory support; and 2) any patient with PaO 2 /FiO 2 < 100.

Chest CT severity score calculation
As proposed by Yang et al. [7], the CT-SS for the assessment of COVID-19 entails the division of the lungs into 20 regions; specifically, the apicoposterior segment of the left upper lobe is divided into the apical and posterior regions, and the anteromedial basal segment of the left lower lobe is divided into medial basal and anterior basal regions. For each patient, CT-SS of 0, 1, and 2 were assigned for parenchymal opacification involving 0%, <50%, and ≥50% of each region, respectively (global range of 0-40 points, including both lungs) as is illustrated in Fig. 1(A-F). Only the initial CT scan was reviewed in patients with multiple chest CT scans (28 patients).
The CT scans, and CT-SS were independently reviewed and calculated for outcomes correlation in a standard diagnostic system workstation by a general radiologist with more than ten years of experience and blinded to the clinical and biochemical data. We assessed the interobserver agreement by including the scores of 25 CTs from two other independent and blinded imaging reviewers: a general radiologist with five years of experience and a radiologist in training who made determinations on his own and had recent experience in calculating the CT-SS of other 25 scans that were not included in the present study, training under supervision the radiologist with more than ten years of experience.

Chest CT scans
Chest CT scans were performed on a 64-detector CT scanner (Revolution EVO, General Electric). All CT examinations were performed with the patient in a supine position, with images acquired from the lung apex to the adrenal glands during a single inspiratory breath. The CT scan parameters were as follows: 120 kV, 350 mAs; rotation time of 0.4 s; the pitch of 1.53:1 (mm/rot); intersection space of 5 mm; multiplanar reconstructions with a slice thickness of 1.5 mm; and a sharp convolution kernel. Thin-section CT scans were read at a window width of 1000-2000 HU and level of − 700 to − 500.

Statistical analysis
A comparative analysis between the survivors and nonsurvivors was performed along with the descriptive analysis (dispersion and central tendency statistics). The hypotheses for comparing numerical variables were tested using Student's t-test, and categorical variables were compared using the chi-square test. Due to a lack of Gaussian distribution of CT-SS within mortality and MV groups (Shapiro-Wilk test p = 0.03), we assess these with a Mann-Whitney U test. The analysis was performed using STATA SE 14.1 software and SPSS software package version 20, considering a significant two-tailed p-value of <0.05.
We used the intraclass correlation coefficient (ICC) to estimate the interrater variability and absolute agreement on the CT-SS among the three observers. We tested the diagnostic yield of the global CT-SS using receiver operating characteristic (ROC) curve analysis to describe the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NVP), positive likelihood ratio (LR+), negative likelihood ratio (LR− ), odds ratio (OR), and area under the ROC curve (AUC) with 95% confidence interval (CI). Days of hospitalization and relevant time intervals between the onset of symptoms, RT-PCR realization, CT acquisition, and the outcomes were reported. Furthermore, we built a Kaplan-Meier curve for mortality regarding the CT-SS for visual assessment and Log Rank regression.
We explored the classification threshold using the Youden index, which was the optimal maximum difference between the true and false positive ratios. Notably, the original identification study reported a predetermined threshold of 19.5 [7].

Clinical and demographic variables associated with mortality
The 224 patients included (Fig. 2) had a mean age of 50.17 years (SD, 13.85) and an age range of 23-91 years; 60.7% (136/224) were men, and 39.29% (88/224) were women. Mortality was observed in 23% (53/224) of the patients and was significantly higher in those with older age than those with younger age (57 ± 14 years vs. 48 ± 13 years, p < 0.001); in men than in women (75.5% vs. 56.1%, p = 0.012); in those who met the criteria for MV than in those who did not (96.2% vs. 16.4%, p < 0.001); in those with diabetes than in those without diabetes (38.5% vs. 21.1%, p = 0.01); and in those with organ transplant than in those without organ transplant (11.3% vs. 3.5%, p = 0.027). Table 1 compares the demographic and clinical features between survivors and nonsurvivors.    Table 2 and Fig. 5(A and B)].
The Kaplan-Meier curves show a significant difference (Log Rank p < 0.001) in mortality between groups separated by the proposed CT-SS threshold during 50 days of the observation period (Fig. 7); the median days of survival in the group with CT-SS > 27.5 was calculated at 22.9 (95% CI 18.4-27.5).

Discussion
We reported a study on the CT-SS classification accuracy, a semiquantitative imaging tool for the clinical outcomes of the Moreover, we found that the proposed CT-SS thresholds of >25.5 points for MV and >27.5 points for mortality had excellent classification performance, with respective sensitivities and specificities of 89.9% and 89% for MV, and 92.5% and 86.5% for mortality, plus robust likelihood ratios.
Several epidemiological studies reported variable rates of invasive MV [12], ranging from 2.3% [12] to 33.1% [13] among hospitalized patients. In our study, 35% of patients met the criteria for MV, and 29% had access to it. The higher mortality rate in this retrospective study than in previous reports [12] reflected the specific status of a cohort of hospitalized patients in a referral center. The mortality rate reported in our study (24%) was similar to the 21% mortality rate reported by Richardson et al. among hospitalized patients with COVID-19 [14]. Even though the OR might overestimate the risk ratio, it has been shown stability among study designs. In our population, the OR was a significant parameter in estimating the mortality and MV risk based on the CT-SS. Determining the need for MV will ultimately be clinical; however, a CT-SS >25.5 points may classify patients at high risk for respiratory deterioration  and mortality. Consistent with Yang et al. [7], we revealed an excellent agreement among readers, suggesting that CT-SS may be highly reproducible in radiology departments and diagnostic centers. However, our study did not include an expert on chest radiology. Although this could be regarded as a limitation, we believed it was our study's strength because an expert chest radiologist is only sometimes

Table 3
Time intervals between Onset of Symptoms, CT, RT-PCR, Hospitalization, and outcomes.  present in the interpretation room. The interpretation of chest CT using the CT-SS can be carried out even by radiologists in training. Additionally, a semiquantitative score could be used for patient follow-up and objectively assessing changes. Furthermore, we validated the predetermined threshold of >19.5 in our cohort and found near-perfect sensitivities (100%) for MV and mortality but low specificities of 52.4% and 44.4%, respectively. Moreover, the AUCs for the predetermined cutoff point of 0.76 for MV and 0.72 for mortality were substantially less than those found for our proposed thresholds (>0.89), as well as OR and likelihood ratios.
Yuan et al. [15] retrospectively studied 27 consecutive patients and proposed a different severity score ranging from 0 to 72 and a threshold of 24.5 as a mortality predictor (85% sensitivity and 86% specificity). In a multicenter prospective study [16], the CT-SS was suggested for the triage and management decision of COVID-19 patients since it showed a significant association with the odds of hospital admission, ICU admission, and 30-day mortality. Also, Salvatore et al. found that CT visual score and software-based quantification of parenchymal involvement were independent predictors of outcomes in patients with COVID-19 pneumonia, proving that CT-SS identifies patients who need more aggressive treatment [17]. To our knowledge, no clinical trial has compared the different available scores.
In a systematic review and meta-analysis pooling 7106 patients, Zakariaee et al. found that the odds of mortality for COVID-19 patients could be accurately predicted using an optimal CT-SS cutoff in visual scores and that higher CT-SS correlates with higher odds of mortality [18]. Another recent meta-analysis showed that higher CT-SS strongly correlates with lung fibrosis [19].
This study had the limitation of having a retrospective cohort design in a single tertiary referral public institution. We only admitted hospitalized patients, and outpatients were excluded. Another limitation was that the subjects needed to be followed up after discharge, hindering information of mortality outcomes after hospital discharge. Multicenter studies on prospective cohorts are necessary for external validation.
In conclusion, in our cohort of hospitalized patients with COVID-19, the CT-SS accurately discriminates the need for MV and mortality. In conjunction with clinical status and biochemical data, the CT-SS may be a useful imaging tool that could be included in these patients' prognostication.

Author contribution statement
Yukiyoshi Kimura, M.D.; Cesar Nicolas Cristancho-Rojas, M.D., M.Sc.: Conceived and designed the experiments; Performed the experiments; Analyzed and interpreted the data; Contributed reagents, materials, analysis tools or data; Wrote the paper.

Data availability statement
Data will be made available on request.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.