Chest CT Scan Features to Predict COVID-19 Patients' Outcome and Survival

Background Providing efficient care for infectious coronavirus disease 2019 (COVID-19) patients requires an accurate and accessible tool to medically optimize medical resource allocation to high-risk patients. Purpose To assess the predictive value of on-admission chest CT characteristics to estimate COVID-19 patients' outcome and survival time. Materials and Methods Using a case-control design, we included all laboratory-confirmed COVID-19 patients who were deceased, from June to September 2020, in a tertiary-referral-collegiate hospital and had on-admission chest CT as the case group. The patients who did not die and were equivalent in terms of demographics and other clinical features to cases were considered as the control (survivors) group. The equivalency evaluation was performed by a fellowship-trained radiologist and an expert radiologist. Pulmonary involvement (PI) was scored (0–25) using a semiquantitative scoring tool. The PI density index was calculated by dividing the total PI score by the number of involved lung lobes. All imaging parameters were compared between case and control group members. Survival time was recorded for the case group. All demographic, clinical, and imaging variables were included in the survival analyses. Results After evaluating 384 cases, a total of 186 patients (93 in each group) were admitted to the studied setting, consisting of 126 (67.7%) male patients with a mean age of 60.4 ± 13.6 years. The PI score and PI density index in the case vs. the control group were on average 8.9 ± 4.5 vs. 10.7 ± 4.4 (p value: 0.001) and 2.0 ± 0.7 vs. 2.6 ± 0.8 (p value: 0.01), respectively. Axial distribution (p value: 0.01), cardiomegaly (p value: 0.005), pleural effusion (p value: 0.001), and pericardial effusion (p value: 0.04) were mostly observed in deceased patients. Our survival analyses demonstrated that PI score ≥ 10 (p value: 0.02) and PI density index ≥ 2.2 (p value: 0.03) were significantly associated with a lower survival rate. Conclusion On-admission chest CT features, particularly PI score and PI density index, are potential great tools to predict the patient's clinical outcome.


Introduction
Coronavirus disease 2019 , caused by severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2), was officially announced as a pandemic by the World Health Organization on March 11, 2020 [1]. Even though most of the patients experience mild symptoms, some may develop a severe type of disease and it may progress to acute respiratory distress syndrome (ARDS) [2]. e global deathto-case ratio is estimated to be 3.5% [3]. However, it varies geographically probably due to the local preventive measures and medical resources. erefore, we need to improve the admitted patients' initial triage not only to optimize the allocation of the medical resources to high-risk patients and minimize the mortality rate accordingly but also to more accurately and reliably predict the outcome of the patients.
Chest computed tomography (CT) scan, as the conventional and relatively accessible imaging modality for pneumonia diagnosis and follow-up, is confirmed to have high diagnostic and prognostic values in the current outbreak of COVID-19 [4,5]. Chest CT findings mainly consist of ground-glass opacities (GGOs), multifocal patchy consolidation, and interstitial changes with a peripheral distribution [6][7][8]. Efforts have been made to underpin the predictive factors for mortality [9]. Among all, the main factors consist of age, underlying disease (i.e., immunocompromised patients and preexisting cardiovascular and pulmonary disorders), laboratory findings (i.e., D-dimer level, neutrophil-to-lymphocyte ratio (NLR), and lymphocyte count), and most recently, imaging features [10][11][12][13]. However, there is still no consensus on the factor with the highest predictive value. Besides, almost all previous studies were retrospective cross-sectional and nonsurvivors comprised only a small population [14,15], which further limit the application of the findings.
In this regard, we aimed to utilize the vastly used modality-chest CT scan, and we hypothesize that on-admission chest CT findings could serve as a potential deterministic factor for risk stratification in hospitalized COVID-19 patients. In order to evaluate the proposed notion, we conducted a case-control study and assessed the adjusted predictive value of CT scan in terms of mortality rate for admitted COVID-19 patients.

Study Design and Participants.
is case-control study was reviewed and approved by the Institutional Review Board of our university. Given the retrospective design of the study and anonymous use of medical records, informed consent requirement was waived by the ethics committee of our institute (IR.TUMS.VCR.REC.1399.054). e current case-control study was carried out in a referral tertiary university hospital from June to September 2020. is study evaluated individuals with the following conditions: (a) all hospitalized COVID-19 patients in whom COVID-19 diagnosis was confirmed by positive real-time reverse transcription-polymerase chain reaction (rRT-PCR) assay on nasopharyngeal or oropharyngeal swap or endotracheal aspirate samples; (b) age equal or greater than 18 years; and (c) a definite outcome of either death or hospital discharge. Case patients consisted of patients who met the inclusion criteria and were deceased. Control patients included discharged patients with equivalent demographic (age and sex) and clinical (underlying diseases and laboratory findings) features. e equivalency evaluation was performed by a fellowship-trained radiologist and an expert epidemiologist. e flow diagram of the study is presented in Figure 1. Of note, admission, discharge criteria, and treatment of all patients were based on the national protocol of COVID-19.

Image Acquisition.
All chest CT images were acquired at the time of admission, in the supine position, with full inspiration with no contrast injection. Examinations were performed on either the Siemens Somatom Emotion (16 slices, Erlangen, Germany) or the Lightspeed 64-detector CT (GE Healthcare, Milwaukee, USA) MDCT scanner. e imaging parameters were set at 5-6 mm section thickness, beam collimation of 0.6-2 mm, 120 kVp tube voltage, tube current of 150-250 mAs, tube rotation speed of 0.75 seconds, and gantry rotation time of 0.5-0.75 s, reconstructed with a mediastinum B20f smooth kernel and a lung B70f sharp kernel (Siemens Healthineers, Erlangen, Germany); coronal and sagittal multiplanar reconstructions were also available with a reconstructed slice thickness of 1.2 mm.

Image Interpretation.
Two fellowship-trained diagnostic imaging radiologists, with respective 9 and 13 years of experience in thoracic radiology and blinded to patients' outcomes, independently interpreted chest CT images' findings. All CT images were reviewed on both lung-and mediastinal-window settings. e intraclass correlation coefficient (ICC) was calculated to assess inter-rater reliability. If ICC was less than 0.8, any disagreement in image interpretation for the case was discussed until resolved. If ICC was greater than or equal to 0.8, the value reported by the radiologist with higher experience was recorded.
Chest CT scan findings were recorded according to the Fleischner Society glossary and published literature on viral pneumonia [16]. Chest CT scan features included the following: (a) predominant pattern, ground-glass opacification/ opacity (GGO) (Figure 2

Pulmonary Involvement (PI) Scoring System.
To assess PI, a semiquantitative scoring tool was proposed and used [17]. All five lung lobes (right upper lobe (RUL), right middle lobe (RML), right lower lobe (RLL), left upper lobe (LUL), and left lower lobe (LLL)) were visually reviewed for GGO and consolidation. en, a score from 0 to 5 was assigned to each lobe according to involvement percentage (0: no involvement; 1: ≤5%; 2: 6-25%; 3: 26-50%, 4: 51-75%; and 5: ≥76%). e total PI score was calculated as the sum of all five lobes' scores. e PI score ranged from 0 (no involvement) to 25 (maximum involvement). Finally, the PI density index was calculated by dividing the total PI score by the number of involved lobes.

Statistical Analysis.
We performed the analyses in SPSS for Windows ver. 18 (Chicago, IL, USA). Descriptive data are presented as mean with standard deviations (SD) for   All p values less than 0.05 were considered statistically significant. We used multivariate logistic regression to predict mortality probability with all imaging parameters as the independent variables. As previously stated, the statistically significant threshold was considered as a p value less than 0.05. To define optimum cutoff values for PI score and PI density index in outcome prediction, receiver-operating characteristic (ROC) curves were drawn and Youden's J index [18] was calculated. e area under the ROC curve (AUC) was considered as the indicator for ROC analysis efficacy.
To determine the impact of any independent variable on survival time, we implemented univariate Cox regressions (in nonsurvivors) considering all demographic, clinical, and imaging parameters as covariates in the model. Multivariate backward Cox regression was performed to find the final model in terms of which variables to include. Kaplan-Meier survival analysis was performed to calculate survival, and the log-rank test was used to compare the survival distribution of two subgroups of interest.

Study Population Characteristics.
After evaluating 384 cases, a total of 186 patients (93 in each group) were admitted to the studied setting, consisting of 126 (67.7%) male patients with a mean age of 60.4 ± 13.6 years ( Figure 1). e most common underlying diseases were HTN (38.7%) and DM (34.9%) ( Table 1).
Two groups were almost equivalent for demographic (age and sex) and clinical (underlying diseases and laboratory findings) variables. Table 1 further illustrates the homogeneity of cases and controls in the aforementioned characteristics. All ICCs for inter-rater reliability were >0.8 for all imaging parameters.
We exploited backward multivariate logistic regressions that included death as the outcome of interest and all imaging parameters with significant association in Table 2

Survival Analysis.
e survival analyses were limited to nonsurvived patients. e mean survival was 4.1 ± 3.5 days, and the median survival was four days. e Kaplan-Meier survival function for death is illustrated in Figure 4. To determine the impact of independent variables on survival, we implemented univariate Cox regressions, considering demographic, clinical, and imaging variables as independent variables. Statistically significant variables included PI score ≥ 10 (p value: 0.02) and PI density ≥ 2.2(p value: 0.03). Afterward, we fitted a multivariate backward Cox regression on the variables with a significant univariate association: PI score ≥ 10 (p value: 0.03) remained significant at the final step (Table 3).

Discussion
ere is still no consensus on the clinical and imaging factors that are associated with and affect COVID-19 patients' outcomes and survival. Our findings contribute to the existing literature by illustrating that patients with higher PI scores, PI density index, axial distribution, cardiomegaly, or pleural or pericardial effusion are more likely to decease. e PI density score is our novel suggested index to distinguish between patients with the same PI score but a different number of involved lobes because it showed a stronger mortality prediction power.
Previously reported chest CT features to predict COVID-19 patients' mortality mainly included crude lung involvement score, number of involved lobes, bilateral or lower lobe involvement, and diffuse pattern [14,[19][20][21][22][23][24][25]. In line with our findings, yet in contrast to previous reports, Yuan et al. [14] reported no significant association of lower lobe or bilateral lung involvement with death. Nevertheless, a significantly higher CT score was the most common reported feature to predict death [14,[19][20][21][22][23]25]. Limited previous studies also reported cutoff values using totally different scoring systems [14,21,23]. For instance, Francone et al. [23] retrospectively evaluated 130 COVID-19 patients (20 patients deceased) and recommended the CT score ≥ 18 (out of 25) as the predictive factor for death. Another study used a 20-scale scoring system in 50 patients (23 deceased, 27 survived) and suggested a cutoff value of 12 with 0.79 AUC in predicting mortality [21]. Moreover, a study used a 72-scale CT score and reported 85% sensitivity and specificity in predicting the mortality of patients with a cutoff value of 24.5 [14]. In general, all previous studies were retrospective and nonsurvivors mostly comprised a small proportion of the studied population. Hence, differences in data sampling and study design (case-control vs. cross-sectional), as well as group matching, in our study remarkably improved the validity of results. Furthermore, details of chest CT findings were not fully reported in previous studies, and there is no consistency on the value of additional findings, including crazy paving and pleural effusion in predicting the outcome [19,21,25], which are both depicted in this study.
Further analyses on patients' survival demonstrated that PI score <10 and PI density index <2.2 were significantly associated with higher survival. To the best of our knowledge, imaging factors associated with the survival days of COVID-19 patients were addressed only in one retrospective study on 20 deceased patients [23]. In line with our findings, they found significantly lower survival days in patients with higher (≥18) vs. lower (<18) CT scores over a 24-day follow-up period [23].
Considering the limitations of the study, we still performed the largest case-control investigation on  Radiology Research and Practice evaluating the clinical and imaging factors to predict COVID-19 patients' outcomes and survival. We provided a novel semiquantitative scale with a defined optimum cutoff, which could serve to better identify high-risk patients and recognize patients with a higher demand for critical care but similar clinical conditions compared to other patients. We posit that patients with a PI score ≥ 10 and a PI density score ≥ 2.2 should be carefully cared for and receive aggressive treatment, as they are highly susceptible to death. Hence, radiologists are expected to report PI and PI density scores in their everyday practice to help clinical physicians manage patients more effectively and efficiently and consequently minimize COVID-19related mortality. Nevertheless, our findings must be interpreted in light of some limitations. Firstly, the study context is prone to a biased selection of cases and controls.
To elaborate, the setting is a tertiary and referral hospital for COVID-19 patients, and the patients who get admitted are probably more clinically severe than the cases in the general population with similar paraclinical and demographic features. On the other hand, there are reports on factors, such as immunological factors and predispositions in patients, that might have an impact on the survival rate of the COVID-19 cases [26,27]. However, we did not have other available data to incorporate in the study. In addition, the admitted patients are not equivalent to outpatient cases in the predicted mortality rate. For example, outpatient cases might have an even lower mortality rate with a similar PI score because they are in a relatively better clinical condition. Another limitation is that we did not follow the discharged patients to evaluate whether they were deceased. e interval time between symptom onset and admission was not exactly the same in all patients; however, we used the on-admission CT scans to minimize this bias. Although all patients were treated based on a unique national protocol, minor treatment differences based on physician clinical judgment could affect mortality and survival, which were not assessed in the present study. is study was conducted before the national COVID-19 vaccination program and further investigations on a larger population after vaccination are recommended to confirm our findings [28].
In conclusion, patients with a higher PI score, PI density index, axial distribution, cardiomegaly, or pleural or pericardial effusion are more susceptible to poor prognosis. Survival analyses revealed that a PI score ≥ 10 and PI density score ≥ 2.2 were significantly associated with lower survival and those patients should be prioritized for higher medical attention. Initial chest CT examination may be able to help in predicting the patient's outcome and survival.

Data Availability
All the data used in this manuscript can be made available upon request to the corresponding author.

Conflicts of Interest
e authors declare that they have no conflicts of interest.