The review board from the centers approved this study. The informed consents of the patients were waived for the retrospective study.
Study Participants
This multi-center, retrospective study composed of two groups of patients. For the COVID-19 group, continuous patients admitted to the centers out of Wuhan from January 17th, 2020 to the February 26th, 2020 were initially screened. Patients with an exposure history, suspected clinical symptoms, pneumonia signs on chest imaging, abnormal laboratory findings were suspected with COVID-19 (Fig. 1). Then, patients diagnosed with COVID-19 were finally enrolled. The diagnosis was based on the RT-PCR testing of the specimen collected from bronchoalveolar lavage fluid or sputum according to the guidelines from WHO and NHC, PRC. No exclusion was applied to this group.
Patient in the contrast group were selected similar in size to the COVID-19 cohort but with no possible exposure history of the SARS-CoV-2. To avoid the confounding effect of seasonal activity and clothing, we retrospectively reviewed the cases admitted in the centers through outpatient emergency service during the same timeframe from January 17th to February 14th in 2019. Patients with a diagnosis of acute pneumonia (within two weeks) were included. Exclusion criteria were applied as follows for this group: (1) lung cancer, tuberculosis, severe interstitial lung disease, pulmonary edema or other conditions that might cause significant increase of the parenchymal opacities (2) the quality of whose HRCT was not good enough for further evaluation.
All the clinical variables, including baseline characteristics, basic signs taken at the emergency department, exposure history, clinical symptoms, duration from illness onset to CT, and the clinical outcome within the study date range, were collected through the history and nursing recordings. According to the sixth version of clinical management guideline of COVID-19 from NHC, PRC, severe illness was defined as patients who fulfilled one of the three criteria or even severer: (1) respiratory distress, respiratory rate (RR) ≥ 30 beats/min; (2) oxygen saturation measured at fingertip ≤ 93% at rest; (3) the ratio between arterial blood oxygen partial pressure and oxygen concentration (PaO2/FiO2) ≤ 300 mmHg (1 mmHg = 0.144 kPa).
Chest CT Scan
The patients were imaged in six centers. 65 and 7 patients were imaged with a 1-mm section thickness, 256-detector CT scanner (Revolution CT, GE Healthcare, Milwaukee, Wis) and a 2-mm section thickness 128-detector CT scanner (Ingenuity Core, Philips Healthcare, Best, The Netherlands) separately at two hospitals in city A; 5 patients were imaged with a 1-mm section thickness, a 64-slice spiral CT scanner (SOMATOM Definition Flash, Siemens, Germany), a 16-detector CT scanner (Brilliance, Philips Healthcare, Best, The Netherlands), or a 16-detector CT scanner (uCT 510, United Imaging, Shanghai, China) at three centers in city B. One patient was imaged with a 1-mm section thickness, 16-detector CT scanner (Philips Ingenuity, Philips Medical System, Netherlands) at one center in city C. (Detailed scanning parameters were provided in supplementary).
Image Analysis
Two radiologists with an over 5-year diagnostic experience analyzed the high-resolution chest CT (HRCT) scans independently and blindly to the clinical variables. Disagreement in imaging interpretation were solved by consensus.
For each case, the presence and number of involved lobes were recorded for the lung lesions including: (1) ground-glass opacity (GGO) (patchy or round morphology like lesions with an increase in lung parenchymal opacification without obscuration of the underlying vessels); (2) consolidation (an increase in lung parenchymal with obscuration of the underlying vessels); (3) Fibrosis (presence of irregular linear opacities, parenchymal bands, traction bronchiectasis and lung distortion based on the above two lesions). Nodule or mass, cavitation, lymphadenopathy, and pleural effusions were also noted.
Concerning the distribution pattern, each case was classified as unilateral or bilateral, unifocal or multifocal, cranial or caudal. In the transverse plane, the distribution pattern was recorded as upper, subpleural (or periphery), peribronchovascular (namely, central) or random(7).
The severity score was graded for each lobe and then summed, according to the visual grades used in previous CT studies of patients with SARS and MERS(8,9). In detail, 0 = no lesion; 1 = single or multiple nodule-like small (with a diameter less than 1 cm) GGO with an extent less than 10% of the lobe; 2 = single or multiple nodule-like medium (with a diameter between 1 to 3 cm) GGO with an extent of 10–30% of the lobe; 3 = single or multiple nodule-like large (with a diameter more than 3 cm) GGO with an extent within 30–60% of the lobe, or any extent of the GGO with consolidation; 4 = any diameter of GGO with an extent of 60–90% of the lobe, or with evidence of fibrosis. 5 = Diffuse extent of GGO with an over 90% extent of the lobe.
Statistical Analysis
Shapiro-Wilk test was used for normality testing, with histograms, and Q-Q plot. Parametrical data was expressed as mean ± standard deviation (SD). Non-parametrical data was expressed as median with interquartile range (IQR). Two-sided non-paired Student’s t test or Wilcoxon’s test was used for comparison between the two groups. Categorical variables were presented as frequency and the corresponding percentage and Chi-square test was used for comparison. The relationship between HRCT severity scores and clinical variables was assessed by Spearman correlation analysis. The agreement between the radiologists on the maximal diameter of abnormal pulmonary findings was evaluated using the intraclass correlation coefficient (ICC), as well as expressed as bias ± SD and graphically shown as difference plots according to Bland and Altman(10). A P value less than 0.05 was considered statistically significant. Statistical analysis was performed with R project (v. 3.3.1).