The institutional review board approved the study and informed consent was waived. This was a cross-sectional study with patients presenting with fever or respiratory symptoms, from March 1st to April 14th, 2020, at a COVID-19 dedicated area in the emergency unit of a private hospital. All patients who collected a reverse transcriptase-polymerase chain reaction (RT-PCR) for the diagnosis of COVID-19 were eligible. The study included consecutive patients with suspected pneumonia that underwent chest CT, at the discretion of the attending physician. Chest CT was ordered according to the institutional protocol taking into account pneumonia severity criteria, laboratory tests and comorbidities. Exclusion criteria were age <18 years, lack of data in medical records, and severe respiratory motion artifacts on chest CT.

COVID-19 was confirmed with one positive RT-PCR result. Patients with one negative RT-PCR result were considered non-COVID-19 if clinical, laboratory, and radiological findings indicated a low likelihood of COVID-19. Two consecutive (at least 48 h apart) negative results were required for excluding COVID-19 diagnosis in those judged to have a higher probability of disease, according to the independent evaluation of two infectious diseases specialists of institutional infection control team, as recommended by Infectious Diseases Society of America Guidelines.17

One oropharyngeal and two nasopharyngeal (from both nostrils) rayon swabs were collected. RNA extraction and real-time RT-PCR were performed at an external laboratory (Grupo Fleury), a reference private laboratory in Brazil, using primer and probes according to the Charité protocol, as previously described.18 Molecular tests for other respiratory pathogens were performed per request of the attending physician, and included either PCR for influenza A (H3N2 and H1N1),19 geneXpert® (Cepheid) for influenza A (H3N2 and H1N1) and B, or FilmArray® PCR Multiplex (Biomérieux), including adenovirus; coronavirus 229E, coronavírus HKU1, NL63 and OC43; human metapneumovirus; influenza A (H3N2 and H1N1) and B; parainfluenza 1,2, 3 e 4; and rhinovirus/enterovirus; Syncytial Respiratory Virus; Bordetella pertussis; Chlamydophila pneumoniae and Mycoplasma pneumoniae.

Imaging acquisitions were obtained with patients in the supine position during end-inspiration without contrast medium injection. Chest CT was performed on a 16-slice CT (Siemens Emotion 16 Slice CT Scanner, Siemens Healthineers, Forchheim, Germany) and 64-slice CT (Siemens Sensation 64 Slice CT Scanner, Siemens Healthineers, Forchheim, Germany). The following technical parameters were used for both CT scanners: tube voltage 130 kV; tube current modulation 100 mAs; spiral pitch factor 1.4; collimation width 0.625. Reconstructions were made with convolution kernel lung and soft tissue at a slice thickness of 1.00 mm. DICOM data were transferred onto a PACS workstation (Carestream Vue PACS version 12.1.6.1005, Carestream Health, NY, USA).

All chest CT images were evaluated independently by two radiologists with two and 10 years of thoracic imaging experience. Both readers evaluated every patient scan twice with a 2-day interval between readout sessions to assess intra and inter-rater agreement. Discordant reports were evaluated by a third thoracic radiologist with 12 years of experience and consensus was reached among the three radiologists. All radiologists were blinded for RT-PCR, clinical and laboratory results, and previous imaging exams.

CT features were classified as "typical," "indeterminate," "atypical," and "negative" for COVID-19 pneumonia", according to RSNA expert consensus.15 COVID-19 typical features are peripheral, bilateral and multifocal rounded ground-glass opacities (GGO) with or without consolidation or visible intralobular lines (“crazy-paving”). Reversed halo sign and other findings of organizing pneumonia can be seen later in the disease. Indeterminate features may occur but are nonspecific for COVID-19 pneumonia, such as diffuse, perihilar, or unilateral GGO. All unusual or unreported findings for COVID-19 were classified as atypical (e.g., centrilobular nodules, tree-in-bud opacities, or lung cavitation). If no CT suggestive findings of pneumonia were present, a negative classification was assigned. A typical appearance was considered a positive chest CT for COVID-19.

Statistical analysis was performed using SPSS version 18.0 (2009, PASW Statistics for Windows. SPSS Inc. Chicago, IL, USA). The association of CT findings and COVID-19 was assessed by univariate analysis using χ² or Fisher's exact test for categorical and Student's t or Wilcoxon Mann-Whitney tests for continuous variables. Sensitivity, specificity, positive and negative likelihood ratio of a positive CT for COVID-19 pneumonia were calculated. Intra- and inter-rater classification agreement beyond chance (appearance and each specific CT findings) and internal consistency reliability were evaluated with Cohen's kappa and Cronbach's alpha coefficients, respectively. Kappa coefficients of 0−0.20, 0.21−0.40, 0.41−0.60, 0.61−0.80, and 0.81–1.00 were considered to indicate none to slight, fair, moderate, substantial, and almost perfect agreement, respectively, and Cronbach’s alpha higher than 0.70 reflects internal consistency.20,21 Positive predictive value (PPV) and negative predictive value (NPV) of the classification according to distinct COVID-19 prevalence were also estimated. A p < 0.05 was considered statistically significant.

A total of 1176 patients underwent clinical evaluation for respiratory symptoms. In 443 (37.6%) RT-PCR for SARS-CoV-2 was collected. Of these, a chest CT with suspected pneumonia was reported for 163, four (2.4%) were excluded, and 159 (mean age, 57.9 ± 18.0 years; 88 [55.3%] males) were included in the study: 86 (54.1%) COVID-19 and 73 (45.9%) non-COVID-19 patients (Fig. 1). Out of 36 (22.6%) patients admitted to the intensive care unit during hospitalization, 24 (66.7%) were COVID-19 and 12 were (33.3%) non-COVID-19 patients.

Fig. 1.

Fow diagram of study participants. Abbreviations: RT-PCR, Reverse Transcriptase Polymerase Chain Reaction; SARS-CoV-2, Severe Acute Respiratory Syndrome Coronavirus 2; COVID-19, coronavirus disease 2019; CT, computed tomography.

(0.21MB).

Baseline, clinical and laboratory characteristics of patients were generally similar between COVID-19 and non-COVID-19 groups (Table 1). Age and sex were not significantly different between both groups. Anosmia was significantly more frequent in COVID-19 group. The non-COVID group had a shorter median duration of symptoms before attendance at emergency. COVID-19 patients had significantly lower oxygen saturation at emergency admission, lower leukocyte, lymphocyte, and platelets counts.

The median time between collecting nasal and oropharyngeal swabs and performing chest CT was 3.6 h (IQR, 1.6–8.3), and the median time from onset of symptoms to undergoing CT was seven days (IQR, 3–9). Eighty (93.0%) COVID-19 patients had their diagnosis confirmed in the first RT-PCR for SARS-CoV-2, while in 6 (7.0%) the first test was negative but the second was positive.

Chest CT features were classified as typical in 80 (50.3%) patients, as indeterminate in 30 (18.9%), as atypical in 17 (10.7%) and as negative for COVID-19 pneumonia in 32 (20.1%). The sensitivity, specificity, accuracy, positive and negative likelihood ratio of typical appearance were 88.3% (95%CI 79.9-93.5%), 94.5% (95%CI 86.7-97.8%), 91.1% (95%CI 85.7-94.6%), 16.1 (95%CI 9.8-26.4) and 0.12 (95%CI 0.10−0.14), respectively. The PPV and NPV of typical appearance were 95.0% (95%CI, 88.0%-98.0%) and 87.3% (95%CI, 79.4% - 92.5%), respectively. PPV and NPV according to distinct expected prevalence of COVID-19 among patients with respiratory symptoms are displayed in Fig. 2.

Fig. 2.

Positive and negative predictive values of a positive chest computed tomography (CT) according to distinct prevalences of COVID-19. A positive chest CT was defined as a typical appearance. Error bars indicate 95% confidence intervals (CI).

Positive predictive values (%) and 95% CI according to each estimated prevalence (P): P 10% - 63.6 (40.8-81.9); P 20% - 80.0 (62.5-90.9); P 30% - 87.5 (74.0-94.8); P 40% - 91.8 (81.1-96.9); P 50% - 94.6 (86.0-98.2); P 60% - 96.5 (89.5-99.1); P 70% - 98.0 (92.2-99.6); P 80% - 98.2 (93.1-99.7); P 90% - 99.2 (95.0-99.9).

Negative predictive values (%) and 95% CI according to each estimated prevalence (P): P 10% - 98.5 (94.3-99.7); P 20% - 96.7 (91.4 - 98.9); P 30% - 94.6 (88.1-97.7); P 40% - 92.7 (85.2-96.8); P 50% - 89.3 (80.1-94.6); P 60% - 84.5 (73.5-91.6); P 70% - 77.6 (64.4-87.0); P 80% - 66.6 (50.9-79.5); P 90% - 46.8 (29.5-64.9).

(0.09MB).

Only four patients presented a typical appearance and were not confirmed as COVID-19 in two RT-PCR exams; an alternative diagnosis could not established in two cases and were considered as possible COVID-19 pneumonia by the attending physicians. Of the remaining two patients, one had pulmonary thromboembolism with pulmonary infarction diagnosed through magnetic resonance angiography performed three days after chest CT and the other patient had bacterial community-acquired pneumonia. The typical finding in these latter two patients was peripheral and bilateral rounded GGO with consolidation.

Commonly typical features reported in COVID-19 pneumonia were multifocal, rounded and peripheral GGO displaying a sensitivity of 95.3% (95%CI 88.5-98.7), 82.5% (95%CI 72.8-89.9) and 81.7% (95%CI 71.6-89.3), respectively (Table 2; Fig. 3). All patients with atypical appearance (n = 17) on chest CT were not diagnosed with COVID-19 by RT-PCR. Of these, according to their attendant physician, 13 had the final diagnosis of bacterial pneumonia, one had decompensated heart failure and three had no confirmed alternative diagnosis. The most common findings among these patients were centrilobular nodules (n = 15) and tree-in-bud opacities (n = 7) (Fig. 4). Among patients with indeterminate appearance (n = 30), seven had a positive RT-PCR for SARS-CoV-2. In this group, common findings were very few and non-rounded GGO (n = 4), diffuse GGO (n = 2), and unilateral features (n = 1) (Fig. 5). One of them also had radiological features of pulmonary fibrosis.

Fig. 3.

Nonenhanced high-resolution chest CT of different patients with confirmed COVID-19 pneumonia and typical findings. A, 74-year-old man presented with 7-day history of fever and cough. Axial CT shows multifocal, peripheral and rounded ground glass opacities (GGO). B, 47-year-old man presented with 10-day history of moderate breathlessness and fever. Axial CT shows reversed halo sign. C, 70-year-old woman presented with 9-day history of mild dyspnea and COVID-19 exposure. Axial CT shows GGO with a perilobular pattern. D, 36-year-old man presented with 5-day history of fever, cough and myalgia. Axial CT shows bilateral areas of crazy-paving pattern.

(0.25MB).

Fig. 4.

Nonenhanced high-resolution chest CT of different patients with atypical findings. A, 52-year-old man presented with 3-day history of fever, cough and adynamia. Axial chest CT shows centrilobular nodules, tree-in-bud opacities and bronchial mucocele. The patient was diagnosed with pulmonary tuberculosis. B, 26-year-old woman presented with 10-day history of cough, sputum, fever and dyspnea. Axial chest CT shows lobar consolidation. The patient was diagnosed with bacterial acquired community pneumonia. C, 47-year-old woman presented with 30-day history of headache, adynamia, cough and chest pain. Axial chest CT shows pulmonary cavitation with satellite centrilobular opacities. The patient was diagnosed with central nervous system and pulmonary cryptococcosis. D, 55-year-old man presented with 3-day history of orthopnea, precordial pain and cough. Axial chest CT shows bilateral pleural effusion, interlobular septal thickening, and centrilobular ground glass opacities. The patient was diagnosed with congestive heart failure due to myocardial infarction.

(0.23MB).

Fig. 5.

Nonenhanced high-resolution chest CT of different patients with indeterminate findings. A, 40-year-old man presented with 5-day history of worsening of chronic cough, dyspnea and fever. Axial CT shows diffuse and bilateral ground glass opacities. The patient was diagnosed with acquired immunodeficiency syndrome and Pneumocystis pneumonia. B, 55-year-old man presented with 3-day history of cough and mild breathlessness. Axial CT shows unilobar, rounded and peribroncovascular ground glass opacity. COVID-19 pneumonia was confirmed.

(0.12MB).

A total of 33 (41.2%) patients with typical findings were tested for influenza by molecular methods (three of them for all other pathogens described in Methods; two were patients with negative SARS-CoV-2 RT-PCR) and none had positive results. Molecular tests for influenza were negative in other five (16.7%; one also negative for other pathogens), three (17.6%) and five (15.6%; one also negative for other pathogens) patients with indeterminate, atypical, and negative CT findings, respectively.

Intra-rater agreement in assigned classification was seen in 282 (88.7%) of 318 observations. Regarding intra-rater agreement, Cohen’s kappa of observer one was 0.847 (P = .13), and observer two was 0.924 (P = .06). Inter-rater agreement ranged from 0.725 (P = .001) to 0.772 (P = .05) between two radiologists (see Tables S1 and S2). The agreement coefficients of selected typical findings are shown in Table S3.