Search strategy

We systematically searched the MEDLINE, Web of Science, and Scopus databases for relevant publications. Because of the novelty of this topic and for the reduction of publication bias, we also searched Google Scholar for the possible missing articles on other databases. Our search terms were the following: “CT”, “CT scan”, “computed tomography”, “imaging”; and “COVID19”, “SARS-CoV-2”, “Wuhan pneumonia”, “Wuhan corona”, “2019nCoV”, and “2019 novel coronavirus”. The search strategy for each database is provided with details in the supplement. No language restriction was applied. All relevant studies published between 1 December 2019 and 15 March 2020 were retrieved. We also manually searched the reference lists of reviews and relevant articles. An additional search was conducted on 4 April 2020 in MEDLINE using the same search terms.

Selection criteria

We included all peer-reviewed and non-peer-reviewed publications that had reported computed tomographic features of RT-PCR-confirmed patients with SARS-CoV-2 infection. Editorials, letters, comments, reviews, and case series with less than 10 reported cases were excluded from our search results. Two authors (P.M. and M.N.) independently conducted the title and abstract screening, and then performed a detailed review of the resulting publications. Any disagreement was consulted with another author (A.S.).

Quality assessment

We used a modified version of the Newcastle-Ottawa scale (NOS) designed for the cross-sectional studies as a tool for quality assessment [21]. This tool is concerned with three main areas of interest: sample selection, comparability of groups, and outcome. The maximum obtained score was ten stars.

Data extraction

Two authors (P.M. and M.N.) independently extracted all the relevant data about the CT features of COVID-19 patients. Because some studies had compared CT findings of confirmed COVID-19 patients with pneumonia patients of other infectious causes, the data of such patients were also retrieved with details. Extracted data consisted of the following parts: first author’s name, year of publication, possible subgroups, number of cases that underwent CT scan, number of patients with normal CT findings, ground-glass opacities (GGO), consolidation, air-bronchogram, bronchial wall thickening, bronchiectasis, crazy-paving pattern, halo sign, reversed halo sign, lymphadenopathy, pleural effusion, pleural thickening, tree-in-bud sign, cavitation, vascular dilation and thickening, bilateral involvement of lungs, peripheral distribution of inflammation, the involvement of more than one lobe, and the number of each of the involved lobes (RUL, RML, RLL, LUL, and LLL). Detailed definitions for each of these signs are provided in Table 1.

Statistical analysis

All proportional and between-group meta-analyses in this study were performed using Stata release 16.0 (StataCorp, College Station, TX, USA). For proportional data, we first transformed the proportions with the Freeman-Tukey double arcsine method so that the normal distribution could be achieved and the problem with proportions being too high or too low and their confidence intervals would be resolved [22]. Reported effect estimates and confidence intervals (CI) were then back-transformed to proportions. Heterogeneity between studies was assessed using the p-value of Cochran’s Q test and its more quantified form, I² statistics. We assessed publication bias using Egger’s test [23]. An I² value of less than 40% was considered as an indicator of low heterogeneity between publications. Based on the results of the I² statistics, we carried out a fixed or random effects model for meta-analysis. For proportional data, the effect size (ES) was the pooled proportion of CT features, reported with a 95% CI. For the between-group analysis, risk ratio and 95% CI were calculated. A p-value of less than 0.05 was considered significant. Because it seems that the course of the disease is generally milder in children [24], we decided to analyse paediatric cases (age < 18 years) separately (n = 147; six publications [25-30]). Also, three types of subgroup analyses were considered, to assess whether the prevalence of CT findings is different between: (1) those with mild-moderate and severe-critical disease; (2) surviving and deceased patients; and (3) those with RT-PCR-confirmed COVID-19 and those with pneumonia caused by other agents. For the severity of the disease, most studies classified their cases based on either the Guidelines for the Diagnosis and Treatment of New Coronavirus Pneumonia [31] or the American Thoracic Society guidelines for community-acquired pneumonia [32].

Study selection and characteristics

Our first search resulted in 547 results. After the removal of duplications, 428 publications underwent title and abstract screening. After the exclusion of irrelevant articles based on title and abstract screening, 109 articles were assessed for eligibility by full-text screening. Among these, we found 65 studies eligible to be considered for quantitative synthesis. Detailed reasons for full-text exclusion are given in Figure 1. With the additional search, 21 new publications were identified, and finally 86 eligible articles were included in the meta-analysis. The total number of patients who underwent at least one CT scan was 7956. The number of cases with at least one CT imaging ranged from 11 to 975 in different studies. The characteristics of the included studies are provided in Table 2.

Figure 1

Flow diagram of study selection for systematic review and meta-analysis of computed tomography findings in patients with COVID-19

Normal CT results

Among 80 publications, 58 had reported the number of patients without any abnormal CT findings (n = 6426). Pooled prevalence of patients with normal CT was 5% (95% CI: 0.03-0.07; p < 0.001; Figure 2). However, for paediatric patients this was higher (40%; 95% CI: 0.27-0.55; p < 0.001; Figure 3).

Figure 2

Forest plot for the pooled proportion of adult patients with unremarkable computed tomography findings

Figure 3

Forest plot for the pooled proportion of paediatric patients with unremarkable computed tomography findings

GGO

This is one of the most important CT findings suggestive of SARS-CoV-2 [33, 34]. The number of patients with GGO was reported in 59 studies (n = 5691). The pooled prevalence of GGO was 73% (95% CI: 0.67-0.78; p < 0.001; Figure 4), which makes it the most common imaging sign in patients with COVID-19. Also, GGO was found in 44% of paediatric patients with COVID-19 (95% CI: 0.26-0.63; p < 0.001; n = 65 patients; Supplementary Figure 1).

Figure 4

Forest plot for the pooled prevalence of ground-glass opacity in patients with COVID-19

Vascular dilation/thickening

Microvascular and macrovascular dilation is probably due to inflammation and the resulting hyperaemia. Thirteen studies reported this sign, although with different nomenclature (vascular dilation, thickening, and enlargement). The pooled prevalence of this feature among COVID-19 patients was 63% (95% CI: 0.52-0.73; p < 0.001; Supplementary Figure 2).

Pattern of involvement

Bilateral involvement of lungs

Fifty-four articles reported bilateral involvement of the lungs (n = 5250). The proportion of patients with this feature was 78% (95% CI: 0.73-0.82; p < 0.001; Figure 5), which is the most common CT scan pattern found in COVID-19 patients in this meta-analysis. Thirty-two per cent of paediatric patients had this feature in their CT images (95% CI: 0.13-0.53; p < 0.001; n = 102 patients; Supplementary Figure 1).

Figure 5

Forest plot for the pooled prevalence of bilateral involvement in patients with COVID-19

Involvement of more than one lobe

Among the included studies, a few reported detailed information about the number of involved lobes for each patient. However, because other publications just reported the proportion of cases with more than one lobe involvement, we took this as a cut-off for reporting the results. For 1687 cases, 75% had involvement of more than one lung lobe, which makes it the second most common pattern of involvement seen in CT images (95% CI: 0.68-0.82; p < 0.001; Figure 6).

Figure 6

Forest plot for the pooled proportion of patients with more than one lung lobe involvement

Severity

We found 14 publications that reported CT findings of patients with the severe-critical and mild-moderate disease, separately [35-48]. There were 1707 and 494 patients with mild-moderate and severe-critical situations, respectively. Subgroup analysis of these patients showed that those with severe-critical illness had a higher risk of having bilateral involvement (RR, 1.37; 95 CI: 1.06-1.78; p = 0.02; Figure 7) and consolidation (RR, 3.12; 95 CI: 1.57-6.23; p < 0.001; Figure 7). Also, the severe-critical group had a lower proportion of patients with normal CT, compared with the mild-moderate group (RR, 0.31; 95 CI: 0.18-0.53; p < 0.001). The details of these findings are summarized in Table 5 and Supplementary Figure 3.

Figure 7

Forest plot comparing the prevalence of consolidation (up) and bilateral involvement (down) between severe-critical and mild-moderate patients. REML – restricted maximum likelihood

Outcome

We found three studies [49-51] comparing CT findings of living and dead COVID-19 patients. It turned out that the presence of GGO is not associated with a worse outcome (RR, 1.97; 95% CI: 0.48-8.00; p = 0.34; Figure 8), but bilateral involvement with an increased risk of mortality (RR, 3.19; 95% CI: 1.07-9.49; p = 0.04; Table 5 and Figure 8).

Figure 8

Forest plot comparing the mortality rate between patients with and without bilateral involvement (up), and with and without ground-glass opacities (down)

Absence of SARS-CoV-2

Nine studies [52-60] compared imaging findings of pneumonia patients with confirmed COVID-19 (n = 968) to those without RT-PCR confirmation of the infection with SARS-CoV-2 (n = 821). COVID-19 cases had a higher risk of having GGO (RR, 1.45; 95 CI: 1.13-1.86; p < 0.001; Figure 9), but there was no significant difference between the presence of pleural effusion, positive CT findings, and bilateral involvement in two groups. However, non-COVID-19 patients showed an increased risk of having consolidation (RR, 1.23; 95 CI: 1.09-1.39; p < 0.001; Figure 9). Detailed information is provided in Table 5 and Supplementary Figures 4 and 5.

Figure 9

Forest plot comparing the incidence of consolidation (up) and GGO (down), between patients with and without COVID-19

Overall radiographic features

Normal chest CT was seen in only 5% of our cases. This, along with some other previously published studies, supports the fact that chest CT is a sensitive diagnostic tool that is appropriate for screening suspected patients, especially in epidemics [67,18]. Common findings in viral pneumonias include patchy or diffuse GGOs and reticulations. Consolidations may be present or absent [71]. Consistent with this phenomenon, our study results show that the most common radiological finding among COVID-19 patients is GGO, followed by vascular thickening (dilation), air bronchogram, and consolidation (observed in 73%, 61%, 41%, and 34% of cases, respectively). Other probable radiologic signs include crazy paving pattern (27%), pleural thickening (24%), bronchiectasis (24%), and halo sign (21%). Pleural effusion, reversed halo sign, and lymphadenopathies were rare, and almost no cases were seen with tree-in-bud sign or cavitations. A relatively low proportion of consolidation positive chest CTs may be a finding that differentiates SARS-CoV-2 infections from those caused by MERS-CoV and SARS-CoV. Conversely, GGOs are seen commonly in patients infected with all three epidemic-causing coronaviruses [72].

Distribution

Among five lung lobes, the left lower and right lower lobes were more likely to be involved in COVID-19 patients, as seen in 67% and 66% of cases, respectively. The least involved lobe was the right middle lobe. So, the infection is more common in the lower lobes, is usually bilateral (78%), and frequently involves more than one lobe (76%) in peripheries (68%). Lesions are usually multifocal (66%), and diffuse distribution (8%) is not common. These findings are also in favour of the distribution patterns proposed previously [73-76].

Concerning the distribution of different lesions observed in COVID-19, peripheral distribution and lower lung infiltrations are the features in common with MERS-CoV and SARS-CoV [72].

Paediatric population

It was previously observed that the disease course might have a milder pattern in children, with a higher proportion of asymptomatic carriers present [77,78]. In our study, a pooled prevalence of 40% was observed for normal chest CTs. Even considering the 95% CI, it is a significantly lower percentage compared with our general population. This implies that chest CT has lower sensitivity for identifying patients in the paediatric group, and relying on chest CT findings may increase the risk of missing patients with normal chest CTs. As a result, and considering the higher proportion of asymptomatic carriers in this age group, an integrated diagnostic approach that relies on thorough examination of clinical findings, exposure history, laboratory results, imaging, and RT-PCR tests might be a better option in children with suspected novel coronavirus infection. In addition, bilateral and unilateral involvement in chest CT images shows closer proportions in our findings, and GGOs are less common. This can mean that the typical imaging findings in children may also differ from those observed in the general population of COVID-19 patients.

Severity

In this meta-analysis, we tried to evaluate the differences in chest CT findings between the severe/critical group and mild/moderate group, classified based on the Guidelines for the Diagnosis and Treatment of New Coronavirus Pneumonia or the American Thoracic Society guidelines for community-acquired pneumonia.

In our study population for disease severity, 25% of cases were classified as severe/critical, considering the 19% report in a study on 72,314 confirmed and suspected COVID-19 patients in China [79]. According to their chest CT findings, it is inferred that the tendency towards developing consolidations and bilateral infiltrations is higher in patients with severe/critical disease. On the other hand, normal lung CT scan had a higher proportion in the mild/moderate group. Like in severe cases of MERS-CoV and SARS-CoV, pleural effusion is a radiological finding with greater prevalence in severe/critical SARS-CoV-2 infection [72]. This, combined with the clinical and laboratory data, may help in the classification of patients into severity groups and guide clinicians in choosing the subsequent plans accordingly.

Outcome

Other than diagnostic values of chest CT, its findings might be useful in predicting disease outcomes and patient prognoses in COVID-19. Regarding the outcome, bilateral involvement of lung parenchyma was associated with a higher risk of mortality. This finding is in accordance with the reports of a recent preprint article showing that the elevated level of inflammatory cytokines (CCL7, CXCL10, and IL-1 receptor antagonist) is associated with the extent of lung injury and a fatal outcome [80].

Pneumonia patients without confirmed SARS-CoV-2 infection

In this subgroup meta-analysis, we compared chest CT findings in conformed COVID-19 pneumonia cases with the negative-PCR group, including suspected patients without PCR confirmations and non-COVID pneumonias such as H1N1 influenza infection [81]. Consolidation, which was observed more commonly in severe/critical patients than in mild/moderate ones, had a higher chance for development in non-COVID pneumonias. In other words, the interpretation for observing consolidations in chest CTs depends on the groups we want to differentiate. If severity of the COVID-19 pneumonia is the question, consolidation favours severe/critical disease. Despite that, if we are going to differentiate COVID-19 pneumonias from non-COVID ones, consolidations in chest CT images make COVID-19 diagnosis less likely. After all, other diagnostic methods may be useful in differentiating such conditions.

Asymptomatic infection

CT abnormalities can also be present in COVID-19 patients without clinical symptoms. Wang et al. reported that 37 of 55 asymptomatic patients (67.2%) had abnormalities in their chest CT scans [82]. Similar observations were reported in two other studies, in which 17 of 24 [83] and 12 of 13 [84] asymptomatic patients had positive CT results (70.8% and 92.3%, respectively). However, due to insufficient data, we could not perform a meta-analysis of the imaging features of asymptomatic patients.