Reverse-transcriptase polymerase chain reaction versus chest computed tomography for detecting early symptoms of COVID-19. A diagnostic accuracy systematic review and meta-analysis

ABSTRACT BACKGROUND: A positive real-time reverse-transcriptase polymerase chain reaction (RT-PCR) for SARS CoV-2, from nasopharyngeal swabs, is the current gold standard diagnostic test for this virus and has sensitivity of 60-70%. Some studies have demonstrated a significant number of false-negative RT-PCR tests while displaying significant tomographic findings, in the early days of symptoms of COVID-19. OBJECTIVE: To compare accuracy between RT-PCR and computed tomography (CT) for detecting COVID-19 in the first week of its symptoms during the pandemic. DESIGN AND SETTING: Systematic review of comparative studies of diagnostic accuracy within the Evidence-based Health Program of a federal university in São Paulo (SP), Brazil. METHODS: A systematic search of the relevant literature was conducted in the PubMed, EMBASE, Cochrane Library, CINAHL and LILACS databases, for articles published up to June 6, 2020, relating to studies evaluating the diagnostic accuracy of RT-PCR and chest CT for COVID-19 diagnoses. The QUADAS 2 tool was used for methodological quality evaluation. RESULTS: In total, 1204 patients with COVID-19 were evaluated; 1045 had tomographic findings while 755 showed positive RT-PCR for COVID-19. RT-PCR demonstrated 81.4% sensitivity, 100% specificity and 92.3% accuracy. Chest CT demonstrated 95.3% sensitivity, 43.8% specificity and 63.3% accuracy. CONCLUSION: The high sensitivity and detection rates shown by CT demonstrate that this technique has a high degree of importance in the early stages of the disease. During an outbreak, the higher prevalence of the condition increases the positive predictive value of CT. REGISTRATION NUMBER: DOI: 10.17605/OSF.IO/UNGHA in the Open Science Framework.


INTRODUCTION
Since COVID-19 pneumonia emerged in Wuhan, China, there has been a search for knowledge that might prevent or minimize its spread. [1][2][3] In just over three months after its initial breakout, it gained worldwide reach such that it affected more than 2.5 million people, with more than 180,000 deaths in more than 200 countries. COVID-19 is caused by the SARS-CoV-2 virus, a member of the Coronaviridae family. 3 Its transmission occurs mainly through respiratory droplets. 1 The clinical spectrum of the disease is variable, and the majority of cases are asymptomatic or oligosymptomatic. The most severe cases, with acute respiratory distress syndrome, commonly affect elderly patients with comorbidities. 3 A positive real-time reverse transcriptase-polymerase chain reaction (RT-PCR) for SARS-CoV-2, from nasopharyngeal swabs, is the current gold standard diagnostic test. The sensitivity of RT-PCR for SARS-CoV-2 is 50-70%; 4-10 around 30-40% of patients with early-stage COVID-19 are false-negative. 4 An inadequate technique for collecting sampling material or low viral load, limited development of nucleic acid detection technology and variation in the detection rate between different manufacturers may all be determinants for false negative results. 4,11 Use of computed tomography (CT) is based on the clinical context and time taken to make the diagnosis, especially in relation to use of RT-PCR and other clinical and laboratory investigations. 4 CT findings do not alter the diagnosis of COVID-19 in cases in which RT-PCR is positive, but they are useful for grading pulmonary involvement and its evolution. 4,6,8 CT has 56-98% sensitivity, 7 and according to Ai et al., 25% specificity and 68% accuracy. 14 Ai et al. found that out of 64 patients with an initially negative RT-PCR test, 15 (23.4%) subsequently had a positive RT-PCR (mean time interval of 5.1 ± 1.5 days); ten of these patients (15.6% of those with initial negative RT-PCR) had typical CT findings at the time of the initial negative RT-PCR. 14 Fang et al. described a 29.4% rate of abnormal CT in patients with initially negative and subsequently positive RT-PCR. 4,11 In the minority of patients with high clinical suspicion in the context of the current pandemic, but with negative initial RT-PCR,

OBJECTIVES
To determine the accuracy of RT-PCR and CT over the first seven days of symptoms of COVID-19 and which method is more sensitive for early case detection.

Study model
The study model followed the guidelines for systematic reviews on diagnostic accuracy studies, i.e. Cochrane Diagnostic Reviewer's Handbook version 5.1.

Inclusion criteria
The search was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. We included comparative studies on diagnostic accuracy among patients who underwent both CT and RT-PCR for making the diagnosis of COVID-19 in the initial days of its evolution, regardless of the severity of the disease.
We did not put any restrictions on patient age, origin, language or publication status of the study. There was no exclusion regarding population size or patient age. In the case of missing information, the authors of the study in question were contacted by e-mail.

Participants
The participants were men and women of all ages with suspected COVID-19 who underwent chest CT and RT-PCR during their first week of symptoms.

Selection of studies and data extraction
The studies selected were those potentially eligible for inclusion in terms of relevance of the articles or abstracts in indexed journals. Two authors performed independent selections for eligibility. In cases of disagreement, a third author was consulted.
Data extraction was performed using a standardized form.

Evaluation of methodological quality
The QUADAS 2 tool, which is used to evaluate bias and precision, was used in relation to all the eligible studies. 16  of the studies included and the main reviews on the subject were also evaluated. Manual searches were also carried out in these reference lists. The full search strategy is displayed in Table 1.

Studies selected
The systematic review yielded 168 studies. At the end of the analysis, five studies. 9,11,14,17,18 were deemed to meet the inclusion criteria and presented acceptable quality according to the QUADAS 2 tool. These studies were thus included in the systematic review (Figure 1). Among these, two studies were included in the meta-analysis. 9,17 In all the studies, there was high concern about applicability. Moreover, in three of the five studies, a high risk of bias was also perceived. It was not clear in most studies whether the radiologist who reported the CT scan had access to the RT-PCR results (Figure 2).        Ai et al. 14 Bernheim et al. 18 Fang et al. 11 He et al. 9 Long et al. 17 High Unclear Low Applicability concerns

Patient selection
Early RT-PCR CT RT-PCR Ai et al. 14 Bernheim et al. 18 Fang et al. 11 He et al. 9 Long et al. 17 In the study by Fang et al., 11 study, 36 CT-positive cases showed typical changes: sparse, subpleural and peripheral ground-glass opacities, commonly in the lower lobes. The CT sensitivity was 98.0%. The first RT-PCR performed on the patients presented sensitivity of 70.5%.
He et al. 9 compared use of CT and RT-PCR among 82 patients with suspected pneumonia, including COVID-19 pneumonia.
The two experienced radiologists who evaluated all chest CT scans demonstrated good interobserver agreement. All the patients underwent chest CT and initial RT-PCR on the same day. The 34 COVID-19 patients had confirmation through RT-PCR, but not necessarily from the initial RT-PCR. The initial RT-PCR had 79% sensitivity, 100% specificity and 92% accuracy. The chest CT had 77% sensitivity, 96% specificity and 88% accuracy. He et al. also analyzed the two tests used in conjunction, and concluded that jointly they presented 88% sensitivity, 100% specificity and 98% accuracy. 9 In the study by He et al., 9  The upper lobes of the lungs were more affected on CT in COVID-19 cases (right: 52.7% versus 37.3%; left: 55.6% versus 33.3%); the other lobes did not show any significant difference. There was also a difference in peripheral involvement, which was more common in cases of pneumonia caused by COVID-19. The sensitivity of CT was 97.2%. The first RT-PCR performed on the patients presented sensitivity of 84.6% and the negative predictive value was 89.4%. In the study by Long et al., 17 51 patients with tomographic findings had pneumonia other than COVID-19. This makes it possible to infer that CT presents high specificity. The positive predictive value for CT was calculated as 58.6%.
All the data in these five studies were retrospective and were obtained during the epidemic period in the regions where these studies were conducted.

DISCUSSION
The symptoms of COVID-19 consist mainly of fever, fatigue and dry cough, with gradual dyspnea in some cases, and acute respiratory distress syndrome and multiple organ dysfunction in severe cases requiring intensive treatment. 1,2,19,20 While the majority of patients, about 80%, have mild symptoms; older patients, especially those above 70 years old and those with underlying conditions, such as cardiovascular disease, diabetes, chronic respiratory diseases and oncological diseases, have a higher mortality rate of up to 15%. 3 In addition to the most common pattern of peripheral and bilateral ground-glass injuries, other patterns of lung injury may be observed. 6,7,[20][21][22][23][24][25] Pulmonary consolidations are present in 2-64% of the cases and form an indicator of disease progression, thus serving as a warning sign for the severity of the patient's condition. Reticular pattern lesions have lower incidence than consolidations and opacities. 6,26 The crazy-paving pattern is present in about 5-36% of the cases, while bronchial wall thickening is present in 10-20%. 6,27 Pleural changes are present in 32%, with pleural thickening; however, pleural effusion occurs in only 5% of the cases. 6,24 Pulmonary fibrosis occurs in 17% of the cases and pulmonary nodules smaller than three centimeters in size, in 3-13%. 6 The incidence of lymph node enlargement is about 4-8% and pericardial effusion occurs in approximately 5%. The latter is an indicator of severity. 6,19 Vascular thickening is characterized in 59% of the cases. 7 The radiological findings tend to become worse seven days after the onset of symptoms and show improvement 14 days after the onset of symptoms. 3 In the current pandemic situation, despite the low specificity of CT (25%), this technique can be used to isolate patients and institute treatment at an early stage, since it presents sensitivity of about 88.9%, starting from the early day of symptoms. 4,10,20,28 In comparison with this, chest X-ray shows abnormalities in 59.1% of the cases and in 76.7% among serious cases. 4,23 Xie et al. 29 reported on a case series in which they performed RT-PCR and CT on the same day, regardless of the duration of the patients' symptoms. They found that out of their 167 patients, 162 were positive according to RT-PCR and 160 were positive accord- Our findings showed that CT outperformed RT-PCR in making an early diagnosis of COVID-19 in suspected cases. Both from previous findings and ours, we suggest that an early evaluation protocol should include applying CT when RT-PCR is negative.
This could guide clinicians' treatment and patient isolation criteria, in order to avoid virus dissemination. Our meta-analysis showed that CT had specificity of 43.8% and sensitivity of 95.3%, and both of these values are higher than those in the recent literature.
All the studies evaluated were conducted among in patients with COVID-19 that confirmed within the first seven days of symptoms by means of RT-PCR. However, this test was not necessarily the first to be performed on suspected patients, within the epidemic period in the country in which these tests were performed.
Therefore, the positive predictive value and detection rate of CT findings in patients with COVID-19 will be higher than it would be outside the epidemic period.
Although RT-PCR is the gold standard for diagnosing COVID-19, it presents a significant percentage of false-negative tests in the early days of symptoms of the disease (0-7 days). On the other hand, even though CT is a test with presumably low specificity, 26 thereby allowing several differential diagnoses, 8 it detects patterns compatible with COVID-19. It has presented very high sensitivity and significant positive predictive value and detection rate in the epidemic period. 8,26

CONCLUSION
The high sensitivity and detection rate of CT demonstrate that it has a high degree of importance in the early stages of the disease, even greater than RT-PCR. During an outbreak, the higher prevalence of the condition raises the positive predictive value of CT. However, the low specificity of CT (43.8%) also needs to be considered. Outside of pandemic times, its positive predictive value for this condition should decrease proportionally with the decline in the prevalence of the disease in the population.