Abstract
Objectives To compare the performance of chest computed tomography (CT) scan versus reverse transcription polymerase chain reaction (RT-PCR) in the initial diagnostic assessment of coronavirus disease 2019 (COVID-19) patients.
Methods A systematic review and meta-analysis were performed as per the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines. A search of electronic information was conducted to identify studies comparing the diagnostic performance within the same patient cohort of chest CT scan versus RT-PCR in COVID-19 suspected cases. Sensitivity, specificity and accuracy were primary outcome measures. Secondary outcomes included other test performance characteristics, discrepant findings between both investigations and main chest CT findings. Random effects modelling was used for the analyses.
Results Thirteen non-randomised studies enrolling 4092 patients were identified. Accuracy was statistically significantly higher for RT-PCR versus chest CT (Odds Ratio [OR] = 0.22, P = 0.001). Chest CT is also less specific than RT-PCR. Ground-glass opacities and consolidations were the most common chest CT manifestations. Importantly, early small studies tended to favour chest CT versus later larger studies.
Conclusion Chest CT is inferior to RT-PCR for the initial detection of COVID-19 and has more false positives. It may still be useful in confirming COVID-19, however, in patients with a suspicious clinical presentation, but who have a false-negative SARS-CoV-2 RT-PCR test.
Highlights
- Chest computed tomography (CT) is less specific in the diagnosis of coronavirus disease 2019 (COVID-19) when compared to reverse transcription polymerase chain reaction (RT-PCR).
- The accuracy of RT-PCR is statistically significantly higher than chest CT for COVID-19.
- Chest CT, however, can detect false-negative and true-positive RT-PCR cases.
1. Background
In December 2019, Coronavirus Disease 2019 (COVID-19), caused by Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-COV-2) emerged from Wuhan city, Hubei province, China [1, 2]. SARS-COV-2 is part of the Coronavirdae family that includes the common cold, Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS). These viruses can cause Acute Respiratory Distress Syndrome (ARDS), bilateral pneumonia and pulmonary failure leading to mortality [3]. There is yet to be a highly effective pharmacological therapy identified for COVID-19, further highlighting the importance of early detection and isolation of COVID-19 patients to prevent the spread of the disease [4].
Reverse-transcription polymerase chain reaction (RT-PCR) tests are currently the gold standard diagnostic tool for COVID-19. RT-PCR assays can be performed on nasopharyngeal and/or oropharyngeal swabs, sputum, blood samples, body fluids, stool samples and bronchoalveolar lavage fluid [5]. RT-PCR is limited by its low sensitivity and associated relatively high number of false-negative results that can lead to the erroneous assumption that a patient who does actually have COVID-19 does not have it and is not infectious, thus leading to an undetected transmission risk in either the community or within an institutional setting [4].
CT has become a standard of care in the diagnosis and assessment of a variety of respiratory conditions, such as interstitial lung disease and lung cancer, and optimises the management process [6, 7]. Although CT scans are not routinely used to diagnose ARDS, certain complications relating to mechanical ventilation, including pneumonia, pneumothorax and emphysema, are sometimes identified by CT but not chest radiography [8]. In addition, CT imaging can be used to identify lung atelectasis due to poor positioning of the endotracheal tube as well as potentially directing ventilation to achieve optimal pressures and air recruitment [8]. This is important in relation to COVID-19 since patients in ICU require optimisation of ventilatory settings and it is increasingly recognised that prone ventilation appears favourable [9]. Nevertheless, in the clinical setting, the benefits of routine CT imaging must be weighed against the considerable practicalities and risks, including that of infection transmission associated with transporting a patient from the intensive care unit, or elsewhere in the hospital, to the radiology department [8]. The Royal College of Radiologists does not recommend the use of CT-scanning as a diagnostic tool for COVID-19, as a negative CT finding does not rule out COVID-19 infection, except for when CT-scanning results can change the management course such as those who require emergency surgery [10]. Furthermore, both the American Society of Emergency Radiology and the Society of Thoracic Radiology advise against the use of CT-scanning as a diagnostic tool for COVID-19 [11]. They also recommend restricting the use of CT-scanning for confirmed and highly suspected cases of COVID-19 when the result may help manage complications such as a lung abscess [11].
Nevertheless, despite the above guidance, CT-scanning has been commonly used in some countries during the pandemic. Given this, and recent observational studies of chest CT and RT-RCR in detecting COVID-19 cases [4, 12–23], there is a need for ongoing systematic reviews and meta-analyses of studies that have directly compared the performance of these tests in the same patient cohorts. This study aimed to compare the outcomes of chest CT-scan versus RT-PCR for the diagnosis of COVID-19 by systematic review and meta-analysis of truly comparative studies.
2. Methods
A systematic review and meta-analysis were conducted as per the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [24].
2.1. Eligibility criteria
All relevant studies comparing chest CT with RT-PCR in the diagnosis of patients with suspected COVID-19 were included [4, 12–23]. Chest CT was the intervention group of interest and RT-PCR was the comparator. All patients were included irrespective of age, gender or co-morbidity status. Studies that did not compare the two tests directly within the same patient cohort were excluded. Studies enrolling only confirmed COVID-19 cases or studies performing chest CT for prognostic rather than diagnostic purposes were also excluded.
2.2. Primary Outcomes
Sensitivity, specificity and accuracy for the diagnosis of COVID-19 cases were the primary outcomes. Sensitivity (true positive rate) is the proportion of actual COVID-19 cases a test correctly identifies. Specificity (true negative rate) is the proportion of non-COVID-19 cases a test correctly identifies. Accuracy is the proportion (true positives and true negatives) of total cases examined a test correctly identifies.
2.3. Secondary Outcomes
The secondary outcomes were positive predictive value (in the event of a positive test the probability the patient is truly positive), negative predictive value (in the event of a negative test the probability the patient is truly negative), positive (PLR) and negative likelihood (NLR) ratios (respectively, the probability a person who has COVID-19 testing positive divided by the probability of a person who does not have COVID-19 testing positive [PLR] and the probability of a person who has COVID-19 testing negative divided by the probability of a person who does not have COVID-19 testing negative [NLR]) and discrepancy of findings between both investigations as well as main chest CT findings.
2.4. Literature search strategy
Three authors independently searched online databases as well as pre-print servers for details of ongoing and unpublished studies. The electronic databases were: MEDLINE, EMBASE, EMCARE, CINAHL and the Cochrane Central Register of Controlled Trials (CENTRAL). The last search was run on 23rd of August 2020. Thesaurus headings, search operators and limits in each of the above databases were adapted accordingly. In addition, World Health Organization International Clinical Trials Registry, ClinicalTrials.gov, ISRCTN Register and medRxiv were also searched for ongoing and unpublished studies. No language restrictions were applied in our search strategies. The search terminologies included “COVID-19”, “Coronavirus”, “SARS-CoV-2”, “severe acute respiratory syndrome coronavirus 2”, “CT”, “computed tomography”, “RT-PCR”, “Reverse transcription polymerase chain reaction”, “PCR” and “polymerase chain reaction”. The bibliographic lists of relevant articles were also reviewed.
2.5. Selection of Studies
The title and abstract of articles identified from the literature searches were assessed independently by two authors. The full texts of relevant reports were retrieved and those articles that met the eligibility criteria of our review were selected. Any discrepancies in study selection were resolved by discussion between the authors.
2.6. Data Extraction and Management
An electronic data extraction spreadsheet using Microsoft Excel was created in line with Cochrane’s data collection form for intervention reviews. The spreadsheet was pilot tested in randomly selected articles and adjusted accordingly. The data extraction spreadsheet included study-related data (first author, year of publication, country of origin of the corresponding author, journal in which the study was published, study design, study size, clinical condition of the study participants, type of intervention, and comparison), baseline demographics of the included populations (age and gender) and primary and secondary outcomes data. Three authors cooperatively collected and recorded the results with any disagreements resolved via discussion.
2.7. Data Synthesis
Data synthesis was conducted using the Review Manager 5.3 software and Microsoft Excel. The extracted data was entered into Review Manager. The analysis used was based on the random effect model. The results are reported in forest plots with 95% Confidence Intervals (CIs).
For dichotomous outcomes, the Odds Ratio (OR) was calculated between the 2 groups. The OR is the odds of an event in the chest CT group compared with the RT-PCR group. An OR of greater than 1 (with the lower 95% CI >1) for accuracy to detect or exclude COVID-19 cases favours Chest CT, an OR of less than 1 (with the upper 95% CI <1) favours RT-PCR, and an OR of 1 (with the 95% CIs bridging 1) favours neither. Pooled mean sensitivity and specificity values were generated. Summary receiver operator characteristic (sROC) curves were also generated and compared.
Heterogeneity among the studies was assessed using the Cochran Q test (χ2). Inconsistency was quantified by calculating I2 and interpreted using the following guide: 0% to 25% may represent low heterogeneity, 25% to 75% may represent moderate heterogeneity, and 75% to 100% may represent high heterogeneity.
2.8. Sensitivity and Subgroup Analyses
To identify areas of heterogeneity and assess the robustness of our results in studies with certain characteristics, additional sensitivity and sub-group analyses were performed.
2.9. Methodological Quality and Risk of Bias Assessment
Three authors independently assessed the methodological quality as well as the risk of bias for articles matching the inclusion criteria. The Newcastle-Ottawa Scale was used to assess the quality of included studies, which uses a star system for analyses [25]. It offers a maximum score of nine stars across three domains including selection, comparability and exposure. The overall rating of good, fair or poor quality was based on the Agency for Healthcare Research and Quality (AHRQ) standards [25].
3. Results
3.1. Literature Search Results
Our search strategy retrieved 1,630 studies. Thirteen studies met the eligibility criteria for inclusion following systematic review (Figure 1).
The PRISMA diagram details the search and selection processes applied during the overview. PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.
3.2. Description of Studies
The baseline characteristics of the included studies are summarised in Table 1. All included studies were comparative with CT scans being assessed using RT-PCR as the reference.
Key NR: Not Reported.
3.2.1. Ai et al
Ai et al. [4] conducted a single centre retrospective study that included 1014 patients (the largest study included) with suspected COVID-19 from January 6 to February 6, 2020. RT-PCR and chest CT results were obtained from medical records and three CT systems with all chest CT images reviewed by two radiologists.
3.2.2. Besutti et al
Besutti et al. [12] conducted a single centre prospective study that included 696 patients with suspicion of COVID-19 who had both chest CT scan and RT-PCR from March 13 to March 23, 2020. Subjects with a time gap between CT and RT-PCR of more than 3 days were excluded. In addition to the routine radiological reporting, each radiologist created a structured report about the probability of COVID-19 pneumonia based on the CT findings (highly suggestive, suggestive and non-suggestive).
3.2.3. Caruso et al
Caruso et al. [13] conducted a single centre prospective cohort study that enrolled a total of 158 study participants that underwent both RT-PCR and chest CT due to COVID-19 suspicion from March 4, 2020 to March 19, 2020. Two experienced radiologists reviewed all chest CT images.
3.2.4. Çinkooğlu et al
Çinkooğlu et al. [14] reported a single centre retrospective study that included 185 patients undergoing both RT-PCR and CT scan from March 15th to April 15th, 2020. A diagnosis of COVID-19 was confirmed via initial or follow-up RT-PCR test. Two radiologists independently reviewed CT images and any disagreements between the two caused a third radiologist to make the final decision.
3.2.5. Falaschia et al
Falaschia et al. [15] conducted a single centre retrospective study that included 773 patients with suspicion of COVID-19 who had both chest CT scan and RT-PCR from March 3 to April 9, 2020. Two radiologists with more than ten years of experience in thoracic imaging, and who were blinded to the RT-PCR results, evaluated the imaging using a consensus approach.
3.2.6. Fang et al
Fang et al. [16] performed a single centre retrospective study that included 51 patients with suspected COVID-19 from January 19 to February 4, 2020. All included patients underwent chest CT and RT-PCR testing within an interval of three days or less.
3.2.7. Gaia et al
Gaia et al. [17] conducted a single centre retrospective study that included 314 patients who presented with clinically suspected COVID-19 from March 3 to March 23, 2020. Patients were evaluated with PaO2/FiO2 ratio from arterial blood gas, RT-PCR assay from nasopharyngeal swab sample and chest CT. Two radiologists who independently evaluated chest CT scans were blinded to the final diagnoses and to the RT-PCR results.
3.2.8. Gietema et al
Gietema et al. [18] conducted a single centre prospective study that included 193 symptomatic emergency department patients, who were suspected of having COVID-19, between March 13th and March 24th, 2020. All patients had a chest CT and RT-PCR test upon arrival to the emergency department. Data were collected from medical records and CT scan results compared to that of RT-PCR. The initial judgment of CT scans was performed by a senior resident and the final reading and reporting was performed by an experienced chest radiologist within 12 hours of scanning.
3.2.9. Guillo et al
Guillo et al. [19] conducted a single centre retrospective study that included 214 patients with suspicion of COVID-19 who had both chest CT scan and RT-PCR within 24 hours from March 1 to March 28, 2020. As part of the clinical routine, the chest CT scans were initially read by a radiology resident and approved by a senior radiologist. The scans were then further independently assessed by an experienced chest radiologist who was blinded for review.
3.2.10. He et al
He et al. [20] performed a single centre retrospective study of 82 patients who were suspected to have COVID-19 between January 10th and February 28th, 2020. All patients underwent both RT-PCR and chest CT-scan testing. Interpretation and final diagnosis of the CT imaging was completed by two trained radiologists.
3.2.11. Long et al
Long et al. [21] performed a single centre retrospective study that included 87 patients with suspected COVID-19 patients between January 20th, 2020 and February 8th, 2020. All patients underwent a CT scan and RT-PCR. All lesions identified on chest CT images were predefined and reviewed by two radiologists.
3.2.12. Ma et al
Ma et al. [22] conducted a single centre retrospective study enlisting 158 children under the age of 17 who had a social or family exposure to COVID-19, from January 21st to February 14th, 2020. RT-PCR was initial performed with a follow up chest CT scan to evaluate lung pathology. Data was collected through access to patients’ electronic medical records. CT scan results were assessed by two independent experienced radiologists. If there was a discrepancy in the diagnosis from the CT scans, consensus was reached by discussion.
3.2.13. Xie et al
Xie et al. [23] reported a single centre retrospective study that included 167 patients at high-risk of COVID-19. All available data, including clinical history, laboratory and epidemic characteristics, were collected. Two blinded radiologists independently reviewed the CT images.
3.3. Primary Outcomes
3.3.1. Discriminatory power (sensitivity and specificity)
Sensitivity to detect COVID-19 cases was reported in 13 studies enrolling 4092 patients (Figure 2). The sensitivity of CT for the identification of COVID-19 cases effectively the same as RT-PCR, with mean sensitivity values of 0.90 and 0.91 respectively (Table 2).
Key CT: Computed Tomography, RT-PCR: Reverse Transcription Polymerase Chain Reaction, COVID-19: Coronavirus Disease 2019.
Specificity in identifying COVID-19 was reported in 10 studies enrolling 3689 patients (Figure 2). The specificity for chest CT was lower than that of RT-PCR in the identification of COVID-19 cases, with mean specificity values of 0.74 and 1.00, respectively (Table 2).
* Assumes no false positive due to non-reported data.
Key TP: True Positive, FP: False Positive, FN: False Negative, TN: True Negative, CI: Confidence Interval, PPV: Positive Predictive Value, NPV: Negative Predictive Value, PLR: Positive Likelihood Ratio, NLR: Negative Likelihood Ratio.
A comparison of the discriminatory power of the two tests in identifying patients with COVID-19 from non-infected patients, based on area under the sROC, is shown in Figure 3.
Key CT: Computed Tomography, RT-PCR: Reverse Transcription Polymerase Chain Reaction.
3.3.2. Diagnostic effectiveness (accuracy)
Accuracy to detect and exclude COVID-19 cases was reported in 13 studies enrolling 4092 patients (Figure 4). There was a statistically significant difference in the identification and exclusion of COVID-19 cases favouring RT-PCR over CT (OR = 0.22, CI = 0.09 to 0.56, P = 0.001). There was a high level of heterogeneity between studies (I2 = 90%, P < 0.00001).
Key CT: Computed Tomography, RT-PCR: Reverse Transcription Polymerase Chain Reaction, COVID-19: Coronavirus Disease 2019.
3.4. Secondary Outcomes
3.4.1. Other test performance characteristics
Other test performance characteristics are summarised in Table 2. CT scan had a lower positive predictive value than RT-PCR, with mean values of 0.85 and 1.00, respectively. The mean negative predictive value was slightly lower for CT scan (0.87) versus RT-PCR (0.90). Additionally, the NLR for CT scan and RT-PCR were similar, with mean values of 0.13 and 0.15, respectively. Comparison between PCR and CT for PLR was impossible due to incomplete data.
3.4.2. Discrepancy of Findings
Ai et al. [4] (n = 1014) reported that 75% (n = 310) of patients with negative RT-PCR results and 97% (n = 583) of those with positive results had positive chest CT findings. Across all studies, only 9.1% (230/2529) of PCR positive patients had a negative CT scan and 79.2% (42/53) of patients who initially had a negative PCR but then had a positive PCR later had a CT scan suggestive of COVID-19 (Table 3).
Column 1 shows the percentage of PCR positive patients who have a negative CT scan. Column 2 shows the proportion of patients with a CT scan suggestive of COVID-19 in patients who initially had a negative PCR, but then later had a positive PCR.
Key NR: Not Reported
3.4.3. Main Chest CT Finding
According to Fang et al. [16] (n = 51), typical manifestations, such as peripheral, subpleural ground-glass opacities, usually in the lower lobes, were seen in 72% of detected cases by CT scan; atypical manifestations were found in 28% of cases. Caruso et al. [13] (n = 158) further highlighted that ground glass opacities (100%) with multi- (93%) and posterior (93%) lobe involvement, and bilateral distribution (91.4%) were the most common manifestations in COVID-19 cases confirmed by PCR. Similarly, Çinkooğlu et al. [14] (n = 185) highlighted pure ground glass opacities (82.3%) to be present in most patients followed by ground glass opacities with consolidation (32.7%) and the crazy paving pattern (21.8%). Similarly, Long et al. [21] (n = 87) found the commonest CT pattern in their cohort of COVID-19 cases was ground-glass opacities with consolidations (52.7%) in the lower lobes. Ma et al. [22] (n = 158) also reported that the most common manifestation was ground-glass opacities (58%). Other studies also highlighted that ground glass opacities and consolidation were the most common CT manifestations [12, 15, 17, 19]. However, Ai et al. [4] (n = 1014) found that consolidation and ground-glass opacities were reported in 50% and 46% of cases, respectively.
3.5. Sensitivity Analyses
Sensitivity analyses were performed for the accuracy of CT scan and RT-PCR to identify and exclude COVID-19 cases. There was no change in the direction of the effect size and the overall heterogeneity when one study was removed at a time. When the odds ratios, risk ratios and risk differences were calculated separately, there was no change in the direction of the effect size. There was similarly no change in the direction when using the random or fixed effects models.
3.6. Subgroup Analyses
3.6.1. China and Non-China Based Studies
There was a statistically significant difference between China versus non-China based subgroups in terms of accuracy in the identification of COVID-19 (P = 0.001); Figure 5a. In China based studies, there was no statistical difference between RT-PCR and CT scan (OR = 0.97, CI = 0.34 to 2.73, P = 0.95). In comparison, non-China based studies showed RT-PCR to be statistically significantly more accurate than CT (OR = 0.05, CI = 0.01 to 0.21, P < 0.0001).
Key CT: Computed Tomography, RT-PCR: Reverse Transcription Polymerase Chain Reaction, COVID-19: Coronavirus Disease 2019.
3.6.2. Larger and Smaller Studies
There was also a statistically significant difference between larger (n ≥100) and smaller (n <100) study subgroups in accuracy for the identification COVID-19 (P = 0.004); Figure 5b. Larger studies showed that RT-PCR is more accurate than CT scan with a statistically significant difference (OR = 0.09, CI = 0.03 to 0.27, P < 0.00001). Smaller studies suggested no difference in accuracy between the two tests (OR = 3.90, CI = 0.38 to 40.45, P = 0.25).
3.6.3. Adult or Mixed Age and Children Based Studies
There was a statistically significant difference between adult or mixed age versus children-based subgroups in accuracy for the identification of COVID-19 (RT-PCR was more accurate in children; P = 0.04); Figure 5c. In adult or mixed age-based studies RT-PCR was significantly more accurate than CT scan (OR = 0.26, CI = 0.10 to 0.67, P = 0.005). Children-based studies also favoured RT-PCR with statistical significance (OR = 0.01, CI = 0.00 to 0.19, P = 0.002).
3.7. Methodological Quality and Risk of Bias Assessment
Although comparability was low in all the studies included, selection and exposure were of higher quality (Table 4). Overall, all studies were of good quality based on the AHRQ standards [25].
4. Discussion
Our findings suggest that RT-PCR is significantly more accurate than CT in the initial detection of COVID-19 in the clinical context of the studies presented (i.e. hospitalised patients within a pandemic). CT had similar sensitivity, but lower specificity (higher false positives) than RT-PCR. The performance characteristics of all tests is context dependent, however, which means that the performance of CT for identifying COVID-19 will be different outside of a pandemic and in areas of the world with different endemicities. When the causes of bilateral CT changes return to the usual heterogeneous state, with SARS-CoV-2 only one of many potential causes, as was the situation pre-pandemic, the performance of CT will likely deteriorate with an even higher number of false positives, in keeping with the relatively modest PLR for CT, even within a pandemic setting, based on the study of Ai et al. [4] For patients requiring hospitalisation with an acute cardio-respiratory presentation, a negative CT scan will likely remain reassuring, particularly if RT-PCR is also negative in keeping with the slightly lower NLR for CT versus RT-PCR. Our results cannot be extrapolated to healthier SARS-CoV-2 patient cohorts who are managed on an outpatient basis or in a community setting.
Diagnostic tests always have inherent strengths and weaknesses. In the case of the COVID-19 pandemic it would be preferable to have a simple test with high sensitivity leading to a low number of false negatives thereby allowing accurate decision-making regarding, for example, who may be infectious and requires isolation in the hospital setting or home isolation in the community. Based on those who had a suggestive CT but tested negative with RT-PCR, our findings suggest that CT scan may be able to detect a greater proportion of hospitalised COVID-19 cases overall and may be useful when PCR testing is initially negative but clinical suspicion high [4]. There were relatively few patients who had a negative CT scan, but who were initially RT-PCR positive. Of the studies that reported patients who were initially RT-PCR negative, but subsequently became positive, 79.2% of patients (42/53) had a suggestive CT. It is important to note that across the studies, however, only a small proportion who were initially RT-PCR negative were systematically retested.
Overall, our results suggest that chest CT scan is inferior to RT-PCR in the initial detection of COVID-19, which supports published guidance and thus we do not advocate routine use. CT scans are subject to their inherent limitations. Firstly, they are best avoided during pregnancy due to the excessive radiation and risk of harming the foetus [26]. In addition, over-scanning patients with suspected COVID-19 can increase the risk of incidental imaging findings and an unnecessary financial burden on the healthcare system [27]. Furthermore, the use of CT-scans for suspected cases is not practical since SARS-CoV-2 has a relatively high transmission rate; each CT machine needs to be sterilised after use, which can waste time and resources as well as overworking the radiology department in times of need and hence may result in delayed reporting of other potentially life-threatening conditions. Lastly, with many countries having problems with providing their healthcare staff with sufficient personal protective equipment (PPE), sending suspected cases for a CT to confirm the diagnosis will unnecessarily consume PPE.
Given the above, and based on our results therefore, we suggest that chest CT is limited to a supplementary diagnostic role for COVID-19 in patients with suspicious symptoms, but at least one negative RT-PCR test result. In such situations, a repeat upper respiratory tract RT-PCR, which is simple to do, rather than a CT scan is also worth considering, although in our clinical experience some patients with suspicious presentations are repeatedly negative or only positive on a third or fourth test. Lower respiratory tract specimens (i.e. sputum or bronchoalveolar lavage) are more likely to be positive in this patient cohort, although most hospitalised patients are neither productive nor ventilated.
In the studies included, the most common manifestations of COVID-19 in chest CT scans were ground-glass opacities and consolidations [4, 12–17, 19, 21–22]. Based on the evidence from previous coronavirus outbreaks, the use of CT-scans may have some clinical and prognostic importance. In MERS, lung radiographic changes such as ground-glass opacifications and consolidations tend to affect the lung bases and can be used as an independent prognostic factor; patients with pleural effusion or pneumothorax have a higher mortality risk [28, 29]. Up to 30% of COVID-19 patients develop ARDS, which cannot always be confirmed on chest X-ray, making CT-scans potentially useful in such circumstances [30, 31]. Comparing the imaging findings of COVID-19 to SARS, despite both having absent features of pleural effusion and lymphadenopathy, COVID-19 tends to cause consolidation in multiple lung zones versus SARS, which tends to cause unifocal involvement [4, 32, 33].
This review provides a summary of the best available evidence using a systematic approach and evaluated the risk of bias of included studies [4, 12–23]. Although Kim et al. recently published a meta-analysis on this topic, many of the studies included were not comparative (i.e. the tests were not compared within the same patient cohorts) [34]. Kim et al. reported pooled sensitivity to be higher in CT scan (Sensitivity = 94%, 95% CI = 91% to 96%, I2 = 95%) than RT-PCR (Sensitivity = 89%, 95% CI = 81% to 94%, I2 = 83%) [34]; our study showed the sensitivity of RT-PCR to be very slightly higher. Our study synthesised comparative studies only, and also included additional recently published studies and those in preprint, so our results may more accurately represent how the investigations can be expected to perform and compare in routine clinical practice. Our study also constructed sROC curves that can overcome some of limitations of univariate pooled sensitivity and specificity analyses [35]. Additionally, we also performed important sensitivity and sub-group analyses that demonstrate how these tests may compare in different contexts, including the fallacy of basing conclusions about a test’s performance early in a pandemic on small studies in one geographic location.
However, the reported outcomes should be considered in the context of the inherent limitations. Only 13 non-randomised studies were identified enrolling a total of 4092 participants, which may not be enough to draw definite conclusions and suggesting the need for further robust prospective research. The performance of molecular tests for SARS-CoV-2 is likely to improve over time. Likewise, the use of, for example, machine learning algorithms to interpret chest CT scan appearances may more accurately confirm COVID-19 than human reporting in the future, and thereby decrease the false-positive rate of CT scanning that can be expected to increase as the pandemic wanes. Some included studies had incomplete data resulting in difficulty in comparing both interventions completely. Additionally, all but one study included a relatively small number of patients and the majority were of retrospective observational design, with all the associated caveats of such studies. It is likely that further large comparative, and possibly prospective, studies will be reported with further meta-analyses required. The between-study heterogeneity for outcomes was high for sensitivity and accuracy. Based on the data presented in the studies included, zero false positives with RT-PCR were assumed, which is unlikely to be the case in real-life.
5. Conclusions
Our study synthesised the best available comparative evidence, using a systematic approach, for the identification of COVID-19 using RT-PCR versus chest CT [4, 12–23]. Our results suggest that chest CT scan is inferior to RT-PCR in the initial detection of COVID-19 in hospitalised patients in the context of a pandemic. We suggest RT-PCR remains the primary diagnostic tool and that chest CT is only considered if there is a strong clinical suspicion of COVID-19 and a negative RT-PCR test result, providing infection prevention and control measures can be maintained. A negative CT scan is likely to be reassuring in a RT-PCR negative patient requiring hospitalisation for possible COVID-19
Data Availability
The author(s) declare(s) that they had full access to all of the data in this study and the author(s) take(s) complete responsibility for the integrity of the data and the accuracy of the data analysis.
6. Declarations
Ethics Approval and Consent to Participate
Not Applicable.
Consent for Publication
Not Applicable.
Availability of Data and Materials
The datasets generated and analysed during the current study are available from the corresponding author on reasonable request.
Competing Interests
The authors declared that they have no competing interests.
Funding
Funding was provided by Kuwait Foundation for the Advancement of Science.
Authors’ Contributions
MK contributed in study concept and design as well as acquisition, analysis and interpretation of data. A Abul, AH, A Alsaif helped in collecting and analysing data as well as drafting the manuscript. SA and MA were responsible for screening additional papers and drafting the manuscript. SA, MA and A Alsaif assessed the quality of included studies. GB was the lead supervisor for critically reviewing the project, including editing of manuscript. All authors read and approved the final manuscript.
Acknowledgements
Not Applicable.
7. References
