The current reference standard for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection is reverse transcriptase-polymerase chain reaction (RT-PCR), which, however, has limited sensitivity.1 Chest x-ray (CXR) and computed tomography (CT) can play a role in the diagnostic pathway,1,2 but to date, only a few small-scale studies have reported on CXR.3 Our aim was therefore to assess the diagnostic performance of CXR performed at presentation at our hospital emergency room (ER) in a relatively large population, negotiating known sensitivity shortcomings of the initial RT-PCR by building a composite reference standard.
This IRB-approved retrospective observational study was carried out at IRCCS Policlinico San Donato (San Donato Milanese, Italy), including consecutive patients presenting to the ER for suspected SARS-CoV-2 infection from February 24 to April 8, 2020, all of them undergoing both digital CXR and nasopharyngeal swab for RT-PCR. Anteroposterior bedside CXR was performed in the ER within 12 hours from admission. CXRs were classified as positive or negative according to original radiologic reports. We took RT-PCR as the reference standard, considering laboratory-confirmed cases as those with positive or weakly-positive swab. RT-PCR-negative patients, without a second swab performed at our institution, were contacted by phone to ask whether, after discharge, they underwent a second swab at another institution, or they were hospitalized for COVID-19 infection in another institution, or whether they had COVID-19-suggestive symptoms during quarantine. Diagnostic performance indexes for CXR were presented as sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and their 95% confidence interval (CI). Statistical analyses were carried out using Microsoft Excel 2019 (Microsoft Corporation, Redmond, WA).
We included in this study 535 patients with concomitant CXR and swab at ER admission (aged 65±17 y, mean±SD, 340 males). The initial swab was positive in 398/535 cases (74.4%) and negative in 137/535 (25.6%). Forty swab-negative patients had a second swab, 9 (22.5%) with a positive result, and 31 (77.5%) confirmed as negative. The composite reference standard converted one weakly positive swab to a negative case, considering a second negative swab performed 5 days after at another institution. Eleven negative swabs were converted to positive cases, 3 of which were because of a third positive swab within 8 days. This finally resulted in 408 positives and 127 negatives at the composite reference standard. Using the composite reference standard, CXR found 363 true positives (Fig. 1A), 45 false negatives (Fig. 1B), 77 true negatives (Fig. 1C), and 50 false positives (Fig. 1D), resulting in an 89.0% sensitivity (95% CI, 85.5%-91.8%), 60.6% specificity (95% CI: 51.6%-69.2%), 87.9% PPV (95% CI: 84.4%-90.9%), and 63.1% NPV (95% CI: 53.9%-71.7%).
As highlighted by various authors,1–3 the role of CXR in the diagnosis of COVID-19 must be driven by several factors: (1) lack of an immediate reliable molecular diagnosis; (2) possible limited availability of CT and difficulties in fast sanitization of CT rooms, also considering the need to use CT for other-than-COVID-19 diseases; (3) ease of CXR performance in isolated rooms in the ER; and (4) high prevalence of COVID-19 in the presence of pandemic outbreak. The last factor, that is, a very high pretest probability, associated with disease spectrum skewed to a high disease severity and strong reduction in the prevalence of non-COVID-19 pneumonia, boosted bedside CXR diagnostic performance.4 In our experience, limited to one center in Lombardy, Italy, and to real-life reporting without independent image review, CXR showed a much higher sensitivity than previously reported,3 approaching that of CT and even surpassing its reported specificity.5 In conclusion, also considering that CXR diagnostic performance could be further enhanced by artificial intelligence applications,6 the adoption of CXR imaging alongside RT-PCR for COVID-19-suspected patients’ triaging can warrant a safe and efficient workflow.
The authors thank their following colleagues at IRCCS Policlinico San Donato for their contribution to the clinical work, which allowed to plan and write this article: Pietro Bertolotti, Bijan Babaei Paskeh, Giuseppe Buragina, Luca Alessandro Carbonaro, Saverio Chiaravalle, Laura Menicagli, Cristian Giuseppe Monaco, and Riccardo Spairani.
1. Rubin GD, Haramati LB, Kanne JP, et al. The role of chest imaging in patient management during the COVID-19
pandemic: a multinational consensus statement from the Fleischner Society. Radiology. 2020. Doi:10.1148/radiol.2020201365
2. Sverzellati N, Milanese G, Milone F, et al. Integrated radiologic algorithm for COVID-19
pandemic. J Thorac Imaging. 2020. Doi:10.1097/RTI.0000000000000516
3. Wong HYF, Lam HYS, Fong AH-T, et al. Frequency and distribution of chest radiographic findings in COVID-19
positive patients. Radiology. 2020. Doi:10.1148/radiol.2020201160
4. Leeflang MMG, Rutjes AWS, Reitsma JB, et al. Variation of a test’s sensitivity and specificity with disease prevalence. Can Med Assoc J. 2013;185:E537–E544.
5. Caruso D, Zerunian M, Polici M, et al. Chest CT features of COVID-19
in Rome, Italy. Radiology. 2020. Doi:10.1148/radiol.2020201237
6. Hurt B, Kligerman S, Hsiao A. Deep learning localization of pneumonia
. J Thorac Imaging. 2020;35:W87–W89.