Comparison between CT and RT-PCR in a cohort of symptomatic patients with suspected COVID-19 pneumonia during the outbreak peak in Italy

Objective: To assess sensitivity and speci�city of CT vs RT-PCR for the diagnosis of COVID-19 pneumonia in a prospective Italian cohort of symptomatic patients during the outbreak peak. Methods: In this cross-sectional study we included all consecutive patients who presented to the ER between March 13 th and 23 rd for suspected COVID-19 and underwent both CT and RT-PCR within 3 days. Using a structured report, radiologists prospectively classi�ed CTs in highly suggestive, suggestive, and non-suggestive of COVID-19 pneumonia. Ground-glass, consolidation, and visual extension of parenchymal changes were collected. Three different RT-PCR-based reference standard de�nitions were used. Oxygen saturation level, CRP, LDH, and blood cell counts were collected and compared between CT/RT-PCR classes. Results: The study included 696 patients (41.4% women; age 59±15.8 years): 423/454 (93%) patients with highly suggestive CT, 97/127 (76%) with suggestive CT, and 31/115 (27%) with non-suggestive CT had positive RT-PCR. CT sensitivity ranged from 73% to 77% and from 90% to 94% for high and low positivity threshold, respectively. Speci�city ranged from 79% to 84% for high positivity threshold and was about 58% for low positivity threshold. PPV remained ≥ 90% in all cases. Ground-glass was more frequent in patients with positive RT-PCR in all CT classes. Blood tests were signi�cantly associated with RT-PCR and CT classes. Leukocytes, lymphocytes, neutrophils, and platelets decreased, CRP and LDH increased from non-suggestive to suggestive CT classes. Conclusions: During the outbreak peak, CT presented high PPV and may be considered a good reference to recognize COVID-19 patients while waiting for RT-PCR con�rmation.


Introduction
The clinical spectrum of SARS-CoV-2 infection ranges from asymptomatic infection to a disease (COVID-19) that can lead to potentially fatal respiratory failure [1,2].
The outbreak of SARS-CoV-2 has rapidly spread worldwide since the end of 2019, with Italy being one of the rst and most affected countries facing the epidemic out of China [3,4].
The diagnosis of COVID-19 currently relies on reverse transcription-polymerase chain reaction (RT-PCR) conducted on oropharyngeal and/or nasopharyngeal swabs.However, while false positives are conceivably rare, false negatives can occur, even in patients with pneumonia, who may have negative nasal/oropharyngeal samples but positive lower airway samples.The true clinical sensitivity of RT-PCR is thus unknown [5,6].Moreover, the huge demand for tests is compromising their availability in some areas [5] and creating frequent delays in diagnosis con rmation, with consequences on timely treatment, isolation, and contact tracing.
Computed tomography (CT) often shows typical ndings in COVID-19 pneumonia, especially bilateral patchy ground-glass opacities and consolidation, with a predominantly peripheral distribution.Crazypaving pattern, peripheral vessel enlargement, and ndings of organizing pneumonia such as reverse halo sign have also been described [7][8][9][10][11].However, CT may show no abnormalities, especially early after symptom onset [12], and CT ndings are not speci c, signi cantly overlapping with other infections [13].
The role of CT in the COVID-19 epidemic is debated.Its use as a screening tool has been proposed [14,15] but is highly discouraged by major radiology societies in Western countries [10,16].In China the diagnostic criteria for COVID-19 include chest CT as major evidence [17], while European and American societies underline the need for RT-PCR for diagnostic con rmation [10,16], even though they suggest repetition of RT-PCR in cases of suggestive CT ndings in symptomatic patients [10].Finally, in patients with respiratory symptoms such as dyspnea and desaturation, CT may help stratify disease severity and patient prognosis [10].The Fleischner Society has recently released a multinational consensus statement suggesting adapting the use of chest imaging to different clinical scenarios according to the severity of clinical features, pre-test probability of COVID-19, and resource constraints [18].
The aim of this study was to assess the sensitivity and speci city of CT vs RT-PCR for the diagnosis of COVID-19 pneumonia in a prospective Italian cohort of symptomatic patients presenting to the emergency room (ER) for suspected COVID-19 during the outbreak peak.We also compared relevant blood tests in classes of patients with different combinations of CT and RT-PCR results.

Setting
In the Reggio Emilia province (Northern Italy, 532,000 inhabitants, six hospitals), the rst case of SARS-CoV-2 infection was diagnosed on February 27, 2020.Up to March 24, there were 1200 RT-PCR-con rmed cases and the epidemic was still spreading.The study was approved by the Area Vasta Emilia Nord Ethics Committee on 04/07/2020 (protocol number 2020/0045199).Patients' written consent to publish their images was obtained, and patients' informed consent to participate in the study was obtained whenever possible, given the retrospective nature of the study.

Study design
This was a cross-sectional study assessing sensitivity and speci city of CT for COVID-19 pneumonia at two different thresholds of suspicion, using RT-PCR as the reference standard.

Study population
All consecutive patients who presented to the Reggio Emilia province ERs between March 13 th and March 23 rd for suspected COVID-19 and underwent both CT and RT-PCR were eligible.Subjects with a time gap between CT and RT-PCR > 3 days were excluded.
During the COVID-19 outbreak, the diagnostic protocol for these patients included nasopharyngeal and oropharyngeal swabs for RT-PCR, blood tests, chest X-rays, and CT scan in cases of suggestive X-ray ndings or negative X-rays but highly suggestive clinical features.A structured CT report was introduced on March 13 th .

Reference standard
Two issues hamper the measure of CT accuracy for COVID-19.Firstly, the clinical sensitivity of RT-PCR, our reference standard, although not yet quanti ed, is not 100% [5].Secondly, the target condition of our index test, i.e., CT, is viral pneumonia, while RT-PCR target condition is SARS-CoV-2 infection.
While the second issue cannot be easily solved, to overcome the rst problem we used different de nitions of reference standard:

CT analysis and structured reporting
During routine reporting, each radiologist completed both the usual radiology report as well as a structured report about the probability of COVID-19 pneumonia based on CT ndings (highly suggestive, suggestive, non-suggestive) (Fig. 1), the presence/absence of ground-glass opacities and consolidations, and the extension of pulmonary lesions using a visual scoring system (< 20%, 20-40%, 40-60%, and > 60% of parenchymal involvement) (Fig. 2).Swab results were unknown when reporting, so radiologists were blinded to RT-PCR.However, they were frequently informed of blood test results and of patients' clinical features.

Blood tests
When available, C-reactive protein (CRP) level, LDH, total leukocyte, lymphocyte, neutrophil, and platelet counts measured on ER admission were collected.The tests were carried out in the Hospital Clinical Laboratories with routine automated methods.Oxygen saturation level (SpO2) was also collected for patients who had it measured before being provided with oxygen support.
These tests were included since previously associated with COVID-19 diagnosis, severity and prognosis: increased CRP re ects host in ammatory response, along with increased total leukocyte and neutrophil counts; elevated LDH concentrations may be a sign of end-organ damage; decreased platelet count may be associated with an underlying coagulopathy; lymphopenia may represent a concomitant immune dysfunction and has been associated with increased disease severity and worse prognosis; SpO2 provides information on lung damage and functionality [19,20].

Statistical analyses
CT scan sensitivity, speci city, and positive predictive value (PPV) for COVID-19 pneumonia were computed for two different thresholds: highly suggestive only, and highly suggestive plus suggestive ndings.
Accuracy measures according to the three reference standard de nitions reported above were calculated with relative 95% con dence intervals (CI) computed on the exact binomial distribution.

Distribution of clinical characteristics across groups of CT and RT-PCR results is reported. Associations between clinical characteristics and CT and RT-PCR classes were measured through Pearson's chi2 and
Fischer's exact distribution.For blood tests, we report mean (±SD) and median (IQR) strati ed by groups of CT and RT-PCR results.Comparisons between groups were conducted with one-way ANOVA and linear regression models adjusted for sex and age.
P values are reported as continuous measures and no pre xed signi cance threshold was used.
We used Stata 13.0 SE (Stata Corporation, Texas, TX) software package.

Population
After excluding 6 patients with a > 3-day interval between CT and RT-PCR, we included 696 patients who underwent CT and RT-PCR in the Reggio Emilia provincial ERs between March 13 th and 23 rd for suspected COVID-19 (Fig. 3).Of these 696 patients, 288 (41.4%) were women.Overall mean age was 59 years (SD 15.8), 58.3 years (SD 16.5) for women, and 59.5 years (SD 15.2) for men.
The proportion of women decreased with the increase in CT suspicion, while age was slightly higher in the group with suggestive CT (Table 1).

CT diagnostic accuracy
The sensitivity, speci city, and PPV of CT at two different thresholds (highly suggestive ndings only and highly suggestive plus suggestive ndings) with respect to the three different reference standards described in the Methods section are reported in Table 3.

Estimation of RT-PCR sensitivity in patients with pneumonia
Our study can also provide a plausible estimate of RT-PCR sensitivity for SARS-CoV-2 in patients with pneumonia.In the hypothesis that all patients with highly suspicious CT have COVID-19 pneumonia, there would be 31 RT-PCR false negatives in this group, for a sensitivity of 93% (423/451).If we only consider RT-PCR false negatives the proportion of patients who retested positive, false negatives would be approximatively 14 out of 451, for a sensitivity of 97%.

Blood tests
Means (SD) and medians (IQR) across CT classes are reported in Table 4.
Among RT-PCR-negative patients, the values of total leukocyte, lymphocyte, neutrophil, and platelet counts, and CRP level were higher than among RT-PCR-positive patients.Total leukocyte, lymphocyte, and platelets counts decreased, whereas CRP and LDH increased from non-suggestive to suggestive and/or highly suggestive CT.Patients experienced similar SpO2 values across CT and RT-PCR groups when the CT was suggestive or highly suggestive, while in patients with non-suggestive, CT SpO2 was higher, particularly for those who tested positive at RT-PCR.

Discussion
In a large sample of consecutive patients presenting to the ER for suspected pneumonia during the peak of the SARS-CoV-2 outbreak in Northern Italy, we estimated CT sensitivity for COVID-19 pneumonia to be between 73% and 77% when adopting a high positivity threshold, which corresponded to a speci city of between 79% and 84%.When adopting a lower positivity threshold, CT sensitivity was between 90% and 94%, but speci city decreased to 58%.Nevertheless, given the very high prevalence of COVID-19 during the epidemic peak, the proportion of patients with SARS-CoV-2 infection among the CT-positive patients (PPV) was always equal to or higher than 90%, whatever the positivity threshold adopted.
Despite the fact that none of the included blood tests could de nitely guide diagnosis, they were signi cantly associated with RT-PCR and CT classes.Leukocyte, lymphocyte, neutrophil, and platelet counts decreased, and LDH increased both in RT-PCR-positive patients and in highly suggestive and/or suggestive CT scans.CRP decreased in RT-PCR-positive compared to RT-PCR-negative patients and increased in highly suggestive and suggestive CT scans.SpO2 was slightly lower in patients with suggestive and highly suggestive CT results.
A recent study has assessed CT diagnostic performance in Italy by using only one threshold (positive versus negative CT scan) and two RT-PCR within 24 hours as the reference standard in 158 patients.
In the largest study conducted in China on over 1000 patients, CT scan presented similar sensitivity (97%) but lower speci city and PPV (25% and 65%, respectively) [22].Given the lower RT-PCR positive rate (59.2%) and high rate of RT-PCR-negative CT-positive patients who were eventually classi ed as probable COVID-19 according to their global clinical course (81%), it is plausible to think that the RT-PCR false negative rate was higher in the study by Ai et al., leading to underestimating CT speci city and PPV.
Accordingly, in our study, when applying different reference standards so as to reduce RT-PCR false negative rates, speci city rose to 84% and PPV to 96% at the higher CT suspicion level.
Furthermore, in China, CT scan has been proposed and is used as a screening tool in asymptomatic or mildly symptomatic patients [14,23], who are more likely to have low viral loads and false negative RT-PCR results [24].Thus, the combination of an increase in clinical sensitivity of the reference standard, the application of CT only to symptomatic cases presenting in ER, and a greater knowledge of COVID-19 CT ndings in the radiology community thanks to the pioneering studies from China may explain the higher CT speci city estimate in our study.
It is not surprising that in the outbreak peak phase, with very high disease incidence, the proportion of RT-PCR positive among non-suspicious CT was quite high, as the patient has not yet developed pneumonia or will never develop pneumonia even if infected.In fact, almost one half of these patients in our study had no CT abnormalities.

Limitations and strengths
The main limitation of this study, in common with all the others with a similar aim, is that CT and RT-PCR target two different conditions, e.g., COVID-19 pneumonia and SARS-CoV-2 infection; using one as the reference standard of the other thus introduces methodological challenges.Any RT-PCR false negative will result in an overestimation of CT false positives.Nevertheless, our estimate of RT-PCR sensitivity in patients with pneumonia ranges between 93% and 97%.Several biases can affect this estimate because retesting of RT-PCR-negative CT-positive cases was not systematic in our study.
Radiologists were not blinded to clinical presentation and blood tests, which is common in real practice, but it should be considered if the results were applied to settings where this information is not yet available.Also, as a consequence of how the reporting was structured and of the need to be rapid and concise during the epidemic phase of the disease, we collected only some of the CT ndings of COVID-19.This approach is surely less accurate than the retrospective review performed by experienced thoracic radiologists adopted by most studies, but it is more representative of the real-life diagnostic process.

Implications for practice
The results of this study are not intended to produce evidence to be generalized to all clinical scenarios.Nevertheless, we show high sensitivity and high positive predictive value of CT for COVID-19 pneumonia in the epidemic setting.
The speci c phase during which the study was conducted re ects one of the scenarios proposed by the Fleischner Society [18], characterized by a high number of symptomatic patients presenting to the ERs, high pre-test probability, and unavailability of rapid virological testing.
Further studies are needed to assess speci city and PPV in lower prevalence settings.Also, a structured report for suspected COVID-19 patients may help in monitoring the proportion of positive results and the PPV of different positivity thresholds in different phases of the epidemic.Moreover, including CT from the pre-epidemic period [13] may help in assessing speci city to be projected in periods when COVID-19 could cyclically shift from endemic to epidemic phases.

Conclusions
During the outbreak peak, CT presented a high PPV and may thus be considered a good reference to help clinicians to recognize and triage COVID-19 patients while waiting for RT-PCR diagnostic con rmation.
Our results also con rm that in case of negative RT-PCR and highly suggestive CT ndings, RT-PCR should be repeated; the patient should remain isolated, given the high probability of RT-PCR false negatives in this group.

Figures
Figures

Figure 2 Visual
Figure 2
CT ndingsGround-glass opacities were present in almost all patients with highly suggestive (453/454) and suggestive (123/127) CT ndings, and in 40% (44/115) of patients with non-suggestive ndings.When considering subgroups of RT-PCR results, in all the three CT classes ground-glass opacities were more frequent in patients with positive RT-PCR.Consolidation was present in 279/454 (61.5%) patients with highly suggestive, 76/115 (59.8%) patients with suggestive, and 41/115 (35.7%) patients with non-suggestive CT ndings.The distribution of consolidation among subgroups of positive and negative RT-PCR patients in the three CT classes varied, being more frequent in negative RT-PCR patients in the intermediate CT class, i.e., suggestive.A limited (< 20%) extension of pulmonary lesions was more frequently present in patients with suggestive and non-suggestive CT ndings.Parenchymal extension was not estimated in 55/115 patients with nonsuggestive ndings; 41/55 had a completely normal CT scan, while 14/55 presented pleural effusion without parenchymal involvement.Interestingly, among the 31 patients with non-suggestive CT and positive RT-PCR, 13 patients had no CT abnormalities and 1 had only pleural effusion (Table

Table 3 :
CT accuracy PCR; if RT-PCR was not repeated the patient was considered negative; 3) as in the second scenario, but negative patients who were not retested with RT-PCR were classified as positive or negative in the same proportion of patients who were actually retested in each group of CT-based probability of COVID-19.
Blood tests expressed in mean (SD) and median (IQR) according to CT and RT-PCR classes.P adj*: P value for coef. of category in a linear regression adjusted for sex and age (< 60, ≥ 60).CRP: C-Reactive Protein.LDH: Lactate dehydrogenase.SpO2: oxygen saturation level.