This is a real-life, prospective, observational study involving 260 patients. All patients received chest CT examination at the emergency department (ED) of Kaspela University Hospital, Plovdiv, Bulgaria and RT-PCR testing for suspected COVID-19 from March 27 to December 31, 2020. Additionally, the overall sample size was split in two: 1st wave of COVID-19 (n=28) starting from March 13, 2020^[12] to May 26 2020^[13] and 2nd wave of COVID-19 (n=232) starting from October 27, 2020^[14] to January 4, 2021^[15]. The timeframe for both waves is based on the official lockdown measures introduced by COVID-19 State of Emergency Measures Act originally announced on March 13, 2020 by a decision of the National Assembly as an Emergency Measures Act.

This study included patients with clinical-epidemiological suspicion of COVID-19 infection based on the manifestation of at least one of the following features: a) fever – temperature >37.8°C; b) one or more clinical findings of lower respiratory illness (e.g., cough, shortness of breath, difficulty breathing)^[16] with or without a history implying exposure to SARSCoV-2 including: (1) close contact with a confirmed case of SARS-CoV-2 disease, (2) close contact with a person with mild, moderate, or severe respiratory illness for whom a chain of transmission can be linked to a confirmed case of SARS-CoV-2 disease in the 10 days preceding the onset of symptoms, (3) travel or residential history in locations with a documented high prevalence of disease, or (4) close contact with individuals with mild-to-moderate symptoms and with a history of travel to a location with documented high prevalence of disease within 14 days prior to the CT scan. Exclusion criteria were set as follows: (a) lack of RT-PCR testing results or “gray-zone” results, (b) a time interval between the CT scan and RT-PCR testing greater than 5 days, and (c) poor/unreadable image quality of the CT scans due to motion artefacts or incomplete data image. The final outcome was expressed as hospital discharge or died.

All patients received RT-PCR laboratory tests before or after the chest CT as a reference standard for the diagnosis of COVID-19. The naso- or oropharynx specimens were obtained according to WHO recommendation.^[17] A patient with CT findings suggestive of COVID-19 and positive RT-PCR results was considered to be infected with COVID-19. A patient with negative CT findings and negative RT-PCR test was considered to be negative if no symptoms worsening or laboratory findings consistent with COVID-19 occurred.

All patients were examined with a multidetector 32-channel CT scanner (Siemens Go Up). The parameters of CT acquisition are: tube voltage 130 kV, quality ref. mAs 54, Eff. mAs 73 with CARE Dose4D dose optimization. Acquisition (mm) 32×0.7; pitch 1.5; rot. time (s) 0.80. All exams were performed in a supine position, at full inspiration without contrast medium. Two reconstructions were made – the first was with 1.5 mm slice thickness with 1.5 mm increment, Br60 Kernel, Lung window, Narrow FAST Planning Width and FAST 3D with Matrix Size 512, and the second was with 1.5 mm slice thickness with 1.5 mm increment, Br40 Kernel, Mediastinum window, SAFIRE strength 3, Narrow FAST Planning Width and FAST 3D with Matrix Size 512. The scans were observed in axial, sagittal, and coronal plane.

All patients admitted to Kaspela University Hospital underwent chest CT examination. Because there were not enough PCR facilities in Bulgaria at the start of the pandemic, samples from Plovdiv were analyzed in the city of Stara Zagora. The result from the CT examination was crucial because this was the only way to confirm the COVID-19 pneumonia suspicions. Patient’s CT scans were interpreted by the radiologist on duty and staged according to the CORADS classification.^[18] Our hospital ward employs five radiologists; three of them (P.S., M.S., and M.G.) has more than 30 years of experience and the other two (L.C. and K.D.) have more than 7 years of experience in the field of radiology; all of them are assistant professors at the Department of Diagnostic Imaging of the Medical University of Plovdiv. The CT readers were not blinded to clinical information, but the RT-PCR results were never available at the time of reading. Apart from the CORADS staging, all patients were classified according to the changes in the parenchyma as mild, intermediate, or severe. Moreover, with the accumulation of knowledge, we were able to determine the duration of the process. The patients with 25% or less affected parenchyma are classified as mild, patients with 25%–75% of affected parenchyma are classified as intermediate, and patients with 75% and higher are classified as severe. In the first wave of the pandemic, it was found that there were patients classified as CORADS 2 or patients with other than COVID-19 pneumonia, probably due to the fact that COVID-19 pneumonia was a new disease and some of the patients were initially misdiagnosed. During the second wave in autumn/winter, almost all the patients had certain changes.

COVID-19 likelihood was assessed by assigning each CT scan to a particular category of the COVID-19 Reporting and Data System (CO-RADS).^[19] The CO-RADS classification is a standardized reporting system for patients with suspected COVID-19 infection developed for a moderate to high prevalence setting based on a 6-point scale of suspicion for pulmonary involvement of COVID-19 on chest CT: CO-RADS 0 – not interpretable (scan technically insufficient for assigning a score); CO-RADS 1 – very low (normal or noninfectious); CO-RADS 2 – low (typical for other infections but not COVID-19); CO-RADS 3 – equivocal/unsure (features compatible with COVID-19 but also other diseases); CO-RADS 4 – high (suspicious for COVID-19); CO-RADS 5 – very high (typical for COVID-19); CO-RADS 6 – proven (RT-PCR positive for SARS-CoV-2).^[19]

Quantitative variables were summarized by mean and standard deviation (mean±SD) or median (25th percentile; 75th percentile), based on the sample distribution. Qualitative variables are presented as numbers/totals and percentages (n, %). The Kolmogorov-Smirnov test was applied to inform about the distribution of the patients sampled. Differences between groups were tested using the independent samples t test, Fisher exact test and z-test as appropriate. A 2-sided p-value of <0.05 was considered statistically significant. Statistical analyses were performed using SPSS Statistics v. 26 software (IBM Corp., Chicago, IL, USA).

The discriminatory power of CO-RADS was estimated by the area under the receiver operating characteristic (ROC) curve. Youden’s index was calculated to indicate the optimal cutoff value, followed by diagnostic measures estimate.

Binary logistic regression analysis was performed to test whether the severity of disease CO-RADS score was associated with age and sex. Associations were quantified by odds ratios (OR).

From March 27 to December 31, 2020, after initial symptom evaluation in the triage of Kaspela University Hospital, Plovdiv, Bulgaria, 260 patients with suspected COVID-19 were referred for chest CT. The male-to-female distribution was 1.4:1 – 150 (57.7%) males vs. 110 (42.3%) females (z=3.5; p=0.0004). The median age was 55 yrs (range 46-65 yrs). We set a CO-RADS score ≥4 as the optimal threshold to discern between patients with PCR+ from those with PCR− results.^[20] We classified 212 (81.5%) patients to CO-RADS –≥4 and the number of false-positive chest CT findings in patients without COVID-19 was 6 (2.8%), confirmed by negative RT-PCR test. PCR-positive patients were 236 (90.8%) and 30 (12.7%) of them had CO-RADS ≤3. Discharged from the hospital were 247 (95.0%) patients, the rest died in the COVID-19 intensive care unit (ICU).

The distribution of patients’ characteristics by RT-PCR results is presented in Table 1. No differences were found in the mean age between RT-PCR positive and negative patients (t=1.4, p=0.164) as such were not observed after split by sex neither between negative vs. positive males (t=0.67, p=0.507), nor between females (t=0.92, p=0.380). RT-PCR positive patients who were reported dead (n=13) had a higher mean age (64.69±12.07 yrs) than the mean age of discharged patients (54.86±14.81 yrs.) (t=2.34, p=0.020).

In the first wave of the pandemic, it was found that there were patients classified as CO-RADS 2 or patients with other than COVID-19 pneumonia, probably due to the fact that COVID-19 pneumonia was a new disease and some of the patients were initially misdiagnosed. In the second wave, almost all patients had certain changes.

The most common finding was ground-glass opacities, which were mostly present in both lungs and were only seen in a few cases in only one lung. In the early stage of the disease, the opacities had low-to-intermediate density while in the later stages, they had higher density. Interlobular septal thickening, crazy-paving patterns, and dilatation of the distal small pulmonary vessels were also observed as the disease progressed. Most often, the changes affected the middle and lower parts of the lungs. Enlarged lymph nodes and pleural effusions were not observed in our patients.

In Fig. 1A, we show a patient with typical COVID symptoms – loss of smell and taste, no fever or cough. The exam was performed 4 days after the onset of symptoms. The CT scan showed no pathological findings. Fig. 1B shows the CT examination of a patient with positive PCR, fever, cough, and shortness of breath. The imaging finding was ground-glass opacity in the left lung. The CT examination in Fig. 1C shows typical COVID-19 changes – ground-glass opacity in the right lung, the lung parenchyma is affected around 10%. Figs 1D through 1I show the CT scans of patients with the typical changes of ground-glass opacities, interstitial thickening, crazy-paving patterns, and dilated distal pulmonary vessels. Depending on the stage of the disease, initially, the ground-glass opacities have low density, and then, after the acute stage, the opacities have mainly two types of progression – the density of the opacities can decrease or get higher but affect less of the parenchyma. Patients who had a severe disease needed at least 6 months to completely recover from the COVID pneumonia. Some patients do not fully recover and have permanent fibrotic changes in the lung parenchyma.

Figs 2 and 3 show the scans of a 49-year-old patient with COVID-19 with thrombosis of the cavernous sinus that developed as a complication. The patient was hospitalized with positive PCR and CT examination showing typical changes. After a week in the hospital, the patient complained of severe headache and underwent an emergency native CT examination, which showed no changes. This patient lost his vision in the right eye the next day, with exophthalmos and swelling of the soft tissues around the right eye. CT angiography was performed showing a hyperdense structure in the right cavernous sinus thrombosis. On the arterial and venous series, thrombosis of the right sinus cavernosus is presented. The internal carotid arteries and other cerebral arteries were normal.

In the second wave of the pandemic, there were patients that had already been cured of COVID-19 infection and such patients received control CT scans. We found that patients were healing generally in two ways: some patients had low-density ground-glass opacities that looked like those in acute COVID-19 pneumonia – such patients were clinically examined and their anamneses were taken, while in other patients, the changes became denser and affected less lung volume compared to the first CT examination. We recommended that such patients should have a control CT scan in no less than 6 months to observe the changes and find whether the changes persisted and if there was pneumonia. It was found that the less volume was affected, the faster the healing process was.

The ROC curves analysis (Fig. 4) demonstrated that CO-RADS ≥3 was the optimal cutoff for discriminating between a positive and negative PCR (Youden’s index J=0.67), with an AUC of 0.825 (95% CI 0.72-0.93), a sensitivity of 91.9% (95% CI 87.7%-95.1%), specificity of 75.0% (95% CI 53.3%-90.2%), negative predictive value of 99.1% (95% CI 98.5%-99.4%), positive predictive value of 24.2% (95% CI 13.8%-39.0%), and an accuracy of 76.4% (95% CI 70.7% to 81.4%). The positive likelihood ratio was 3.78 (95% CI 1.84-7.36), and the negative likelihood ratio was 0.11 (95% CI 0.07-0.18). The interval likelihood ratio was 2.34 (95% CI 0.3-16.6) for CO-RADS 4 and 3.72 (95% CI 1.7-8.1) for CO-RADS 5. The overall model quality was 72%.

Detailed patients’ characteristics split by the established CO-RADS optimal cutoff are reported in Table 2. Additionally, we explored the difference in the mean age between males (n=133) and females (n=90) by the CO-RADS split and proved that although younger approximately by 5 yrs than females (54.70±14.57 vs. 59.43±13.97; t=2.24, p=0.016), males had higher dead rate 76.9% (n=10). The mean age of dead males was 11 yrs higher than that of the discharged males (64.60±11.29 vs. 53.89±14.55; t=2.27, p=0.025). Statistically significant differences were found between the age groups listed in Table 2 and the CO-RADS categories. A difference existed also between the severity (CO-RADS ≥3) of those patients below (31.8%) and above (68.2%) 50 years of age (z=7.7; p<0.0001). The same result about severity (CO-RADS ≥3) was observed for the patients below (68.2%) and above (31.8%) 65 years of age (z=7.7; p<0.0001), but in the opposite direction – a significantly smaller proportion of patients ≥60 yrs experienced severe symptoms of the disease.

The overall sample size was split in two, regarding the peaks of the epidemic curve of COVID-19 incidence in Bulgaria. Overall, four Ministerial Ordinances determined the beginning, length, and end of the two COVID-19 lockdowns.‌^[12–15] They basically defined the two incidence waves. The first lockdown measures were introduced on 13 March 2020 after the registration of the first cases of COVID-19. Due to the strict measures during the first wave between 13 March and 26 May 2020, there were very few cases of COVID-19 registered with 2443 cumulative cases for the period of which 106 for the Plovdiv region.^[21] Тhe beginning of the second wave was on 27 October 2020 and it finished approximately by the beginning of 2021. During this period, the increase of registered COVID-19 cases was very distinctive, with 197384 cumulative number of cases of which 18474 cases were in the Plovdiv region. The incidence in this period peaked in November with the highest registered incidence in our country from the beginning of the pandemic – 660.33/100 000.^[21] The sample size for the first wave was 28 patients and 232 patients for the second wave. Table 3 summarizes patient characteristics measured throughout the two waves.

Fig. 5 illustrates the CO-RADS categories distributed by age and RT-PCR results in each wave. The mean age (57.95±13.82) of the second wave RT-PCR positive patients assigned to CO-RADS 5 was approximately 7 yrs less than the one measured in the first wave RT-PCR positive patients with CO-RADS 5 score (65.80±7.60), although the difference failed to reach statistical significance (p=0.081).

The second wave logistic regression model was statistically significant: χ²(2)=26.04, p=0.000, explained 23.0% (Nagelkerke R2) of the variance in disease severity (measured by binary outcome variable of CO-RADS <4 or CO-RADS‌ ≥4) and correctly classified 90.9% of cases. Males were 4.13 times more likely to be diagnosed with CO-RADS ≥3 score than females. Increasing age was associated with increased likelihood of being classified with higher CO-RADS scores. The results of the logistic regression are shown in Table 4.