Discussion
Among positive RT-PCR COVID-19 patients, bronchial distortion and total lung involvement volume ≥ 50 %, are the only signs associated with a fatal outcome, as shown in Figures 3a, 3b and 4. We also found that vascular dilatation, bronchial distortion, lesional expansion (total lung involvement ≥ 50 % or ≥ 4 damaged lobes) are significantly associated with invasive endotracheal ventilation or ICUH. Maybe some of these signs are not correlated with death due to our sample size. Vascular dilatation, as shown in Figures 3b and 4, could be due to endothelial lesion with inflammatory vasodilatation or maybe to parenchymal retraction associated with fibrosis. The air bubble sign is more frequent in invasive endotracheal ventilation, it could correspond to a bronchioloalveolar dilatation linked with fibrosing damages or a previous existing cyst revealed by the diffuse infection. The significantly higher frequency of interlobular and peribronchovascular thickening in the invasive endotracheal ventilation group could be due to interstitial oedema observed in congestive heart failure (6/109 patients) and are not specific of COVID-19 infection. However, contrary to Zhao et al. , lymphadenopathy and pleural effusion were not associated with a pejorative evolution in our cohort.12
We also have observed a compatible pattern with organized pneumonia (OP) in more than 50 % of patients : an example is shown in Figure 3c. Association between OP and fibrosis lesions has been observed in a pulmonary autopsy of one COVID-19 patient as a histologic pattern of acute fibrinous and organizing pneumonia, by Copin et al. , consistent with the CT aspect.25
Frequency and type of chest CT signs are similar to other previous studies except for consolidation and bronchial distortion, more prevalent in our study probably due to the clinical severity of our patients.13,20–23 In a previous metanalysis, Salehi et al.  have described 80 % of air bronchograms in COVID-19 lung damages.24 Rather than air bronchograms, it seems more pertinent to use bronchial distortion as a typical sign of secondary fibrosing damage observed in severe COVID-19 pneumoniae.
All the CT signs must be analysed independently from each other: indeed, if the presence of ground glass opacities (frequent sign whatever clinical presentation) was not a pejorative evolution predictive sign, its expansion over 50 % of volume the lung was predictive of a pejorative evolution as shown in this study.20,24,26
We have found that chest CT has a strong diagnostic value in COVID-19 infection comparing to RT-PCR, with a sensibility of 93,6 %, specificity of 85,8 %, PPV of 75 % and NPV of 96,7 %, as described in previous studies. Our sensibility was concordant with other studies, e.g. 97 % in Ai et al. 10,29 However, our specificity and PPV are better, perhaps due to the recruitment of our patients in terms of clinical severity.
Figure 2 shows that the proportion of different CT signs increases during the second week and decreases after, so it confirms that pulmonary extension is maximal around the 10th day.27 Furthermore, we noted that all the typical COVID-19 CT signs were present during the first week in similar proportions. Contrary to Pan et al. , we haven’t observed a relative predominance of ground glass opacities in the first week or consolidation predominance in the second week (perhaps due the exclusion of severe patients in these studies).28Indeed, several kinds of CT patterns seem to exist in COVID-19 pneumonia, rather than a unique stereotyped chronological evolution, depending on the type and degree of histological damage.
This study has some limitations. It is a retrospective study. Only one CT scan was included per patient. No CT aspect evolution analysis was performed; however, the rapidity of symptoms evolution could be another predictive factor of pejorative evolution. Other futures studies should consider this point. It is possible that some signs are not associated with pejorative evolution because of a relatively small number of deaths.