Perfusion Abnormalities Shown On Subtraction CT Angiography In Apparently Normal Lungs And Their Impact On Patient Outcome. A Prospective Cohort Study: What Could We Be Missing?

Background: COVID-19 pneumonia seems to affect the regulation of pulmonary perfusion. In this study, through iodine distribution maps obtained with subtraction CT angiography, we quantified and analyzed perfusion abnormalities in patients with COVID-19 pneumonia and correlated them with clinical outcomes. Methods: : 205 patients were included in this cohort, from two different tertiary-care hospitals in Chile. All patients had RT-PCR confirmed SARS-CoV-2 infection. CT scans were performed within 24 h of admission, in supine position. Airspace compromise was assessed with CT severity score, and the extension of hypoperfusion in apparently healthy lung parenchyma with perfusion score. CT severity and perfusion scores were then correlated with clinical outcomes. Multivariable analyses using Cox Proportional Hazards regression were used to control for clinical confounders. Results: Fourteen patients were excluded due to uninterpretable images. This left 191 patients, 112 males and 79 females. The mean age was 60.8±16.0 years. The median SOFA score on admission was 2 and average PaFi ratio was 250±118. Patients with severe perfusion abnormalities showed significantly higher SOFA scores and lower Pa/Fi ratios when compared to individuals with mild or moderate anomalies. Severe perfusion abnormalities were associated with an increased risk of intensive care unit (ICU) admission and the requirement of invasive mechanical ventilation (IMV). Conclusion: Patients with severe perfusion anomalies have a higher risk of admission to the ICU and IMV. Perfusion alterations could be considered as an independent risk factor in patients with COVID-19 pneumonia. Summary: Statement: Lung perfusion abnormalities in patients with COVID-19 pneumonia were associated with admission to Intensive Care Unit and requirement of invasive mechanical ventilation. Perfusion abnormalities could be considered as an independent risk factor in patients with COVID-19 pneumonia.


Background
Infection caused by SARS-CoV-2 progressed into a global pandemic, and over one year and a half after its onset more than 172 million infected patients and more than 3,7 million deaths worldwide have been reported at the time of this submission [1].
In general, COVID-19 is considered self-limited, and many of those infected have mild symptoms or appear to be asymptomatic; however rapid progression to acute hypoxemic respiratory failure has been reported in up to 20% of pneumonia cases [2]. Severe gas exchange impairment can occur even in early stages, with only minor lung airspace disease [3][4][5]. This may suggest that the shunt associated with the gasless lung parenchyma is not su cient to explain this hypoxemia. SARS-CoV-2 seems to affect the regulation of pulmonary perfusion, with clinical impact early in the course of the disease [4,6].
Angiotensin-converting enzyme 2 (ACE2) is an important regulator of the renin-angiotensin system (RAS), and SARS-CoV-2 binds to host ACE2 as the functional receptor for invasion of human cells [7]. ACE2 is found mostly in the basal layer of the non-keratinizing squamous epithelium of nasal and oral mucosa, heart, kidneys, intestines, lungs and is widely distributed in the endothelium, and it can be key since early stages of lung perfusion disorders [8,9].
In this prospective cohort, we evaluated and quanti ed lung perfusion disturbances in areas of apparently healthy lung parenchyma in conventional chest CT images in patients with COVID-19 pneumonia as a severity predictor, through iodine distribution maps obtained with subtraction CT angiography (sCTA), and correlated perfusion abnormalities with clinical outcomes. Our goal is also to identify and propose perfusion alterations as an independent COVID-19 pneumonia severity predictor.

Patient cohort
This study was approved by the Institutional Review Board at the Hospital Naval Almirante Nef and written informed consent was waived. A prospective cohort study was undertaken to assess the prognostic ability of lung perfusion using sCTA. Patients (> 18 years old) with con rmed SARS-CoV-2 infection by real-time reverse transcription polymerase chain reaction (RT-PCR) that required hospitalization were consecutively enrolled at two hospitals between June and September 2020. These centers included the Hospital Naval Almirante Nef (Viña del Mar, Chile) and Hospital San Martín de Quillota (Quillota, Chile). These facilities are tertiary-care hospitals able to provide on-site CTA scans and intensive care management of severe COVID-19 cases when needed. Every participant had RT-PCR con rmed SARS-CoV-2 infection. In all cases, a basic clinical and demographic pro le was obtained that included information regarding sex, age, duration of clinical symptoms, Pa/Fi ratio at baseline and SOFA score. Baseline lung CT characteristics were recorded as well such as the predominant imaging pattern (ground-glass opacities, consolidation or mixed pattern), the extension of involved lung, presence of pleural effusion, evidence of right-ventricular overload and pulmonary embolism. These data were recorded along with sCTA features (see below) in an anonymized registry.
All participants were followed-up until hospital discharge searching for the development of any of the study endpoints. These endpoints included intensive-care unit (ICU) admission, initiation of invasive mechanical ventilation (IMV) and death. The decision to intubate or admit to an ICU facility was left to the attending physician's discretion.
Ct Scan Protocol And Image Acquisition All CT scans were performed within 24 h of admission to the hospital, in supine position. Imaging data were acquired with multidetector-CT (Canon Aquilion Prime 80, and Canon Aquilion RXL 16).

CT scan protocol
After positioning, an unenhanced scan was obtained, followed by IV injection of 100 mL iodinated contrast medium at a rate of 5 mL/seg (Visipaque 320, GE Healthcare, Milwaukee, WI and Optiray 320, Mallinckrodt Medical, St Louis, MO). After bolus triggering at the level of the pulmonary artery with a relative threshold of 150 HU, an early pulmonary arterial angiographic phase was obtained, followed eight seconds later by a delayed pulmonary arterial phase. All the exams were performed with the same acquisition parameters: kV (100 kV), automatic exposure control (Standard), similar range and eld-ofview (FOV: L or LL), collimation (1.0 × 16 for Canon Aquilion RXL 16; 0,5 × 80 for Canon Aquilion Prime sCTA is a technique that uses software-based motion correction between an unenhanced and an enhanced CT scan to obtain an iodine distribution map of the lung parenchyma. Iodine distribution maps of the early and delayed arterial phases were obtained using SureSubtraction software (version 7.0; Canon Medical Systems, Japan). Iodine maps were generated with gray-scale and identical color window tables, ranging from blue (low iodine enhancement) to yellow (high iodine enhancement).

Image analysis
Areas of injured parenchyma in both lungs were assessed for a predominant pattern. These were characterized as ground glass opacities, consolidation and mixed pattern. Airspace compromise was assessed for each of the 5 lobes considering the extent of anatomic involvement, as follows: 0 points for no involvement; 1 point for < 30% involvement; 2 points for 31-60% involvement; and 3 points for > 61% involvement. The resulting CT severity score was the sum of each individual lobar score (range: 0-15). Peripheral vascular dilatation and vascular tortuosity were also assessed in each lobe.
Areas of apparently healthy lung parenchyma were assessed qualitatively for preservation of anteroposterior and apicobasal perfusion gradient or hypoperfusion. Hypoperfusion in apparently normal lung parenchyma was characterized according to its distribution (focal or diffuse), and extension. The latter was graded using a scoring system: rst, both lungs were divided into ve lobes (upper right and left lobes, middle lobe, and lower right and left lobes). Then, the extension of hypoperfusion in apparently healthy lung parenchyma in each lobe was categorized as normal perfusion (0 points), less than 50% of the lobe affected (1 point), and 50% of the lobe or more affected (2 points). This resulted in an overall score that ranged from 0 to 10 points, with higher scores indicating more severe hypoperfusion. Using this score, patients were divided into three groups. Patients with 3 points or less were considered to have mild perfusion abnormalities, those with 4-6 points had moderate abnormalities and those with 7 or more points had severe abnormalities.
Inter-rater reliability was assessed using Cohen's Kappa in a subsidiary sample of 90 subtraction CT angiography scans. Two radiologists, IB with 7 years experience and MS with over 15 years experience reading CT analyzed images of included patients.
The observed agreement was very high, with a median Kappa statistic of 0.95. The domain with least consensus between observers was vascular tortuosity ( =0.9). On the other hand, evidence of pulmonary embolism showed the highest level of agreement between radiologists ( =1.0).
The attending physicians did not have information regarding sCTA perfusion results. However, data regarding overall CT characteristics of included participants was made available to clinicians as per current protocols at study centers.

Sample Size
Using previous data from an early report regarding subtraction CT angiography ndings amongst patients with COVID-19 [4], it was estimated that 190 patients would provide 80% statistical power required to conduct this study. This calculation assumed a HR of 2.0 for intensive care unit admission amongst patients with severe perfusion de cits, an intensive care unit requirement rate of 15% and a twosided p-value of 5%. The targeted sample size was increased by an additional 10% in order to account for artifacts that might render CTA (CT angiography) results uninterpretable.

Analysis Plan
Descriptive statistics including medians, means, standard deviations, interquartile ranges (IQR) and absolute and relative frequencies were used rst to describe participant characteristics. Bivariate comparisons between groups were performed using Fisher's Exact test for categorical variables and oneway Analysis of Variance (ANOVA) or Kruskall-Wallis' test for continuous variables after reviewing data distribution and variances. Association between quantitative variables were sought using Kendall's correlation coe cient. The development of any of the aforementioned study endpoints was evaluated using Kaplan-Meier survival curves that were then compared using the logrank statistic. Hazard ratios (HR) were calculated to quantify the strength of relevant associations alongside 95% con dence intervals.
A Cox proportional hazards regression model was constructed to control for potential confounders. All candidate models considered potential associations between independent variables in their development. The proportional hazards assumption was assessed graphically and by the analysis of Schoenfeld residuals. The overall goodness of t was evaluated using Harrell's C statistic. Analyses were undertaken by an independent statistician who had no participation in the decision to admit or intubate any of the included patients, using STATA 16.1 SE® (College Station, TX: StataCorp LLC). A two-sided p value of < 5% was considered to be statistically signi cant.

Patient Characteristics
Two hundred and ve patients were included in this cohort. However, due to uninterpretable iodine map results, fourteen (6.8% 95% CI 3.7%-11.2%) of these participants had to be excluded from analyses. This left 191 patients in the study. Every participant had RT-PCR con rmed SARS-CoV-2 infection. The mean age was 60.8 ± 16.0 years, 58% were male and the median duration of symptoms was 8 (IQR 5-12) days.
The median SOFA score on admission was 2 (IQR 0-2) and average Pa/Fi ratio on admission was 250 ± 118. Patients with severe perfusion abnormalities showed signi cantly higher SOFA scores and lower Pa/Fi ratios ( Fig. 1) when compared with individuals with mild or moderate anomalies (ANOVA p < 0.001 for both comparisons). A summary of baseline clinical characteristics is shown in Table 1.  Pulmonary CT Evaluation The most common patterns seen in lung CT scans were ground-glass opacities (40.8%), followed by mixed patterns with predominant ground-glass opacities (40.8%) and mixed patterns with predominant consolidation (14.6%). A pure consolidative pattern was rare, with only 3.6% of included patients showing this nding on admission. Patients with severe perfusion abnormalities tended to show a higher frequency of consolidative patterns, but this difference did not reach statistical signi cance (p = 0.06). The median pulmonary CT severity score was 9 (IQR 5-12) points. As described in Table 1, participants with severe perfusion defects showed CT severity scores that were signi cantly higher than those observed in the other groups (p < 0.001).
Signs of pulmonary embolism were found in twenty patients (10.5%), and all of them had an abnormal perfusion CT. Most cases of pulmonary embolism were found amongst individuals with severe perfusion defects and localized at segmental or subsegmental level. Pleural effusion was infrequent amongst included participants (16 patients, 8.3%) and only 2 patients (1.1%) showed signs of right ventricular overload. Both belonged to the severe perfusion abnormalities group.

Perfusion ndings and study outcomes
Perfusion abnormalities were common amongst included participants, with 189 (98.9%) patients having abnormal results. We found 63.8% (122/191) of artifacts in the iodine map images, but these artifacts were not signi cant enough to preclude image interpretation.
Diseased lung areas showed increased blood ow in most patients (85.8%), followed by hypoperfusion (10.0%). A predominant pattern could not be established amongst the remaining participants.
In apparently normal lung parenchyma, the median perfusion-CT score was 9 points (IQR 6-10). Patients were then categorized in three groups using the aforementioned scoring system. Sixteen patients (8.4%) were identi ed as having mild perfusion anomalies, 34 patients (17.8%) had moderate perfusion compromise and 141 (73.8%) had severe abnormalities (Fig. 2). A signi cant correlation was found between the extent of lung parenchymal disease and perfusion abnormalities (Kendall Tau B = 0.50, p < 0.001), which is shown in Fig. 3. Survival analyses showed signi cant differences in the probability of ICU admission and initiation of mechanical ventilation between perfusion score groups, which are shown in Fig. 4. Differences were mostly explained by a signi cant increase in the probability of requiring any of the aforementioned outcomes amongst individuals with severe perfusion abnormalities. The unadjusted hazard ratio for ICU admission was 5.4 (95% CI 2.2-13.5, p < 0.001) and 16.2 (95%CI 2.22-117.5, p = 0.006) for the initiation of mechanical ventilation when contrasted to patients with mild disease. Twenty-ve patients in the severe perfusion anomalies group (17.7%) died during the hospitalization, a rate that was higher than the observed in the other groups as shown in Fig. 5. However, this difference did not reach statistical signi cance (p = 0.44). A summary of study outcomes is shown in Table 2.  Table 3.

Discussion
In this study we evaluated lung perfusion disturbances in areas of apparently healthy lung parenchyma in conventional chest CT images in patients with COVID-19 pneumonia as a severity predictor, through iodine distribution maps obtained with sCTA. The main ndings of this prospective cohort were highly correlated and reliable with our prior small prospective cohort [4] study: (1) perfusion abnormalities were very common amongst included participants; and (2) with increasing severity of hypoperfusion abnormalities, patients had a statistically signi cant increase in their chance to require admission to ICU (p = 0.001), and to require IMV (p < 0.001). The higher rate of mortality was in the severe perfusion anomalies group (17.7%) in relation to the other groups but did not reach statistical signi cance (p = 0.44). This could be explained by the large number of patients with moderate and severe perfusion alterations, and fewer with mild alterations.
ACE2 is an important regulator of the renin-angiotensin system, and SARS-CoV-2 binds to host ACE2 receptors as the functional receptor for invasion into human cells.
Initially, when the virus reaches the air space it produces local in ammation with important local endothelial dysfunction, thrombosis, angiogenesis and vasodilation [10,11,12].
When the virus spreads to the pulmonary blood circulation, SARS-COV-2 binds to ACE2, and due to viral blockade and down-regulation, both Angiotensin I (Ang I) and Angiotensin II (Ang II) accumulate. Also, as angiotensin-converting enzyme (ACE) is not engaged by the virus, the conversion of Ang I to Ang II continues unabated, leading to unopposed accumulation of Ang II [9]. Probably in the early phases, Ang II produces vasoconstriction and endothelial dysfunction with less production of nitric oxide, also resulting in vessel constriction and nally causing hypoperfusion and establishing a progressive V/Q mismatch, with extensive areas of apparently healthy but hypoperfused lung that functions as alveolar dead space [9,13]. This phenomenon could explain in part the L phenotype proposed by Gattinoni et al. at the beginning of the pandemic [14,15], which found severe hypoxemia in patients without signi cant air space compromise. Late and in more advanced stages, excessive Ang II it would end up producing in ammation, vasodilatation, capillary leakage, edema, and a pro-coagulant state, eventually accompanied by microvascular thrombosis [9,15].
A third important nding was that the presence of vascular tortuosity was signi cantly associated with the admission to ICU (p = 0.001) and the requirement of IMV (p = 0.001). Once SARS-COV-2 reaches the alveolus, active replication and release of the virus cause the host cell to undergo pyroptosis, liberating damage-associated molecules, with disruption of the alveolar-capillary barrier, resulting in vascular leakage and alveolar edema. Also, it can locally cause pulmonary endothelitis, thrombosis, angiogenesis and vasodilation resulting in high perfusion to areas of hypoventilated lung and an abnormally low V/Q ratio that can promote hypoxemia [6, 10,12].
We found pulmonary embolism in twenty patients (10.5%), and all of them had an abnormal perfusion CT. Most cases of pulmonary embolism were found amongst individuals with severe perfusion defects localized at segmental or subsegmental level. It has been suggested that SARS-Cov-2 in severe forms of the disease induces an excessive in ammatory state via a cytokine storm combined with endothelial injury and pulmonary vascular microthrombosis, which could considerably increase the risk for venous thromboembolism and mainly pulmonary embolism, for which some meta-analyses have shown a pooled prevalence ranging from about 13-30% [16][17][18][19]. In non-severe COVID-19, these microthrombi are broken down by the highly active brinolytic function in the lungs to allow gas exchange which is noted as an elevation in D-dimers, but in severely ill patients, the pulmonary coagulation system becomes markedly activated, which can manifest clinically as increased oxygen requirements [20,21]. The role of empirical anticoagulation in some COVID-19 subgroups has been considered but is not recommended in general population because it is associated with higher bleeding risk and does not improve the survival rates [22,23].
Another salient nding to highlight from our study is that a signi cant correlation was found between the extent of lung parenchymal disease and perfusion abnormalities (p < 0.001) with lung perfusion scores of 7 or more points (severe perfusion abnormalities) being a signi cant predictor of ICU admission after adjustments for parenchymal disease extension, vascular tortuosity, sex and age were made (p = 0.010).
The limited sample size of our preliminary cohort study was hampered due to its inability to control for relevant clinical confounders. In this second study, we were able to isolate lung perfusion as an important prognostic marker in COVID-19.
Finally, it should be noted that the observed agreement between two radiologists was very high, with a median Kappa statistic of 0.95. Vascular tortuosity was the domain with least consensus between observers ( =0.9), and on the other hand, evidence of pulmonary embolism showed the highest level of agreement between radiologists ( =1.0).
This study has limitations. First, some of the iodine maps were uninterpretable and 63.8% had artifacts that did not preclude their interpretation. Second, sCTA is a new postprocessing technique and thus, has not been as extensively validated as Dual Energy Computed Tomography (DECT). However, it is a promising approach that digitally subtracts a precontrast CT scan from a contrast-enhanced CT scan after motion correction and has higher contrast-to-noise ratio at same dose than DECT. Since it is software-based, it is signi cantly less expensive and potentially more available than costly DECT equipment. Nevertheless, being a novel approach, concerns regarding potential bias resulting from subtraction of non-contrast images from contrast-enhanced images need to be addressed, and greater experience and better knowledge of typical pitfalls is needed to improve the diagnostic accuracy [24,25,26].
Third, perfusion imaging involves more radiation than conventional CT techniques. Finally, it is unclear whether these perfusion abnormalities are unique to COVID-19 or if they can also be found in other multifocal pneumonias and other causes of ARDS.

Conclusion
In conclusion, this study allowed us to correlate perfusion alterations with clinical outcomes, evidencing that patients with severe perfusion anomalies have a higher risk of admission to the ICU and connection to invasive mechanical ventilation. Furthermore, these ndings support the theory that vascular alterations, both in damaged and apparently healthy lungs, would be key to the pathophysiology of the disease. Perfusion alterations could be considered an independent prognostic factor in patients with COVID-19 pneumonia. This study was approved by the Institutional Review Board at the Hospital Naval Almirante Nef. The requirement to obtain informed consent from patients enrolled in the aforementioned study was waived.

Consent for publication
Not applicable Availability of data and materials The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.