Causes of Hypoxemia in COVID-19 Acute Respiratory Distress Syndrome: A Combined Multiple Inert Gas Elimination Technique and Dual-energy Computed Tomography Study

The data reporting follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cohort observational studies. The study was approved by the hospital ethical board, registered (Deutsches Register Klinischer Studien DRKS00025503), and conducted in the intensive care unit of the University Medical Center of Göttingen (Germany). The only inclusion criteria was the diagnosis of ARDS⁴  from COVID-19. The diagnosis of COVID-19 infection was confirmed with an RT-PCR test performed on a nasal swab. The measurements were performed within 7 days from the endotracheal intubation. Exclusion criteria were pregnancy, hepatic cirrhosis, extracorporeal membrane oxygenation, and patent foramen ovale. Patent foramen ovale was excluded with an agitated saline bubbles bolus during transthoracic echocardiography. The recruitment of patients ran from November 2021 to March 2022. The written informed consent was acquired from a patient’s legal representative.

A femoral or brachial artery was cannulated with a 4F PiCCO catheter (Getinge, Sweden). We recorded respiratory mechanics and hemodynamic data after PiCCO calibration. The cardiac output was measured by transpulmonary thermodilution. Samples for the MIGET were taken in duplicate from mixed expired gas and arterial blood, as explained below, together with an arterial and central venous blood gas analysis. After the measurement, the patient was transported to the computed tomography scan facility within 5 h. For technical and safety reasons, it was not possible to perform the whole measurement in the computed tomography scanner.

Six inert gases, namely sulfur hexafluoride-SF6, ethane, and cyclopropane (Air Liquide, Germany); isoflurane (Baxter, USA); diethyl ether and acetone (Merck, Germany), were dissolved in 500 ml of 0.9% NaCl (Baxter), before the measurement.^5,6  The infusion rate was set at 1/1,000 of the patient’s minute ventilation. During the infusion, the ventilator circuit was modified, substituting the entire expiratory limb with a specifically designed 1-l, baffled, flow-through mixing box, heated above body temperature from the mouth opening to the gas sampling point.⁷  After 30 min, an arterial blood sample and a time-aligned sample of the mixed expired gas were collected into ungreased, glass gas-tight syringes (Cadence Science, USA). The subsequent preparation of the samples (adapted to our own setup) is detailed in the supplemental digital content (https://links.lww.com/ALN/D303). The prepared probes were sent to the gas chromatography facility, and the data obtained were entered into the Fortran software kindly provided by Peter D. Wagner, M.D. (University of California, San Diego). Finally, the $V˙A/Q˙T$ distributions were recovered.

The PiCCO was once more calibrated. DECT was carried out with the same ventilatory setup the patient had during MIGET. DECT was performed with a dual-source spiral unit (Siemens Somatom Force, Siemens Healthcare, Germany) at two different energy levels, after contrast injection.⁸  The threshold for the bolus trigger was 120 HU, the injection rate was 3 ml/s, 5 seconds of delay before scanning. The scan was performed with bolus-tracking. From the study, we extracted two Digital Imaging and COmmunications in Medicine (DICOM) series through a three material mass fraction decomposition (syngo.via, Siemens Healthcare): a virtual unenhanced and a perfusion map. Each slice was manually contoured along the lungs, excluding the great vessels and bronchi. On the virtual unenhanced series, we performed a quantitative analysis as previously described.⁹  The analysis carried out on the perfusion map is detailed in the supplemental digital content (https://links.lww.com/ALN/D303).¹⁰  From -1000 HU to +100 HU, for each interval of 100 HU, we calculated fractions of the total for tissue and the blood volumes and the ratio between them. Summing these ratios, in those regions where this ratio was less than 1, led to the calculation of the overperfusion coefficient. Finally, a radiologist consultant (L.B.) diagnosed the presence of pulmonary embolism.

A formal sample size has not been calculated a priori. We chose to enroll patients during what we hypothesized to be a “winter wave” of COVID-19, namely between November 2021 and March 2022. Given the stringent entry criteria, and the nature of our institution, we aimed to recruit between 10 to 15 patients.

The normal distribution of the variables was assessed with the Shapiro–Wilk test. The data are presented as means ± SD or medians [interquartile range] as appropriate. The quantitative relationship between variables was assessed with linear regression with the ordinary least-squares method. Two-tailed P values < 0.05 were considered statistically significant. The data analysis was performed in Julia 1.8¹¹  and the following packages: DataFrames.jl¹² , Makie.jl¹³ , GLM.jl¹⁴ , and Distributions.jl¹⁵ .

Between November 2021 and March 2022, 12 patients were eligible for inclusion. Two patients were excluded because of relevant patent foramen ovale (n = 1; table E1 supplemental digital content, https://links.lww.com/ALN/D303) and technical problems during the measurement (n = 1). A demographic characterization of the resulting sample (n = 10) is presented in table 1 and table E2 (supplemental digital content, https://links.lww.com/ALN/D303). Nine patients were infected with B.1.617.2 (Delta variant of COVID-19), one was infected with B.1.1.529 (Omicron variant), and only two were partially vaccinated (single dose). The laboratory analysis showed d-dimer elevation as commonly observed in COVID-19 (table E3, supplemental digital content, https://links.lww.com/ALN/D303). Table 1 also displays common respiratory and invasive hemodynamic variables measurable at the bedside.

The $V˙A/Q˙T$ distribution of each patient is represented in figure 1, while in figure E1 (supplemental digital content, https://links.lww.com/ALN/D303) are presented the plots of the retention–excretion of the inert gases. As shown, the patients presented with overall heterogeneous patterns of $V˙A/Q˙T$ alterations. The residual sum of square of the gas data fitting procedure was 4.7 [3.5 to 6.9], ensuring that MIGET was overall well performed. More than 50% of the cardiac output was distributed in regions with a $V˙A/Q˙T$ ratio between 0.1 and 1, while ventilation was more widely distributed among ratios 0.1 to 10. Both the scattering of perfusion (logarithm of the SD, LogSD, Q) and the one of ventilation (LogSD, V) were abnormal, indicating a relevant presence of $V˙A/Q˙T$ mismatch. LogSD, Q and LogSD, V are the second moment of perfusion and ventilation distributions, respectively. They represent the dispersion on both sides of the mean of perfusion (and ventilation) and they are therefore used as a measure to describe the heterogeneity of the distributions. Dead space was 53 ± 11%, and true shunt was 25 ± 16%. Table 2 summarizes these findings.

The recovered $V˙A/Q˙T$ distribution was used to calculate the predicted Pao₂ of the patient, as previously described,¹⁶  and to estimate the degree of diffusion limitation (and of postpulmonary shunting, if present). In figure 2, the plot between the predicted Pao₂ and the measured Pao₂ shows that there was a tendency toward overestimation, suggesting the presence of diffusion limitation or postpulmonary shunting.

Dual-energy computed tomography was used to quantify the anatomical distribution of the tissue and lung volume. The overall lung weight was 1,427 ± 357 g,⁹  and the excess tissue mass was 541.1 ± 248.4 g.¹⁷  The blood volume maps showed the presence of defects of perfusion in all patients. Pulmonary embolism was present in five of ten patients on DECT. The distribution of the fractions of tissue mass and blood volume is displayed in figure 3A. A large fraction of the tissue volume was distributed in the range between –100 and 0 HU and above 0 HU, as expected for ARDS lungs. The blood volume, instead, was distributed across all the density compartments, with a larger fraction in the well aerated and overinflated tissue. In the overinflated and the well aerated lung compartments, we detected a relevant blood pooling (i.e., those areas were possibly overperfused). In contrast, the denser the tissue, the less the blood volume proportionally entering those areas. There was a remarkable correlation between the fraction of atelectasis (nonaerated lung tissue) and the blood volume distributed to those areas (R² = 0.90, P < 0.001; fig. E2, supplemental digital content, https://links.lww.com/ALN/D303). The shunt measured by MIGET was directly proportional to the fraction of atelectasis (R² = 0.56, P = 0.012; fig. E3a, supplemental digital content, https://links.lww.com/ALN/D303) and to the blood volume distributed to these regions (R² = 0.70, P = 0.003; fig. E3b, supplemental digital content, https://links.lww.com/ALN/D303). $V˙A/Q˙T$ mismatch was correlated with the fraction of blood volume distributed to the poorly aerated tissue (R² = 0.54, P = 0.016; fig. E4, supplemental digital content, https://links.lww.com/ALN/D303), while diffusion limitation or postpulmonary shunt was not correlated with any of the measurements we acquired. Importantly, $V˙A/Q˙T$ mismatch and diffusion limitation were not related to shunt (P = 0.376 and P = 0.154, respectively) or to each other (P = 0.813).

In figure 2 (B and D), we explored the relationship between the overperfusion coefficient as an arbitrary index of the degree of blood pooling with respect to the lung tissue and measurements of injury severity. Increasing overperfusion coefficient meant worse oxygenation, higher excess tissue mass, and worse gas exchange efficiency.¹⁸  Of note, the coefficient was also larger in patients with pulmonary embolism (fig. 2E).

This study describes the complex pathogenesis of hypoxemia in COVID-19 respiratory failure. A few aspects are worth mentioning.

The past years have seen many theories speculating about the mechanisms of hypoxemia in COVID-19 patients.¹⁹  In this study, we showed, using a state-of-the-art technique, that the main cause of hypoxemia in COVID-19 respiratory failure is shunt. This is not a new finding in ARDS; indeed, a milestone article from Dantzker et al.¹  and other works later^20,21  showed that shunt is the main cause of hypoxemia in ARDS from various etiologies. Despite a rather different setup (higher positive end-expiratory pressure, lower Fio₂, and tidal volumes), we confirm this finding even in COVID-19–related ARDS.

In our sample, we found that ventilation and perfusion were mismatched, mainly because more than 50% of the cardiac output was distributed in regions with a $V˙A/Q˙T$ less than 1 and greater than 0.1. Lower $V˙A/Q˙T$ regions were substantially absent. While low $V˙A/Q˙T$ can be also found in ARDS from different causes,²¹  the cardiac output distributed there is usually less than 20%. The fact that $V˙A/Q˙T$ mismatch could be a more relevant cause of hypoxemia in COVID-19 has been hypothesized by our group at the beginning of the pandemic.^3,22  Later reports showed a high incidence of thromboembolic events,^23–25  known to increase the $V˙A/Q˙T$ dispersion.²⁶  Our findings may indirectly confirm that the macro- and microvascular thromboembolisms typical of this disease may have a substantial effect on gas exchange.

A number of studies in ARDS from different etiologies firmly exclude the diffusion limitation as a mechanism for hypoxemia.^1,21  By providing a complete description of the $V˙A/Q˙T$ ratios in the lung, the MIGET allows estimating the patient’s Pao₂ (predicted Pao₂) if that specific $V˙A/Q˙T$ distribution was solely responsible for the gas exchange. However, the actual Pao₂ (measured Pao₂) is also sampled directly at the bedside. The difference between the predicted and the measured Pao₂ can be considered a proxy for the magnitude of diffusion limitation. A caveat, however, is represented by the postpulmonary shunts. With this term, we refer to the shunts that occur after the blood has already transited across the capillary bed (mainly through bronchial and thebesian veins). After transiting through the lung, the inert gases used by MIGET do not undergo any further change in concentration. On the other hand, the blood returning from the lungs will subsequently perfuse the bronchial circulation, providing an aliquot of deoxygenated blood in the left atrium, uncoupling the exchange of oxygen from that of the inert gases. However, it has been shown that the postpulmonary shunting is of minor importance under most physiologic and pathologic conditions.^16,27 

Quite unexpectedly, we found some evidence of diffusion limitation or postpulmonary shunts in our patients. While diffusion limitation is most commonly caused by lung fibrosis,²⁸  it is hard to imagine that the required anatomical changes would take place so rapidly in the course of the disease (less than 7 days from intubation). Moreover, our patients were ventilated with a high Fio₂. Therefore, the postpulmonary shunting seems, in this context, more likely. Interestingly, it is known that in chronic thromboembolic disease, for example, the bronchial arteries can dilate and carry as much as 30% of the cardiac output, as a consequence of microvascular occlusions.^29,30  Recent anatomical studies have shown the opening of intrapulmonary bronchopulmonary anastomosis in COVID-19.³¹  Shunting through the bronchial system could therefore explain our findings and also the positive contrast-enhanced transcranial Doppler³²  and orthodeoxia.³³  This, however, needs to be taken with care, as experimental error at multiple levels during the MIGET procedure may lead to a falsely increased difference between the measured and predicted Pao₂.

With DECT, we found increased lung excess tissue mass, a sign of edematous lungs, confirmed by the increased Extravascular Lung Water index. We quantitatively measured the distribution of the tissue mass and, importantly, of the blood volume. Although DECT is frequently used in various contexts, it has only recently gained popularity as an imaging technique for ARDS.³⁴  At the time of writing, its quantitative analysis has been performed only by a few centers.^35,36  The techniques used to evaluate the distribution of tissue and blood volume, despite small differences, are all variations of the original voxel quantitative analysis of a “traditional” CT scan.⁹ 

Regarding the distribution of the blood volume, it is key to understand whether hypoxic pulmonary vasoconstriction is preserved or lost.^34–36  In both healthy and diseased lungs, hypoxic pulmonary vasoconstriction diverges the blood flow away from the gasless tissue, helping maintain the matching between ventilation and perfusion.³⁷ 

The tissue mass followed the expected distribution of ARDS lungs, with proportionally more tissue in the nonaerated regions. The blood volume followed in each interval a similar distribution, but with a tendency toward a relative hypoperfusion of the atelectasis and a relative hyperperfusion of the well aerated tissue. Importantly, the blood flowing through the atelectasis was the main determinant of shunt as measured by MIGET, confirming few other studies during both general anesthesia and ARDS.^38,39  This is relevant because it opens the way for the use of the quantitative analysis of DECT as a functional tool for the evaluation of both anatomy and gas exchange. Interestingly, we also found an association between the poorly aerated tissue (ground glass opacities) and the $V˙A/Q˙T$ mismatch, which implies that in earlier stages of the disease, when ground glass opacities are the main radiologic finding, $V˙A/Q˙T$ mismatch may play a more relevant role.

The redistribution of the blood flow away from the atelectasis, leading to hyperperfusion of the well aerated lung, is challenging to interpret. On one hand, this might represent a sign that hypoxic pulmonary vasoconstriction is working efficiently. On the other hand, when we calculated the hyperperfusion coefficient, we found an impressive association with several indices of ARDS severity, namely Pao₂/Fio₂, excess tissue mass, and Etco₂/Paco₂. Moreover, the coefficient was larger in patients with pulmonary embolism. This seems to support the hypothesis that the microvascular alterations⁴⁰  typical of COVID-19 may force the blood to flow through the well aerated tissue, and this redistribution might be seen as a powerful marker of severity.

Despite combining the complex methods of MIGET and quantitative analysis of DECT, as done here, is not likely applicable on a broader scale for clinical decision-making, these tools offer an unprecedented understanding of the mechanisms underlying this disease and ARDS in general.

However, it is important to highlight some limitations of the techniques we used. The relevant finding regarding diffusion limitation or post-pulmonary shunting is based on an adjunctive calculation on the MIGET data. Importantly, this analysis works well when the Pao₂ is low. In our case, however, some patients had a Pao₂ close to the atmospheric one. This may lead to a rapid equilibration between air and the syringe, decreasing the measured Pao₂. Therefore, we highlight once more the need to interpret this finding with care.

Throughout the paper, when commenting on the DECT findings, we tried to avoid writing about “perfusion”, preferring the wording “blood volume” instead. Indeed, DECT simply detects iodine and, through the calculation of the three material decomposition, separates a iodine map from a tissue map. Given the short duration of the acquisition, we assume iodine remains intravascular so that the distribution of iodine closely mimics the distribution of blood. Importantly, this is a volume, not a flow. While studies have found a good association between flow and volume⁸ , this might not be true when pulmonary capillary recruitability is compromised, which may be the case of our patients. Therefore, more than overperfusion, blood pooling could be likely a more appropriate wording.

The study has another obvious limitation: the low number of patients studied. However, the study also benefits from a rigorous, and until now unattempted, experimental measurement in this highly challenging context. In absence of a typical ARDS group it is difficult to generalize our findings. Nevertheless, we believe these results highlight how hypoxemia in COVID-19 is exceptionally multifactorial, combining alveolar factors, typical of ARDS, with vascular factors, more typical of macro and microembolism. Moreover, our analysis revealed a very solid correlation between some DECT-derived data and the ones derived from the much more complex MIGET. This, in our opinion, opens the door to the use of DECT in ARDS, as an interesting clinical and experimental tool.

The authors thank P.D. Wagner, M.D. (University of California, San Diego, California) for his kindness and the invaluable technical and conceptual assistance in developing MIGET at our institution. Thanks also to Alessio Gasperetti, M.D. (Johns Hopkins University, Baltimore, Maryland) for the uncompromised support during the early phases of the project. The authors are also thankful to Peter Herrmann, M.Sc. (University Medical Center, Göttingen, Germany) for his continuous assistance with the computed tomography scan data. They thank all the intensive care unit staff at the University Medical Center, Göttingen, Germany, the study nurses Loredana Benea and Andrea Kernchen, as well as the nuclear medicine (in particular, Dr. Daniel Modemann, M.Sc., for helping develop the methodology at the gas chromatography–mass spectrometry) and radiology facilities for the collaboration and compliance while performing this complex study.

Supported by departmental sources and an unrestricted grant for respiratory medicine by Sartorius AG (Göttingen, Germany).

Dr. Biggemann received a travel grant from Bayer AG (Leverkusen, Germany) and speaker fees from Bristol–Myers Squibb (New York, New York). Dr. Harnisch reports consultancy for Shionogi (Osaka, Japan). Dr. Vogt reports consultancy for MSD Merck (Rahway, New Jersey). Dr. Saager reports consultancy for the Surgical Company (Amersfoort, The Netherlands). Dr. Lotz is a board member of Röntgenfortbildung Neuss GmbH (Grevenbroich, Germany). Dr. Gattinoni reports a consultancy for SIDAM (Mirandola, Italy); he also receives lecture fees from Estor (Pero, Italy). Dr. Moerer is a member of the COVID-19-Evidenz-Oekosystem (CEOsys) Network (Germany) funded by the Federal Ministry of Education and Research; he is also a member of the national Preparedness and Pandemic Response in Deutschland (PREPARED) and Nationales Pandemie Kohorten Netz - Therapeutische Interventionsplattform (NAPKON-TIP) networks, which is funded by the Federal Ministry of Education and Research; Dr. Moerer also received honoraria for sponsored lectures by Getinge (Sweden) and Advitos GmbH (Münich, Germany) and speaker fees for workshops at the European Medical School (Oldenburg, Germany). The other authors declare no competing interests.

Supplemental methods and results, https://links.lww.com/ALN/D303