SARS-CoV-2, a virus that manifests clinically as a disease called Covid-19, is causing a serious worldwide pandemic1,2. Although this disease can affect various organs, the main target is the lung, where it can cause both pneumonia with reactive tissue inflammation and also vascular and blood disorders2,3. As a result, many cases present respiratory failure that can become very serious if it develops into acute respiratory distress syndrome (ARDS). The clinical management of these processes requires pulmonary gas exchange monitoring, determined preferably by arterial blood gas tests, although oximetry is often performed to reduce the invasiveness of the procedure. The variable most widely used to assess the severity of hypoxemia and the clinical course is partial oxygen pressure (PaO2) and its ratio to the concentration of gas delivered (fraction of inspired oxygen [FiO2]), a variable known as PaFi (PaO2/FiO2)4. PaO2 is important for determining the total oxygen content delivered to the tissues from arterial blood (CaO2), as it is an indicator of both the amount of gas bound to hemoglobin and the amount dissolved in plasma5. The first is far more important (95% of CaO2), and is expressed by percentage oxygen saturation (SaO2). However, a recent in silico study reported that some SARS-CoV-2 proteins (e.g., ORF8)6,7 might inhibit the binding of hemoglobin with oxygen and even destroy hemoglobin molecules6,7. An even more recent study, conducted on red blood cells from Covid-19 patients using proteomic and lipidomic determinations, also suggested that the affinity between oxygen and hemoglobin in these patients may be reduced8. The clinical consequences of these 2 studies could be significant, as they would limit the usefulness of PaO2 and PaFi in assessing the real oxygenation status of Covid-19 patients7. Since numerous validated equations can be used to calculate the oxygen saturation that corresponds to a particular partial oxygen pressure while taking into account the other main factors involved, the difference between the estimated oxygen saturation and the patient’s real saturation status will help determine whether the affinity of oxygen for the hemoglobin molecule is affected by the virus. The objective of this study was to analyze the concordance between measured SaO2 and SaO2 calculated from PaO2, in order to assess whether the infection actually does impact the affinity of oxygen for hemoglobin.

This study was performed by analyzing all arterial and venous blood gas samples obtained from Covid-19 patients (confirmation by polymerase chain reaction [PCR] in nasal or pharyngeal smears) received by our center’s central laboratory between March and April 2020. A total of 549 arterial blood samples were collected from 139 patients, and 544 venous blood samples were collected from 215 patients. Other blood gas determinations that were collected and processed in other healthcare facilities with slightly different instrument calibration and maintenance criteria were excluded. We also studied blood gas determinations from 2 control groups: the first comprising patients tested synchronously during the same period as the experimental group who presented febrile symptoms but radiology not suggestive of SARS-CoV-2 and PCR negative for this virus (251 arterial samples from 73 patients and 1069 venous samples from 379 patients), and the second comprising subjects admitted for non-parenchymal lung disease in the 2 months prior to the detection of the first cases of SARS-CoV-2 infection (2614 arterial samples from 1089 patients and 3098 venous samples from 1461 patients). We included arterial and venous samples because we were interested in analyzing oxygen affinity for hemoglobin over a wide range of arterial oxygen pressures. In line with the requirements of the institution’s clinical research ethics committee, patients initially gave verbal consent for the inclusion of their clinical and analytical data, an approach that was subsequently approved within the framework of a study on patients with severe Covid-19 (ref. 2020/9458).

Table 1 shows the most relevant general and clinical characteristics of patients, while Table 2 shows mean arterial and venous blood gases for each of the 3 groups. In the Covid-19 group, SaO2 measured by cooximetry was almost identical to the results calculated using the different equations (Table 3). The results were similar in venous blood. The correlation between measured and calculated SaO2 and between measured and calculated SvO2 was also very high, with the regression line very close to the identity line (Fig. 1 of the manuscript and figure A(a) of the online supplement Fig. 1); these results were also corroborated by the consistency analysis, in which ICC values were always higher than 0.98 (Table 3). This is also shown in the Bland-Altman plot (Fig. 2). The patients showed mean partial pressure of carbon dioxide (PCO2) and pH within normal ranges in both arterial and venous blood. There were no relationships of interest between clinical variables and the differences between measured and calculated saturations.

Fig. 1.

Relationships between saturation measured by cooximetry and saturation calculated using the equations of (a) Kelman, (b) Severinghaus, and (c) Siggaard-Andersen for arterial blood, and for venous blood (d), (e), and (f), respectively.

(0.36MB).

Fig. 2.

Bland-Altman plots for the differences between measured saturation and saturation calculated by the three equations used in the study, for both arterial blood (a), (b), and (c), and systemic venous blood (d), (e) and (f). The solid line represents the mean between the measured and calculated values. Dashed lines represent the mean ± 1.96 x standard deviation (95% confidence interval).

(0.4MB).

The results did not differ substantially in either control group, nor did the synchronous group of patients with negative PCR for SARS-CoV-2 show differences between measured saturation and saturation calculated with the different equations in either arterial or venous blood. The same was true of the group of patients without infectious disease who had gas analysis performed in the months before the pandemic. In these 2 cases, correlations and ICC were also highly significant. Hemoglobin and ferritin were somewhat higher in the Covid-19 group compared with both control groups.

In the post-hoc analysis, which excluded measurements with PaO2 values greater than 100 mmHg, no differences were observed between measured SaO2 and SaO2 values calculated using any of the 3 equations and for any of the 3 study groups (Appendix B Tables A and B, and Figure A of the online supplement).

The main finding of this study is that active SARS-CoV-2 infection does not substantially change the PaO2:SaO2 relationship. This is of particular importance in the clinical management of patients with pulmonary involvement due to Covid-19. Furthermore, there was no impairment of oxygen affinity for hemoglobin in venous blood, nor was this molecule reduced by infection.

A recent in silico study suggested that various viral surface proteins (such as ORF8 and perhaps some other glycoprotein) might inhibit binding between the iron of the hemoglobin molecule and oxygen6,14, thus impeding the transport of oxygen to tissue, despite adequate PaO2 being maintained. This would also mean that PaO2 is only of relative use in the management of patients, since it does not reflect their real oxygenation status7,14. More recently, another group using omic techniques described protein and lipid alterations that affect the membrane of the red blood cell, a situation that could also reduce the affinity of oxygen for hemoglobin at a certain partial pressure of oxygen8. Under normal circumstances, this pressure helps predict hemoglobin saturation with acceptable accuracy, since both variables maintain a known relationship, similar to an italic S. At high or very low oxygen pressures, saturation changes secondary to PaO2 modifications are relatively small; however, when the oxygen pressure falls below 60 mmHg, saturation declines dramatically5,15. Of course, certain factors can modify the relationship between both variables, shifting the curve to the left or right. The most important of these are temperature, PaCO2, pH and concentration of 2–3-diphosphoglycerate5, as well as the presence of abnormal hemoglobins, as would be the case in SARS-CoV-2 infection. In clinical circumstances, these shifts are not usually taken into account, but they are used to calculate saturation when it cannot be determined directly. This is no longer common because most current gas analyzers include cooximetry. However, as this technique was not widely available until relatively recently, various equations for predicting saturation were developed; of these, Kelman, Severinghaus, and Siggaard-Andersen offer the best concordance with the measured saturation10–12,16. In our study, these equations predicted the saturation that would correspond to a certain oxygen pressure, taking into account the other factors mentioned, in both arterial and venous blood, in a large sample of patients and blood samples. Our results indicate that saturation calculated using any of the foregoing equations shows no clinically significant difference with saturation measured by cooximetry, strongly undermining in real-world practice the hypothesis that SARS-CoV-2 infection significantly alters the affinity of oxygen for hemoglobin. Very recently, a British group also suggested that this hypothesis could be misleading after a small pilot study of 14 arterial blood gas samples from patients with Covid-1917. Although they used a different methodology based on the calculation of P50, their results are consistent with ours. However, the real situation regarding oxygen affinity for hemoglobin in these patients is still under discussion18,19. Our results also remind us of the enormous gap between in silico or in vitro estimates and real-world practice.

The above-mentioned in silico study suggested that other viral proteins, such as orf1ab, ORF3a and ORF10, might denature the structure of the hemoglobin molecule, releasing iron and transforming the preexisting hemo groups to simple porphyrins that lack the capacity to carry oxygen6. This would affect CaO2 levels, which depend predominantly on hemoglobin-bound oxygen, decreasing supply to the tissues5. Our results do not support this hypothesis, since we found that blood hemoglobin values in patients with Covid-19 were even slightly higher than those of the control groups. However, some studies have reported lower hemoglobin levels in these patients20–22. In general, these studies have not included control groups of patients with severe disease, in whom blood collection and reduced hemoglobin are very common. However, like us, other authors have reported increases in serum ferritin in patients with Covid-1923, possibly related to this molecule’s role in buffering iron released by hemoglobin22. Iron is essential for viral replication, and this might explain why several viruses can interfere in iron homeostasis24. However, it is not yet known whether this could also be the case with SARS-CoV-2. Macrophages also play an important role in the regulation of iron homeostasis, as they prevent the release of hemoglobin molecules by red blood cells25. However, an excess of hemo groups can also help activate the innate immune system, triggering proinflammatory macrophage orientation26,27, which appears to be a key trigger of the Covid-19 cytokine storm28.

Funded by an unrestricted grant from Boehringer-Ingelheim and the Catalan Foundation for Respiratory Medicine (FUCAP).