Discrepancy between S_pO₂ and S_aO₂ in patients with COVID‐19

In patients admitted to our critical care unit during the COVID‐19 pandemic, we observed that oxygen saturation measured by pulse oximetry (S_pO₂) was consistently lower than arterial oxygen saturation (S_aO₂) measured directly by blood gas analysis.

Over 2 days, S_pO₂ and corresponding S_aO₂ were recorded from patients with severe COVID‐19 (n = 17). The S_pO₂ was measured using a NellcorTM (Medtronic, Watford, UK) reusable sensor. The GEM Premier 5000 gas analyser (Instrumentation Laboratory, Werfen, Germany) was used to directly measure S_aO₂.

Peripheral oxygen saturation underestimated S_aO₂ by > 3% in 15 patients. In nine patients, this gap was > 5%. A Bland‐Altman analysis suggested S_pO₂ consistently under‐read S_aO₂ by an average of 5.3% with 95% limits of agreement. However, our small sample size could be prone to bias, and we did not have a matched control group.

Pulse oximetry is a simple, cheap and non‐invasive method of measuring S_pO₂. The pulse oximeter consists of two light‐emitting diodes which transmit light at two wavelengths; 660 nm and 940 nm, and a photodetector that is sited across a tissue bed, for example, a finger. It is assumed that absorbance at these wavelengths is due to de‐oxyhaemoglobin or oxyhaemoglobin [1]. The accuracy of pulse oximeters is generally quoted as ±2%’ [1]. In the critically ill, S_pO₂ does not reliably predict equivalent changes in S_aO₂ [2]. This is expected as the original calibration is based on calculations made from employing healthy volunteers. Peripheral oxygen saturation can underestimate S_aO₂ in low perfusion states, arrhythmias, vasoconstriction, venous pulsations, oedema and severe anaemia [2, 3, 4]. Nail polish can result in erroneous signal measurement whereas the presence of dyshaemoglobins, or haemoglobin variants can interfere with absorbance [4]. Elevated glycosylated haemoglobin results in an overestimation of S_aO₂ by the S_pO₂ [5]. In sepsis and septic shock, there are conflicting reports on how S_pO₂ is biased [2, 3, 4].

In our patients, we were able to confirm good quality of the pulse oximeter trace and known causes for S_pO₂ underestimation [2, 3, 4] were excluded. An explanation for our observations remains unclear. Suggested hypotheses may include the following; firstly, high ferritin, d‐dimer or other proteins in patients with COVID‐19 may have different spectral properties at 660 nm and 940 nm [6]. These proteins may adversely affect the signal‐to‐noise ratio, thereby reducing the precision of pulse oximetry. Secondly, arteriolar dilatation secondary to tissue hypoxia may lead to venous pulsations, which in turn contributes to falsely low S_pO₂ readings because venous oxyhaemoglobin saturation is also measured in the pulsatile vein [3, 4]. COVID‐19 may contribute through microvascular complications to tissue hypoxia. In addition, there is anecdotal evidence that anaerobic respiration due to secondary infection by anaerobic bacteria in COVID‐19 might inhibit mitochondrial cytochrome oxidase, thereby causing hypoxia at the cellular level (Chakraborty and Das, unpublished observations, https://osf.io/s48fv/). Finally, a possible formation of a complex between the virus and haemoglobin may result in increased red light absorbance relative to infrared absorbance, thereby resulting in a lower S_pO₂.

Our observations in a relatively small number of patients with COVID‐19 pneumonia in critical care suggest that S_pO₂ does not reliably predict S_aO₂, with S_pO₂ consistently underestimating S_aO₂. On our unit, oxygen titration is mostly guided by S_pO₂, and therefore patients may have been administered a higher inspired oxygen fraction than was necessary. It is also possible that the phenomenon of 'happy hypoxia' described in patients with COVID‐19 at an earlier stage in their presentation could be explained in part by these observations.