To the Editor

We read with interest the recent manuscript by Lin et al. ^{Reference Lin, Mills, Jergel, Aljiffry and Oster1} examining pulse oximetry bias in neonates less than or equal to 7 days of age. We commend the authors for assembling a very large cohort and for contributing to our knowledge of the limitations of pulse oximetry in the neonatal population.

The work included 1645 infants hospitalised in a single institution’s NICU and CICU over a 12-year period. The mean saturation bias was 8.7 percentage points, a value larger than has previously been reported in the literature. ^{Reference Foglia, Whyte and Chaudhary2,Reference Andrist, Nuppnau, Barbaro, Valley and Sjoding3} It could be argued, prima facie, that a difference of this magnitude would substantially limit the clinical reliability of pulse oximetry in this population. For historical reference, the largest difference reported previously in paediatrics was 6.6%. ^{Reference Ross, Newth and Khemani4} The rate of occult hypoxaemia (SpO₂ > 95% and SaO₂ < 90%), despite a more conservative definition than that devised by Sjoding et al. ^{Reference Sjoding, Dickson, Iwashyna, Gay and Valley5} (Spo₂ > 92% and SaO2 < 88%), is also notably higher (20.5%) than previously reported values (allowing for multiple measures from each patient, Andrist et al found that 16.5% pediatric patients experienced occult hypoxemia). ^{Reference Andrist, Nuppnau, Barbaro, Valley and Sjoding3,Reference Ruppel, Makeneni and Faerber6,Reference Savorgnan, Hassan, Borges and Acosta7}

We re-analysed the data behind our work ^{Reference Coghill, Law, Webb, Asfari and Borasino8} with a subset similar to the cohort presented by Lin et al. We found that the mean bias for blood gas pairs in patients less than or equal to 7 days of age was 4.74 percentage points (2.95 percentage points when patients were limited to a single data pair post-bypass surgery). The incidence of occult hypoxaemia (SpO₂ ≥ 95%, SaO₂ < 90%) was 6.03%.

There are several methodological differences in how we acquired our data compared with how Lin et al. acquired theirs. We used high-fidelity data storage recorded by Etiometry (Dedham, MA) with time-averaged pulse oximetry values to reduce noise. Lin et al. matched discrete data pairs within a 10-minute period. The greater proportion of cyanotic patients in our cohort would be expected to increase the mean bias based on the underlying data (increase the difference at lower SaO₂).

We suspect that another major contributor to the observed bias may be related to technical aspects of data acquisition. The use of the i-STAT device (Abbott), as mentioned in the manuscript, relies upon the derivation of SaO2 using pH, pO₂, pCO₂, and a proprietary algorithm based on a standard adult haemoglobin dissociation curve. ⁹ This approach has important limitations in neonates, who have substantial fractions of fetal haemoglobin. ^{Reference Murthy, Hicks and Soldin10,Reference Porath, Sinha, Dudenhausen and Luttkus11} Fetal haemoglobin exhibits a “left-shifted” dissociation curve that would result in higher SaO₂ for any given PO₂. This would suggest that the derived SaO₂ values may underestimate true arterial oxygen saturation, thereby artificially amplifying the observed difference between pulse oximetry and arterial measurements and inflating estimates of occult hypoxaemia.

We again commend the authors on establishing a large cohort of vulnerable patients. We agree with the conclusion that pulse oximetry significantly overestimates arterial oxygen saturation. The development of pulse oximeters that are more specifically calibrated for this population is necessary, and prospective studies using objective measures of skin pigmentation and directly measuring oximetry data would benefit this population.

Sincerely,

Matthew Coghill, MD