Non-invasive monitoring of blood gases has become standard procedure in neonatal intensive care units (NICU).1 In particular, continuous monitoring of oxygenation is now considered indispensable to prevent retinopathy of prematurity (ROP) and brain damage that can result from too much or too little oxygen,2^–4 despite randomised trials never having shown continuous monitoring to have an effect on clinically meaningful outcomes.5 6

Two standard techniques are used to monitor oxygenation continuously in the NICU: transcutaneous monitoring of the partial pressure of oxygen (TcPo₂) measures the amount of oxygen dissolved in tissue (which corresponds reasonably well to arterial oxygen tension (Pao₂) when the skin underneath the sensor is heated to 44°C); and pulse oximetry, which measures the proportion of haemoglobin molecules in arterial blood that are loaded with oxygen. TcPo₂ monitoring was introduced in the early 1970s, but was soon replaced by pulse oximetry after that technique became available. Many factors influence the precision and accuracy of TcPo₂ monitoring, including skin thickness, sensor site and temperature, amount of contact gel used and state of peripheral perfusion. It may cause skin burns and requires frequent re-siting and calibration.7 In contrast, pulse oximetry, introduced into the NICU in the 1980, was much easier to do but was prone to motion artefact. There were also considerable differences between brands in their measurement bias and precision, at least until a new generation of more motion-resistant instruments became available about 10 years ago. Ideally both monitoring techniques should be used in combination, particularly in critically ill preterm neonates, although this is costly and time consuming.7 However, the question arises whether it is justified to rely on just one technique for the continuous monitoring of oxygenation in the NICU.

In this issue of the Archives, Quine and Stenson report a randomised crossover study of 19 preterm infants with a mean weight of 1 kg in which they compared saturation versus transcutaneous monitoring. They identified a potentially important advantage of TcPo₂ monitoring, namely that it is associated with less fluctuations in blood or tissue oxygen levels, and, thereby, a higher proportion of time spent with these levels within target range. Displaying either TcPo₂ or Spo₂, they found that Spo₂ monitoring was associated with 10 times more time spent hyperoxaemic and three times more time spent hypoxaemic than TcPo₂ monitoring.8

Shall we now throw away our pulse oximeters and in future solely rely on TcPo₂ monitoring? We do not think so, for the following reasons.

• First, it is possible that much of the reduction in extreme values (either hypoxic or hyperoxic) in the above study8 was due to differences in averaging and/or response times. TcPo₂ monitors respond to changes in Pao₂ only with a delay of approximately 15–30 s.9 This will dampen any variability in their readings considerably if Pao₂ fluctuates, as also acknowledged by the authors. No information is given on the averaging time of the pulse oximeter used in their study, but it most likely was considerably shorter than that of the TcPo₂ monitor.

• Second, we are not informed on how motion resistant their pulse oximeter was. Excluding readings only if pulse and heart rate differ by more than 10 beats per minute (which happened to be the case for only 0.3% of recording time in their study) might not exclude artefacts sufficiently reliably. Thus, it is possible that the actual fluctuations in arterial Po₂ were similar with both monitoring techniques, but resulted in a higher proportion of pulse oximeter readings that were too high or too low because the oximeter responded much faster to these fluctuations, or was more likely to display erroneous readings during motion than the TcPo₂ monitor. It is equally possible, however, that the shorter averaging time of the pulse oximeter resulted in overadjustments in fraction of inspired oxygen (Fio₂) by nursing staff, in response to spontaneous or induced variations in tissue or blood oxygen levels, thereby causing the wider fluctuations in these levels. To separate between these potential explanations would have required to download data on Fio₂, which was not possible in their setting. In any case, to overcome this potential problem would require better staff education, not necessarily different types of monitor.

Perspective on the paper by Quine and Stenson (see p F347)

• Third, although one might expect TcPo₂ monitors to be better suited to detect hyperoxaemia than pulse oximeters, given the shape of the oxygen dissociation curve, this has not been confirmed by previous studies.7 These studies analysed the sensitivity and specificity of TcPo₂ and Spo₂ monitors in detecting hyperoxaemia and hypoxaemia (for a review, see Poets and Southall7). The average sensitivity of TcPo₂ monitors for hyperoxaemia detection was 87% versus 93–97% with pulse oximeters, depending on instrument brand and alarm threshold.7 10 These findings, relying on arterial blood gases as the reference, measured during steady state, are different from those reported by Quine and Stenson.8 The explanation for these differences may lie in the methodological issues raised above.

• Fourth, to ensure optimal agreement between TcPo₂ and Pao₂, transcutaneous monitoring requires frequent blood sampling for arterial blood gases, either from an arterial line or repeated arterial puncture. Neonatologists might consider this too invasive for routine monitoring in daily practice, and survey data from Germany suggest that in only 8% of NICUs arterial blood gases are obtained to validate the transcutaneous measurements.11 However, relying on values based on capillary or venous blood gases might further impair the accuracy of transcutaneous monitoring.

What we need before deciding on the best technique for oxygen monitoring in the NICU are data on how best to prevent ROP and other oxygen-related disorders—that is, whether the absolute level of Pao₂, fluctuations in this level, or both, are relevant to the development of ROP. Currently, the question “which is the best baseline level for Spo₂” is being addressed in large randomised trials in Australia as well as in Canada and Europe. Fluctuations in Pao₂ levels, however, may be equally important, but have yet only been addressed in retrospective studies.12 One way to reduce such fluctuations in Pao₂ would be an automated control of oxygen supply, which is now within reach.13 14 However, a randomised controlled trial is required to show that an increased stability in oxygenation will minimise the incidence of ROP. Until such data become available it might be too early yet to decide on the best method of monitoring stability of oxygenation in preterm infants.