A long long time ago, in a galaxy… actually quite near here, the idea of servo control of inspired oxygen was already in the air. At the time I first heard about it, the idea was to control FiO2 based on PO2 derived from electrodes on the tip of an umbilical catheter (paper from the mists of time… 1979), so you can see how old an idea it is! Other systems were developed which used transcutaneous PO2, and although the systems worked, there were the obvious limitations of using TcPO2. Now multiple systems have been tested to control FiO2 based on the pulse oximeter, different algorithms compared, systems that work with high-flow nasal cannulae developed, and even use in the delivery room during resuscitation had been trialled.

What will it take for this to become standard of care?

Most babies receiving oxygen have adjustments to their oxygen supply several times a day, this is usually triggered by alarms from the pulse oximeter, which have an enormously high frequency, one recent study, for example, noted up to more than 300 alarms per day, which leads to alarm fatigue and even important alarms being ignored or with very delayed responses. Babies as a result may spend substantial amounts of time below and above target ranges. In our unit, as in many others, we have a target range of saturation for most babies receiving oxygen between 90 and 94%, but we set the alarms 2% wider than that in order to avoid too many alarms. In addition, our oximeter monitors have an adaptive alarm system, so if they are just 1% outside the alarm limit they don’t ring for 60 seconds, at 2% they ring after 30 seconds, and so on, until they ring immediately if more than 5% outside the limits. Even with these approaches, the alarms are very annoying to parents, nurses, and I dare say babies (and to me!).

What we really need are systems which reduces the number of episodes of hypoxia and hyperoxia, and reduces the duration and severity of over and undersaturation when they occur. I am not sure if the most important thing is to reduce the total duration of hypoxia and hyperoxia, or to reduce the number of episodes of hypoxia and hyperoxia. It may be that saturations rapidly changing up and down are worse than persistently low (or high) saturations, even if the total duration is the same. Re-saturation after hypoxia is a potent source of oxygen free radicals, so my guess is, that for the same total duration of hypoxia, it would be less harmful to have one long episode of mild hypoxia, rather than 100 brief episodes with intermittent re-saturation.

Hyperoxia is bad for your retina and your lungs, and probably your brain also, but is a persistent slightly high saturation worse than a saturation that goes up and down to 99% then back to normal? There might be animal data out there that address the issue, but I don’t know them.

There are a number of complicating factors, in designing automated control systems. Much hypoxia is caused by apnoea, which won’t be affected by giving more oxygen! Post apnoeic hyperoxia is already a real problem, which probably often occurs when caregivers increase oxygen during apnoeic spells (partly because mixed and obstructive apnoeas don’t look like apnoea, so people try increasing the oxygen anyway). An automated system which routinely increased oxygen administration during apnoeas could be problematic. But, with the appropriate safeguards built into the algorithms it should be possible to safely reduce overall oxygen instability, and reduce alarms.

Another fairly trivial problem for an automated system, would be to turn off the high saturation alarm when the baby is in 21% oxygen, and turn it back on again when the oxygen concentration is increased; this is by no means a trivial problem in the NICU currently, babies in variable very low concentrations of oxygen who have a period when they saturate above 96% in 21% oxygen, will currently, appropriately, have the high alarm switched off by caregivers, and then when the baby needs an increase in FiO2 for a temporary desaturation, they often end up in low supplemental oxygen concentrations, with saturations above the desired limits and the high alarms turned off.

Two years ago Carlo Dani reviewed the then available literature, finding 16 trials which compared oxygenation outcomes between standard and automated oxygen control. He showed that overall the proportion of time spent in normoxia was increased, with less time spent hyperoxic, and a smaller impact on hypoxia, with automated control. Since then there have been several other trials published, including one by his own group, but we still don’t know if any clinical outcomes might be improved.

One of the outcomes that always seems improved in the trials, when reported, is the number of manual oxygen adjustments that were required, which is usually dramatically reduced. Alarm frequency is not something that I have seen reported much, but is presumably substantially reduced. One interesting study (Warakomska M, et al. Evaluation of two SpO2 alarm strategies during automated FiO2 control in the NICU: a randomized crossover study. BMC Pediatr. 2019;19(1):142) randomly compared 2 alarm strategies in a cross-over study, among babies who are all on automated oxygen control, they crossed-over babies who had normally a target range of 88-95% saturation, and had either alarm limits set to 87 and 96%, with a 30 second delay, or looser alarms, set 2% wider with a 90 second delay. They only studied 21 babies and started on day 1 of life for up to 6 days, and the alarm strategy crossed over every 24 hours, they used the Avea ventilator for both invasive and non-invasive support and stopped the study when they were off CPAP. Although the alarm limits were wider and slower, and therefore rang much less frequently in the “loose” group, oxygen saturation profiles were identical, being 95% of the time normoxic (88-96% unless in room air, when >96% still considered normoxic). The number of actual alarms is shown below.

.

The holy grail, of course, to use an overused metaphor, would be to show that automated servo controlled FiO2 decreased retinopathy and shortened resolution of lung disease. I guess equipment manufacturers would then charge enormous amounts of money for their systems, so maybe we don’t want that! Just making life easier for families in the NICU by reducing alarms, and making life easier for caregivers (mostly nurses) by reducing alarm fatigue, that should be enough to make automated control standard of care, but it will be hard to find the money to buy the systems unless we can show there are fewer complications.

One of the reasons for writing about this now is this new publication from the Pediatrix group (Srivatsa B, et al. Oxygenation factors associated with retinopathy of prematurity in extremely low birth weight infants. J Pediatr. 2022), in a retrospective analysis of oxygenation changes during ventilatory assistance, which was respiratory support of HFNC of 2 litres or anything more than that (CPAP etc.) whether or not the infants were getting supplemental oxygen. They included 101 ELBW babies who had at least 45 days of data during the first 2 months of life and survived to have an eye exam, from a single NICU, of whom 33 developed mild and 15 had severe RoP (6 needed treatment). They found the babies with severe RoP were exposed to higher FiO2 and had more frequent FiO2 adjustments compared with the group without RoP. Room air hyperoxia (>95%) occurred more frequently in the group without RoP whereas iatrogenic hyperoxia occurred more frequently in the group with severe RoP. Babies with severe ROP had more hypoxic episodes and a longer time spent in severe hypoxia (<80%) than those without ROP. You would hope that reducing those fluctuations would reduce RoP, but of course, you can’t be sure of that from an uncontrolled observational study like this.

A recent pre-post study (Salverda HH, et al. The effect of automated oxygen control on clinical outcomes in preterm infants: a pre- and post-implementation cohort study. Eur J Pediatr. 2021;180(7):2107-13) was unable, however, to show a benefit of automated oxygen adjustments, they analyzed cohorts of 24 to 29 week gestation babies over about 3 years before and 3 years after routine use of automated oxygen adjustment. The babies in the cohorts look quite similar in terms of risk factors, during the after period, duration of intubation and CPAP was less, and HFNC was more, but that might of course have nothing to do with the O2 system. Rates of retinopathy were low and stable, and there was a bit more severe BPD and a bit less moderate BPD after implementation, but that could easily just be due to practice patterns changing and the use of HFNC, which tends to inflate rates of diagnosis of severe BPD (more about that another time!)

Showing a reduction in RoP, for example, when only 10% of the eligible babies develop the condition, and only half of them need treatment (as in these cohorts in Salverda et al) will need a very large trial. Which is why I am grateful that an international European collaborative will enrol over 2300 babies <28 weeks.

But, as I already mentioned, even if there is no improvement in clinical outcomes, as long as they are not worse, I think there are major benefits to families of having fewer alarms, and, if I had access to devices to fit to my ventilators/CPAP/HFNC devices right now, I would do so.