A Computer That Automatically Adjusts a Premature Baby's Oxygen Was Safe, but Did Not Prevent the Illnesses Doctors Hoped It Would

What a Large International Study of More Than 1,000 Extremely Premature Babies Found About Automated Oxygen Control

Premature babies born many weeks early often need extra oxygen to breathe, but too little or too much oxygen can each cause lasting harm. A major international study tested whether a computer that automatically fine-tunes a baby's oxygen level works better than a nurse adjusting it by hand. The computer system was safe over months of use, but it did not reduce death, serious bowel disease, chronic lung disease, or serious eye disease compared with careful hand adjustment [1].

Why Getting Oxygen Right Is So Hard for the Tiniest Babies

Babies born before 28 weeks of pregnancy — sometimes twelve or more weeks before their due date — have lungs and breathing-control centers in the brain that are still very immature. Their oxygen levels drift up and down many times a day, sometimes many times an hour, as their breathing pauses briefly or becomes shallow. Doctors and nurses respond by turning a dial that controls how much oxygen is added to the air the baby breathes. Get it wrong in one direction — too little oxygen for too long — and the baby's organs and brain can be starved of oxygen when it matters most. Get it wrong the other way — too much oxygen for too long — and the excess oxygen itself can injure the delicate blood vessels growing in the retina and the air sacs still forming in the lungs. Because a single nurse is often caring for two or three babies at once, and because these swings can happen faster than any person can react, this has been one of neonatal intensive care's most persistent, unglamorous problems for decades.

How Families and Doctors Faced This Before Automated Systems Existed

The dangers of getting newborn oxygen wrong were discovered the hard way, generations ago, when doctors realized that giving premature babies generous, unmonitored oxygen was blinding some of them through a condition now called retinopathy of prematurity. Part of why these babies' oxygen levels swing so much in the first place is that the brain's breathing-control centre and breathing muscles are still learning to work together, producing frequent short pauses and shallow breaths that researchers have mapped out in detail [2]. Those dips are not harmless: careful follow-up of premature babies found that the ones with a heavier day-to-day burden of low-oxygen episodes went on to have more deaths and disability later, regardless of which oxygen target their care team was aiming for [3]. At the same time, a landmark analysis combining data from thousands of babies in earlier trials confirmed there was no single "safe" oxygen target that avoided every risk at once — lower targets protected the eyes and lungs but meant more deaths and bowel injuries, while higher targets did the reverse [4]. Too much oxygen causes its own separate problem for the lungs: the excess triggers inflammation that interferes with the tiny air sacs still forming, contributing to the chronic lung disease of prematurity [5]. For years, the practical answer available to families and clinicians was simply more vigilant hand-adjustment: closer monitoring, tighter alarms, and hoping the bedside team could keep up. It was against this backdrop, more than twenty years ago, that engineers and neonatologists began building computer systems designed to do the fine, constant adjusting that no human nurse can sustain around the clock, and the first randomized test of such a system showed real promise — babies spent more time in their target oxygen range and needed far fewer manual adjustments [6]. Neonatal care has seen this pattern before: brain-oxygen monitors for premature babies were also shown, in earlier research, to improve the measurement doctors were watching, years before anyone tested whether that improvement actually changed outcomes for the baby [7]. A thorough 2023 review of the automated-oxygen evidence confirmed these systems reliably kept oxygen levels steadier, but it also pointed out something important: none of those earlier, shorter studies had actually tested whether steadier oxygen levels translated into healthier babies at the time of hospital discharge [8]. That was the open question this new trial, planned and registered years in advance [9],[10], was built to answer.

What the Study Did

Researchers in Germany, the Netherlands, China, and the United Kingdom enrolled 1,082 babies born between about 23 and 28 weeks of pregnancy at 32 different newborn intensive care units — the great majority (27 units, 80% of babies) in Germany. Researchers call the study, for short, the FiO2-C trial — shorthand for "closed-loop automatic control of the fraction of inspired oxygen." Half the babies were randomly assigned to have a closed-loop computer system continuously fine-tune their oxygen level, working alongside the usual nursing care; the other half received oxygen adjusted by hand, exactly as is standard in most NICUs around the world today. Everything else — the oxygen saturation target the team aimed for, the caffeine medication routinely given to help preemies breathe, and the overall quality of nursing care — was kept the same between the two groups, so that any difference in outcome could be attributed to the automated system itself rather than to some other difference in care. The babies were then followed to see whether they developed any of four serious problems: death, the intestinal disease necrotizing enterocolitis, chronic lung disease called bronchopulmonary dysplasia, or severe retinopathy of prematurity affecting the eyes.

What the Study Found

The automated system made no meaningful difference. Among babies with the computer-assisted oxygen control, 39% experienced at least one of the four serious problems, compared with 41% of babies who received hand-adjusted oxygen only — a difference small enough that it could easily have arisen by chance [1]. Looking at each problem separately told the same story: about 9% of babies died in each group; roughly 5% versus 7% developed the bowel disease; about 21% versus 23% developed chronic lung disease; and 18% versus 19% developed severe eye disease. Even when the researchers looked only at babies who had reliably used the automated system as intended, or looked separately at boys and girls, different gestational ages, and different brands of equipment, the result held: no clear benefit emerged. The one clearly reassuring finding was safety — over roughly five years of use across all these hospitals, the automated systems caused no pattern of harm, and the handful of software glitches that did occur were fixed without any baby being hurt.

What This Means for Families

If your baby's NICU uses one of these automated oxygen-control systems, this research suggests you can feel reassured about its safety — it has now been tested in over a thousand of the tiniest, most vulnerable babies over months of continuous use with no signal of danger. What this study does not support is an expectation that the technology, by itself, will lower your baby's individual risk of the four serious problems it was designed to prevent, at least not more than attentive hand-adjustment already does. Where it may genuinely help is behind the scenes: by taking over a repetitive, exhausting task, it frees nurses to spend more attention on other aspects of your baby's care, which matters even if it does not show up as a difference in this particular set of outcomes. It is worth asking your care team which system, if any, their unit uses, since the researchers noted hints that not all of the different computer systems tested may perform identically — a question future research is expected to sort out in more detail. It is also worth knowing that a computer system was never meant to replace the nurses and doctors watching your baby — in this study, it worked alongside routine bedside care the whole time, with staff free to step in and adjust the oxygen by hand whenever they judged it necessary, and that combination of person-plus-computer, rather than computer alone, is what was actually tested and found safe.

What Researchers Are Working on Next

The study's authors are still analyzing an enormous amount of additional data — roughly a million hours of recorded oxygen readings — that should help clarify whether the automated system truly did keep oxygen steadier for these specific babies, and if so, why that steadiness did not translate into fewer illnesses. Researchers are also expected to look more closely at whether particular computer systems, among the several tested here, perform better than others, since this study was not designed or large enough to make that comparison with confidence. Because the study stopped earlier than planned due to slower-than-expected enrollment, a somewhat larger or longer follow-up study could still detect a modest real benefit that this one was not quite big enough to see. For now, the clearest, most useful message for families is that this technology appears safe, that it eases staff workload, and that its ability to actually prevent the illnesses of extreme prematurity remains, honestly, unproven. Future trials may also test starting the automated system sooner after birth, keeping it running for longer before switching it off, or studying it in hospitals with fewer nurses per baby than the mostly European hospitals in this study — any of which could turn up a benefit that this particular study was not quite designed to find.

References

  1. Franz AR, Maiwald CA, Niemarkt HJ, et al., for the FiO2-C Study Group. Automatic versus manual control of oxygen and neonatal clinical outcomes in extremely preterm infants: a multicentre, parallel-group, randomised, controlled, superiority trial. Lancet Child Adolesc Health. 2026;10(3):179–188. doi:10.1016/S2352-4642(25)00351-700351-7) ↩
  2. Martin RJ, Di Fiore JM, Macfarlane PM, Wilson CG. Physiologic Basis for Intermittent Hypoxic Episodes in Preterm Infants. Adv Exp Med Biol. 2012;758:351–358. doi:10.1007/978-94-007-4584-1_47 ↩
  3. Poets CF, Roberts RS, Schmidt B, et al., Canadian Oxygen Trial Investigators. Association Between Intermittent Hypoxemia or Bradycardia and Late Death or Disability in Extremely Preterm Infants. JAMA. 2015;314(6):595–603. doi:10.1001/jama.2015.8841 ↩
  4. Askie LM, Darlow BA, Finer N, et al., NeOProM Collaboration. Association Between Oxygen Saturation Targeting and Death or Disability in Extremely Preterm Infants in the Neonatal Oxygenation Prospective Meta-analysis Collaboration. JAMA. 2018;319(21):2190–2201. doi:10.1001/jama.2018.5725 ↩
  5. Kalikkot Thekkeveedu R, Guaman MC, Shivanna B. Bronchopulmonary dysplasia: A review of pathogenesis and pathophysiology. Respir Med. 2017;132:170–177. doi:10.1016/j.rmed.2017.10.014 ↩
  6. Urschitz MS, Horn W, Seyfang A, et al. Automatic control of the inspired oxygen fraction in preterm infants: a randomized crossover trial. Am J Respir Crit Care Med. 2004;170(10):1095–1100. doi:10.1164/rccm.200407-929OC ↩
  7. Hyttel-Sørensen S, Pellicer A, Alderliesten T, et al. Cerebral near infrared spectroscopy oximetry in extremely preterm infants: phase II randomised clinical trial. BMJ. 2015;350:g7635. doi:10.1136/bmj.g7635 ↩
  8. Stafford IG, Lai NM, Tan K. Automated oxygen delivery for preterm infants with respiratory dysfunction. Cochrane Database Syst Rev. 2023;11:CD013294. doi:10.1002/14651858.CD013294.pub2 ↩
  9. Maiwald CA, Niemarkt HJ, Poets CF, et al., FiO2-C Study Group. Effects of closed-loop automatic control of the inspiratory fraction of oxygen (FiO2-C) on outcome of extremely preterm infants — study protocol of a randomized controlled parallel group multicenter trial for safety and efficacy. BMC Pediatr. 2019;19:363. doi:10.1186/s12887-019-1735-9 ↩
  10. König J, Stauch A, Engel C, et al., FiO2-C Study Group. Statistical analysis plan for the FiO2-C trial: effects of closed-loop automatic control of the inspiratory fraction of oxygen (FiO2-C) on outcomes of extremely preterm infants — a randomized-controlled parallel group multicentre trial for safety and efficacy. Trials. 2024;25:756. doi:10.1186/s13063-024-08615-7 ↩