Article Text

Download PDFPDF

Randomised cross-over study of automated oxygen control for preterm infants receiving nasal high flow
  1. Peter R Reynolds1,2,
  2. Thomas L Miller3,4,
  3. Leonithas I Volakis5,
  4. Nicky Holland1,
  5. George C Dungan5,6,
  6. Charles Christoph Roehr7,8,
  7. Kevin Ives7
  1. 1 Neonatal Intensive Care Unit, Ashford and St Peter’s Hospitals NHS Foundation Trust, Chertsey, Surrey, UK
  2. 2 School of Biological Sciences, Royal Holloway University of London, Egham, UK
  3. 3 Sidney Kimmel Medical College, Philadelphia, Pennsylvania, USA
  4. 4 Vixiar Medical, Baltimore, Massachusetts, USA
  5. 5 Vapotherm, Exeter, New Hampshire, USA
  6. 6 Canisius College, Buffalo, New York, USA
  7. 7 Oxford University Hospitals NHS Foundation Trust, Oxford, UK
  8. 8 Medical Sciences Division, Department of Paediatrics, University of Oxford, Oxford, UK
  1. Correspondence to Dr. Peter R Reynolds, Neonatal Intensive Care Unit, Ashford and St Peter’s Hospitals NHS Foundation Trust, Chertsey KT16 0PZ, UK; peter.reynolds1{at}


Objective To evaluate a prototype automated controller (IntellO2) of the inspired fraction of oxygen (FiO2) in maintaining a target range of oxygen saturation (SpO2) in preterm babies receiving nasal high flow (HF) via the Vapotherm Precision Flow.

Design Prospective two-centre order-randomised cross-over study.

Setting Neonatal intensive care units.

Patients Preterm infants receiving HF with FiO2 ≥25%.

Intervention Automated versus manual control of FiO2 to maintain a target SpO2 range of 90%–95% (or 90%–100% if FiO2=21%).

Main outcome measures The primary outcome measure was per cent of time spent within target SpO2 range. Secondary outcomes included the overall proportion and durations of SpO2 within specified hyperoxic and hypoxic ranges and the number of in-range episodes per hour.

Results Data were analysed from 30 preterm infants with median (IQR) gestation at birth of 26 (24–27) weeks, study age of 29 (18–53) days and study weight 1080 (959–1443) g. The target SpO2 range was achieved 80% of the time on automated (IntellO2) control (IQR 70%–87%) compared with 49% under manual control (IQR 40%–57%; p<0.0001). There were fewer episodes of SpO2 below 80% lasting at least 60 s under automated control (0 (IQR 0–1.25)) compared with manual control (5 (IQR 2.75–14)). There were no differences in the number of episodes per hour of SpO2 above 98% (4.5 (IQR 1.8–8.5) vs 5.5 (IQR 1.9–14); p=0.572) between the study arms.

Conclusions The IntellO2 automated oxygen controller maintained patients in the target SpO2 range significantly better than manual adjustments in preterm babies receiving HF.

Trial registration number NCT02074774.

  • automatic oxygen control
  • closed loop; IntellO2
  • neonatal intensive care
  • blood oxygen saturation
  • preterm babies
  • high flow
  • high flow nasal cannula
  • monitoring

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

What is already known on this topic?

  • Many neonatal units target higher SpO2 ranges (eg, 90%–95%) for preterm babies.

  • Automated control devices have been shown to improve SpO2 targeting.

  • Hypoxic episodes with SpO2 <80% for ≥60 s are associated with poorer outcomes in preterm infants.

What this study adds?

  • The IntellO2 device significantly improved the time spent in SpO2 target range from 49% under manual control to 80% under automated control.

  • Under automated control there were fewer and shorter episodes of SpO2 <80% for ≥60 s.

  • There were overall reductions in the proportion of time in hypoxic and hyperoxic SpO2 under automated control, with more time spent in air.


A recent meta-analysis from the NeoProm Collaboration confirmed that a lower SpO2 range was associated with a higher risk of death and necrotising enterocolitis.1 As a result, many neonatal units, including our own, target a higher range of SpO2, for example, 90%–95%. For staff, maintaining SpO2 targets presents a compliance challenge.2 Avoiding both hypoxia and hyperoxia is an important goal.3 While additional training improves compliance,4 manual maintenance of the target range 90%–95% may only be achieved less than 50% of the time.5

A meta-analysis concluded that improved targeted SpO2 with reduced hypoxia and hyperoxia can be achieved using automated control of inspired oxygen concentration (FiO2) in babies requiring both invasive and non-invasive ventilation (NIV).6

This study evaluates a new device (IntellO2, Vapotherm, USA) which controls the Precision Flow (Vapotherm, USA) delivering nasal high flow (HF). We hypothesised that, for preterm babies on HF, the IntellO2 would maintain an SpO2 target range of 90%–95% for a greater percentage of time compared with standard (manual) practice.


Study design and setting

Our study was a prospective, two-centre, order-randomised, cross-over trial of HF employing automated versus manual oxygen titration, conducted in the neonatal intensive care units (NICU) at St Peter’s Hospital, Surrey, UK, and Oxford University Hospitals NHS Foundation Trust, Oxford, UK. Both units predominantly use HF in preference to nasal continuous positive airway pressure (CPAP) for NIV according to published guidelines.7–9 Staff received device and study protocol training, including a reminder of the importance of maintaining SpO2 targets. Written parental consent was obtained for all patients.

Study patients

Patients were eligible to participate if they were preterm and receiving HF at FiO2 ≥25%. An a priori inclusion criterion was that patients would require at least 12 adjustments of FiO2 during the manual arm to ensure adequate algorithm testing.10 This would only be apparent once the manual arm had been completed. Exclusion criteria were: (A) presence of major congenital abnormalities; (B) haemodynamic instability; (C) seizures; (D) ongoing sepsis; (E) meningitis; or (F) clinician’s concern regarding infant stability. By these criteria, we aimed to avoid studying babies whose clinical condition was either deteriorating or improving at a rate that would have impacted on the 48 hours’ study window. Nursing staff were very experienced in the use of HF in the study population and manual adjustments would reflect their standard practice. In accordance with UK practice the ratio of nurse to patient was 1:2 or 1:3.

IntellO2 device

The IntellO2 works with the Precision Flow, using a modified closed-loop control algorithm,11 employing pulse oximetry as the primary input signal with signal averaging set at 8 s (Masimo, Irvine, USA) to target a user-set SpO2 value. Further details are provided in online supplementary appendix 1.

Study protocols

Babies were randomised to commence either on manual or automated mode. SpO2 (alarm range 90%–95%) was continuously monitored on the NICU monitors as per normal standard of care12; a second pulse oximetry probe was placed initially on the right wrist for the IntellO2 input. Staff were instructed to care for the baby normally in both study arms, including SpO2 probe repositioning.

In manual mode, all FiO2 adjustments were made by clinical staff as needed. In automated mode, FiO2 was adjusted by the IntellO2, set to maintain a single SpO2 value of 93%. Staff could adjust all settings, including oxygen, in both modes, depending on clinical judgement.

In both arms, data for FiO2, SpO2, pulse rate, flow, mode and manual adjustments were logged by the IntellO2. Staff recorded the time and reason for all manual oxygen adjustments, as well as any cares or procedures which could affect SpO2. The study period in each arm was 24 hours’ run consecutively.

The primary outcome was the time in target SpO2 range 90%–95% (90%–100% if FiO2=21%) by pulse oximetry. Secondary outcomes included the number of episodes of SpO2 below 80% lasting at least 60 s, duration and frequency of episodes with SpO2 above 95% or below 90%, frequency of FiO2 adjustments, and overall mean FiO2 and flow.

Statistical design and analysis

A sample requirement of 30 patients was determined based on a pilot study (Saslow JG; unpublished) to provide 90% power, and 0.05 significance (two sided) to show difference of 19% in time spent within target SpO2 range. Each subject acted as their own control. The study was defined as intention to treat, thus manual adjustments in the automated arm would be included in that arm. Quantification of SpO2 episodes above, in and below the target range was performed. Mean flow (L/min) and FiO2 were calculated for each patient over the 24 hours’ study period. The known limitations of the Precision Flow device, where displayed FiO2 21%–23% delivers FiO2=21%, and displayed FiO2 98%–100% delivers FiO2=100%, were accounted for in the analysis. Babies were not deemed to be ‘out of range’ if their SpO2 was above 95% and their FiO2 was 21%, in line with clinical practice.

Subjects were block randomised using statistical software, and assignments were maintained in consecutively numbered opaque envelopes until randomisation and assignment to treatment arm by the consenting researcher. Statistical analysis was performed using Wilcoxon signed-rank test, with findings reported as median values with IQRs (median (IQR)), unless otherwise annotated. Ranges are also reported for demographics. Categorical variables are presented as proportion of subjects in each category. Statistical significance was accepted where p<0.05. Analyses were performed using MedCalc (V.18, MedCalc Software, Brussels, Belgium) and Minitab (V.18.1, Minitab, PA, USA).

The study was registered at NCT02074774. Independent Data Safety Monitoring Board (DSMB) review was undertaken at two points during the study.



Data sets from 30 different patients with mean gestational age of 26.4 weeks were evaluated for the recruitment period of December 2016 to November 2017. Patient characteristics are presented in table 1.

Table 1

Patient demographics

Data analysis

Seven data sets were excluded from five subjects. Additionally six different patients had some issues during the study; these are described in more detail in online supplementary appendix 2. After discussion with the DSMB they were included in the analysis on an intention-to-treat basis.

Time in target SpO2 range

Babies in the automated arm spent significantly more time in the target range versus the manual arm (automated 80% (70–87) vs manual 49% (40–57); p<0.0001). Figure 1 shows the pooled histogram for SpO2, with 93% as the most common SpO2 in both study arms. Figure 2 shows the histogram of FiO2 values for all patients in both study arms. There was less variation for the SpO2 values in the automated arm versus the manual (coefficient of variation, respectively, 0.03 (0.03–0.04) vs 0.06 (0.05–0.07); p<0.00001). The most common FiO2 in automated mode was 21% (range 21–100) versus FiO2 of 30% in manual mode (range 21–73). Babies had more episodes per hour within the target SpO2 range in the automated arm (54 events/hour (39–62) vs 39 events/hour (34–48); p=0.0002) with similar episode durations between the arms (45 s (36–74) vs 43 s (26–56); p=0.025).

Figure 1

Composite SpO2 histogram of all patient data (n=30) with paired bars as automated control (white) and manual control (black). The frequency of SpO2 values denotes the proportion of total time (%) spent at each SpO2, with aggregated SpO2 values <80%. The target SpO2 range for babies receiving oxygen (90%–95%) is illustrated by the shaded region.

Figure 2

Proportion of all patient data (n=30) as a percentage of total study duration in each arm by FiO2 (%). Values for automated control (grey) and manual control (black) are shown for each 1% FiO2interval.

Improvement for time in target range

All babies showed improvements in time spent in the SpO2 target range with mean improvement of 30% (range 10%–60%) (figure 3).

Figure 3

Comparison of percent time in target range of paired manual and automated control. Individual paired values (connected by line) denote the same patient. All completed patients shown (n=30). Horizontal bars denote the median of the associated control arm data. Vertical bars denote the IQR. Data represent the SpO2 target range 90%–95% (or 90%–100% if FiO2=21%).

Hypoxia and hyperoxia

The babies spent less time at any SpO2 below the target range (12% vs 28%; p<0.0001) and specifically less time at SpO2 <80% (0.5% vs 2.3%; p<0.0001) during automated control. There were no differences in the number of episodes below the target range, but they were of shorter duration in the automated arm as compared with the manual arm (17 vs 42 s; p<0.0001). There were fewer episodes of SpO2 below 70% in the automated arm (0.21 vs 0.98 episodes/hour; p<0.001), of slightly shorter average duration (11 vs 14 s; p=0.0006). Likewise, babies spent a lower proportion of time with SpO2 >95% in the automated versus manual arm (12% vs 23%; p<0.0001). There were more oversaturation episodes in the automated arm than the manual arm (37 vs 18 episodes/hour; p<0.0001) but the episodes were shorter in duration (12 vs 48 s; p<0.0001). At SpO2 above 98% there were no differences in the episode frequency. These data are presented in table 2.

Table 2

Oxygen saturation range compliance

Occurrence of hypoxia within 60–180 s after a return to FiO2=21% was higher in the automated arm compared with the manual arm (10 events/hour (5–18) vs 0 (0–0.1); p<0.001). Likewise following this event, hyperoxia within 60–180 s after a return to FiO2=21% was also higher for the automated arm compared with the manual arm (8 events/hour (3–15) vs 0 (0–0.1); p<0.001). Overshoot, defined as hyperoxia lasting at least 5 s above SpO2 range following hypoxia, was higher (as % of total study time) in the automated group (13% (6–18) vs 8% (3–19); p=0.021), but the episodes were shorter.

Flow and FiO2

There was no difference in flow between automated and manual control (6.0 (5.0–6.6) L/min vs 5.8 (4.9–7.0) L/min; p=0.2790). The mean FiO2 was greater in the automated arm than in the manual arm (34% (0.29–0.38) vs 29% (0.27–0.36); p<0.0001). The FiO2 at the start of each patient study ranged from 25% to 53%.

Adjustments to FiO2

The overall dynamics of adjustments to FiO2 were different between automated and manual modes. There were more FiO2 adjustments per hour during the automated versus manual mode (96 (93–101) adjustments/hour vs 1.6 (1.1–2.4) adjustments/hour; p<0.0001). Manual ‘over-ride’ adjustments of FiO2 during automated mode occurred less than 0.001% of the time, and the median number of FiO2 ‘over-ride’ adjustments was 0 (0–1), with range 0–11. Three patients had one manual adjustment during automated mode. Three patients had two manual adjustments during automated mode. Fewer manual adjustments were made in automated versus manual mode (p<0.0001).

Example of patient SpO2 and FiO2 traces

Figure 4 depicts examples of 24 hours’ SpO2 and FiO2 traces (automated and manual modes) from two babies labelled A and B. Baby A was born at 26+6 weeks and recruited into the study on day 13 in an initial FiO2=38% and flow of 6.0 L/min. Baby B was born at 23+3 weeks, with the study commenced on day 54, in FiO2=34% on a flow of 6 L/min. The graphs demonstrate the variability in both SpO2 and FiO2 over the 24 hours’ study periods, with greater variability in SpO2 in manual mode and greater variability in FiO2 in automated mode. Note the rising FiO2 in baby B automatically applied to maintain a stable SpO2 range. The baby was maintained on HF throughout and the following day completed the manual arm.

Figure 4

Top row Baby A. Bottom row Baby B. Complete 24 hours’ recordings for each arm. Left manual control and right automated control recordings of SpO2 (black line; y-axis is SpO2) and FiO2 (grey line, y-axis is FiO2).


Under automated control by the IntellO2 the babies in this study spent 31% more time in the target SpO2 range (80% vs 49%), which was a significant improvement and supported the first part of the study hypothesis. Patients also spent less time in hypoxic or hyperoxic SpO2 ranges under automated control. Every baby showed an improvement in their SpO2 targeting although we did not identify why some showed greater improvements than others. In a recent systematic review, SpO2 targeting compliance was also improved during automated control, both in infants being mechanically ventilated as well as NIV, with a total mean difference of 12.9% (range 6.5%–19.2%).6

Plottier et al studied 20 preterm babies on HF or CPAP and showed a 25% improvement in SpO2 targeting compared with manual control, with reductions in hypoxia and hyperoxia.13 In their study, the automatic arm was 4 hours’ duration, with a researcher present throughout. In our study, the babies received normal nursing care and interventions with no additional staff present, with each arm running for 24 hours, including weekends. Like Plottier et al, our study was unblinded, and it is possible that our nursing staff controlled the SpO2 range in the manual arm more diligently as they were aware that the values were being recorded, although this would be expected to have the effect of reducing the magnitude of any differences seen. Manual targeting of SpO2 to a precise range is difficult, and a systematic review showed that while compliance was generally low (about 33%–54%)2 it can be improved by improving the nurse to patient ratio.14 In our study, 73% had a 1:2 care ratio and 17% a 1:3 care ratio (not recorded in 10%). The higher care ratios in the majority, and the high starting FiO2 in many babies, give an indication that while they were sufficiently stable to be enrolled, many had considerable oxygen requirements and nursing workload. We reported flow as it is an important variable in babies receiving HF and is, along with FiO2, a proxy for respiratory stability.

The IntellO2 reduced the amount of time babies spent below the target range, especially at SpO2 less than 80% mainly through reducing the duration of episodes rather than their frequency. Poets et al described the significantly increased risks of death or developmental delay in babies with hypoxia where the episodes lasted for 60 s or more.15 Hypoxic episodes were generally short in both arms, but we specifically examined SpO2 values below 80% lasting for a minute or more, and found a reduction during the automated arm. Hypoxia increases the risks of mortality and necrotising enterocolitis1; our study was not designed to detect these important but infrequent events.

The variation in adjustments between the modes shows that while there were many more adjustments in the automated mode, the number of manual over-rides in the automated mode was very low (maximum two per patient). Nursing staff reported that the automated adjustments were often far greater in number and magnitude than they would normally perform, but also that the adjustments were rapid. The tendency of the algorithm to alternate between oversaturation and undersaturation may be due to excessive FiO2 adjustments, and ‘damping’ of this response may be required to further improve the stability of oxygen control. It is not known whether this degree and/or rate of fluctuation of FiO2 is clinically important at tissue level, given that these episodes were brief and overall babies spent more time adhering to the target SpO2 range.

In our study, babies spent more time with FiO2=21% in the automated arm, and while the minimisation of unnecessary oxygen exposure is desirable, the stability of the median FiO2 may also be an important consideration. Larger studies are thus needed to examine clinical outcomes affected by oxygen exposure, such as Retinopathy of Prematurity (ROP)16 , Bronchopulmonary Dysplasia (BPD)17 and neurodevelopmental outcomes1 in the context of automated oxygen control. In the meantime, we cannot necessarily assume that improved targeting will lead to a reduction in these morbidities.

Our study was a pragmatic study of oxygen requirement and adjustment, so that we enrolled a heterogeneous group of preterm babies who were sufficiently clinically stable to be enrolled, but who required oxygen and needed regular adjustments to their FiO2 when in the manual mode to stay in SpO2 target range. The cause of their oxygen requirement was not recorded. With a cross-over design we considered that the stability of the baby would not be expected to vary greatly over the 48 hours’ study period and that babies with episodes of apparent intrapulmonary shunting should not be excluded from the study provided they remained on HF.

On a cautionary note, automated control of oxygen has the potential to reduce the reliability of desaturation episodes as a clinical diagnostic indicator of the baby’s stability. If automated devices are to be routinely introduced, alternative indicators of clinical deterioration must be robust to prevent false reassurance provided by a stable SpO2.


The IntellO2 device in automated control mode maintained the babies’ SpO2 in the target SpO2 range significantly more effectively than manual control, and reduced the duration of hypoxic and hyperoxic episodes. A larger study is needed to determine if this better targeting would improve clinical outcomes.


We thank Nancy Gordon and Manya Harsch for providing statistical advice. We thank Professor Ben Stenson and Dr Nigel Kennea for their Independent Data Safety Monitoring Board work.



  • Funding This study was sponsored by Vapotherm, Exeter, NH.

  • Competing interests PRR and KI have received travel support and undertaken consulting work for Vapotherm. TLM, LIV and GCD were employees of Vapotherm during the study. NH and CCR have no declarations.

  • Patient consent Written parental consent was obtained for every study participant

  • Ethics approval MHRA Devices Division and Research Ethics Committee (London-Chelsea 16/LO/1272).

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement We are happy to share aggregated data; individual data are not made available to ensure that individual subjects cannot be identified.

Linked Articles