Neonatology

Performance and safety of the PRICO closed-loop oxygen saturation targeting system in neonates: pragmatic multicentre cross-over study (TarOx Study)

Abstract

Objective This study aims to evaluate the performance of the fabian-Predictive-Intelligent-Control-of-Oxygenation (PRICO) system for automated control of the fraction of inspired oxygen (FiO2).

Design Multicentre randomised cross-over study.

Setting Five neonatal intensive care units experienced with automated control of FiO2 and the fabian ventilator.

Patients 39 infants: median gestational age of 27 weeks (IQR: 26–30), postnatal age 7 days (IQR: 2–17), weight 1120 g (IQR: 915–1588), FiO2 0.32 (IQR: 0.22–0.43) receiving both non-invasive (27) and invasive (12) respiratory support.

Intervention Randomised sequential 24-hour periods of automated and manual FiO2 control.

Main outcome measures Proportion (%) of time in normoxaemia (90%–95% with FiO2>0.21 and 90%–100% when FiO2=0.21) was the primary endpoint. Secondary endpoints were severe hypoxaemia (<80%) and severe hyperoxaemia (>98% with FiO2>0.21) and prevalence of episodes ≥60 s at these two SpO2 extremes.

Results During automated control, subjects spent more time in normoxaemia (74%±22% vs 51%±22%, p<0.001) with less time above and below (<90% (9%±8% vs 12%±11%, p<0.001) and >95% with FiO2>0.21 (16%±19% vs 35%±24%) p<0.001). They spent less time in severe hyperoxaemia (1% (0%–3.5%) vs 5% (1%–10%), p<0.001) but exposure to severe hypoxaemia was low in both arms and not different. The differences in prolonged episodes of SpO2 were consistent with the times at extremes.

Conclusions This study demonstrates the ability of the PRICO automated oxygen control algorithm to improve the maintenance of SpO2 in normoxaemia and to avoid hyperoxaemia without increasing hypoxaemia.

What is already known on this topic

  • Oxygenation targeting in newborn infants is essential but successful manual titration of oxygen is challenging.

  • Automated control of inspired oxygen is now available and shows great promise, but efficacy and safety may vary between the different available algorithms.

What this study adds

  • In a multicentre study, the fabian-Predictive-Intelligent-Control-of-Oxygenation (PRICO) automated system was shown to be effective and safe in neonates of different maturities among a range of ventilator modalities.

How this study might affect research, practice or policy

  • Based on this and other smaller studies, the fabian-PRICO can be routinely used in clinical practice.

  • Additional research is indicated to evaluate automated oxygen control systems to not only determine how they can be improved but also how their use can be optimised in different clinical environments.

Background

Respiratory instability leading to frequent hypoxaemic episodes is a common problem in preterm infants. Therefore, the SpO2 is continuously monitored, aiming to keep it within a target range by manually adjusting the fraction of inspired oxygen (FiO2).1 However, nurses struggle with this task, and as a result, hyperoxaemia and hypoxaemia are prevalent.2 3 Both are associated with adverse outcomes in preterm infants.4–8 Improving oxygen targeting in preterm infants is, therefore, needed.

Modern neonatal ventilators offer the option of automated closed-loop oxygen control (A-FiO2). During A-FiO2, a software algorithm adjusts the delivered FiO2 based on the measured SpO2 and the set target range. Many studies have reported that A-FiO2, compared with manual control (M-FiO2), improves the time spent within the target range and reduces the exposure to hypoxaemia and/or hyperoxaemia.9 10 However, A-FiO2 control algorithms differ considerably,11 and therefore, the performance of each algorithm needs to be studied under different conditions before conclusions can be drawn on its relative effectiveness and limitations. The Predictive-Intelligent-Control-of-Oxygenation (PRICO) A-FiO2 algorithm, introduced in 2016, had only been evaluated in two small single-centre studies.12 13 Therefore, the aim of this study was to evaluate the efficacy and safety of the PRICO algorithm in a wide range of infants supported with different modes of respiratory support, across multiple centres.

Methods

This was a multicentre randomised cross-over study in five European neonatal intensive care unit (NICUs), one in the Netherlands and four in Poland. Potential sites were only considered if they had experience with the fabian ventilator (Vyaire Medical, Mettawa, USA) and A-FiO2. The study was sponsored by Vyaire and designed as a postmarket clinical follow-up study conducted to fulfil the EU Medical Device Regulations and registered at ClinicalTrials.gov (NCT 04957472). There was no patient or public involvement in designing the study. An independent data safety monitoring committee conducted interim safety reviews masked to the sponsor and the investigators. A contract research organisation managed the study activities including site monitoring and collection of adverse events. This is an independent report from the investigators, which was neither commissioned nor influenced by the sponsor.

A-FiO2 system

The PRICO A-FiO2 option of the fabian ventilator is available for all invasive and non-invasive respiratory support modalities. PRICO uses a rule-based algorithm, first described by Hütten et al.14 It monitors the SpO2 every second using an integrated Masimo pulse oximeter (Masimo Corporation, California, USA). Based on a weighted average of these data, an adjustment in FiO2 is made every 30 s, if warranted. When within the set SpO2 target range, the FiO2 adjustment is ±1% towards the midpoint. When outside the target range the adjustment varies (±1%–10%), based on a proprietary algorithm that takes into account the depth and the trajectory of the predicted response to a change in oxygen. In certain conditions (no SpO2 reading (drop-out) or exceeding operator set parameters) the system switches off automated control and falls back to manual control at the FiO2 previously set by the clinician, while generating an alarm. The system returns to A-FiO2 control when the condition resolves or after manual reactivation.

Study population

Infants in the NICU needing a target of 90%–95% SpO2, regardless of weight or gestational age, were eligible for enrolment if they required invasive or non-invasive respiratory support and supplemental oxygen (>0.21). Infants were excluded if they had congenital anomalies or uncontrolled haemodynamic instability. In addition, infants were also excluded if the attending physician did not believe participation was in the best interest of the infant, or if parental consent was not obtained.

Study protocol

Subjects, a sample of convenience, were assigned to sequential 24-hour periods of A-FiO2 and M-FiO2 control in random order. Except for the order of intervention, all other aspects of care, including all PRICO settings (oximeter averaging, SpO2 alarm levels and target range thresholds), were according to each unit’s standard practice. For infants ready to start caffeine therapy, entry was delayed at least 12 hours to avoid caffeine wash-in effects. Subjects exited the study early if they required a change in the mode of respiratory support, or if the attending physician or the parents withdrew support for participation.

An electronic web-based system (CastorEDC, Amsterdam, Netherlands) was used for the collection of baseline characteristics and subject randomisation. The SpO2 values and ventilator settings were recorded on a laptop computer attached to ventilator during both arms of the study. These data were uploaded to the sponsor’s study portal and automatically incorporated into a database. The analyses were based on 4 s data points. Remote monitoring reports were automatically generated and sent to the study monitor after each enrolment, with potential inconsistencies adjudicated. Periodic site visits also audited study conduct.

Endpoints

The proportion of time in normoxaemia was the primary endpoint and specified as SpO2 90%–95% with FiO2>0.21, plus time 90%–100% with FiO2=0.21).9 15 Avoiding SpO2 extremes associated with severe hypoxaemia (<80%)5 16 and severe hyperoxaemia (>98% with FiO2>0.21)15 16 were also specified outcomes.

Other descriptive endpoints were (1) the proportion of time in each SpO2 bin between ≤80% and 100%, with and without supplemental oxygen; (2) the mean proportion of time above and below normoxaemia; (3) the mean FiO2 and SpO2, as well as the median SpO2 for the 2 min after oximeter drop-outs; (4) the number of FiO2 adjustments and (5) the frequency and duration of episodes outside target range. Episodes were defined as consecutive 4 s data points. Prolonged episodes were defined as those lasting for at least 60 s. Two post hoc analyses were conducted. The time with SpO2 between 90% and 95% during supplemental oxygenation was calculated, to assess the impact of differences in time on room air between automated and manual control. In addition, the primary outcomes were cross-tabulated by invasive and non-invasive respiratory support. To account for possible transition effects data from the first 10 min and last 5 min of each arm were excluded. All calculations of the proportion of time used time with an SpO2 signal as the denominator.

Endpoints were unchanged except for the post hoc analyses noted above.

Statistical considerations

A meta-analysis of 14 similar A-FiO2 cross-over study cohorts reported that A-FiO2 improved time within the SpO2 target range by 12.9%.9 We determined that we would be able to detect a 4% (±6% SD) difference between A-FiO2 and M-FiO2 with an alpha of 5% and a power >90% with 40 infants. Assuming 15% would not complete the study, an enrolment of 47 patients was specified.

Based on intention to treat, all subjects with exposure to both interventions of any duration were included in the analyses. The three outcome endpoints were also calculated for all enrolled subjects, as a sensitivity analysis to determine if their exclusion affected the results.

Variables and differences between paired variables were tested for normality (Shapiro-Wilk test), which showed that nearly all were not normally distributed. Therefore, non-parametric tests were used to evaluate the difference between automated and manual control periods (Wilcoxon signed-rank or Mann-Whitney for paired and nonpaired data, respectively). To facilitate comparison to previous and inclusion in future meta-analyses, times in, above and below the target range are reported as mean and SD. For consistency of presentation, other parameters are reported as median and IQR. A two-tailed p<0.05 was considered statistically significant for all comparisons. 95% CI in the difference in mean was determined for the primary outcome. Statistical tests were conducted with XLSTAT V.11.5 (Lumivero, New York, USA).

Results

Patients were enrolled between January and October 2022. Centres enrolled between 6 and 12 infants, accounting for a total of 47 infants. Eight subjects were excluded from the analysis: seven were not exposed to both control methods, and in one, no data were captured due to a technical problem (figure 1). Thus, 39 subjects, exposed to both A-FiO2 and M-FiO2, were included in the final intention-to-treat analyses.

Figure 1
Figure 1

Subject accounting. The two patients exited in the first A-FiO2 arm were at 22 and 24 hours (one was an intubation and the other an operator error (data logging turned off during manual intervention)). The five patients’ exits from the first arm of M-FiO2 were at 8, 19, 21, 24, 24 hours (two were weaned from HFNC, one was intubated and two were withdrawn by the attending). There were no early exits in the second arm during A-FiO2. Four patients exited early in the second arm of M-FiO2 at 5, 12, 14, 15 hours (two were an operator error with collection of SpO2 data, one withdrawal by attending and one was extubated). FiO2, fraction of inspired oxygen; HFNC, high-flow nasal cannula.

The characteristics of the study population are shown in table 1. Most of the subjects were less than 2 weeks at enrolment. More than half were less than 1000 g when enrolled, and a quarter between 2900 and 4160 g. A total of 27 infants (69%) were receiving noninvasive support (13 intermittent nasal positive pressure ventilation, 10 continuous nasal positive airway pressure, 4 high-flow nasal cannula (HFNC)) and the remaining 12 infants (31%) were supported by invasive mechanical ventilation. Eight infants (21%) were receiving A-FiO2 before study entry.

Table 1
|
Baseline enrolment

The primary outcome, time within normoxaemia and time at SpO2 extremes are detailed in table 2. During A-FiO2 control, the time in normoxaemia was higher than during manual control (74%±22% vs 51%±22%, respectively; p<0.001, mean difference 23%, 95% CI 16% to 30%). A-FiO2 was associated with a decrease in time with SpO2>98% when receiving supplemental oxygen (1% (IQR 0%–3.5%) vs 5% (IQR 1%–10%), p<0.001. Exposure to SpO2<80% was low in both arms and not significantly different. The first post hoc analysis, which included all enrolled subjects (ie, those not exposed to both A-FiO2 and M-FiO2), was consistent with these findings. Finally, these results were similar for infants supported non-invasively or invasively (post hoc results in online supplemental materials).

Table 2
|
End points

Secondary descriptive endpoints were consistent with the primary outcome findings. As shown in table 2, there were fewer total episodes and fewer prolonged episodes of hyperoxaemia during A-FiO2. In addition, there was no difference in number of prolonged episodes of hypoxaemia. There was, however, a slight reduction in the total number of severe hypoxaemic episodes associated with A-FiO2.

Table 3 provides information on the course of the interventions. During A-FiO2, automatic adjustments of FiO2 were quite frequent (80/hour (IQR: 50–108), but manual adjustments were markedly lower than during M-FiO2 (8.0/day (4.5–12) vs 30/day (17–106), p<0.001). The median FiO2 and SpO2 were similar in the two arms. However, during A-FiO2 the infants spent more time on room air, with a slightly lower median SpO2 (92.9% (92.4%–94.6%) vs 94.0% (92.2%–95.4%) p=0.025). The SpO2 histogram in figure 2 shows the SpO2 distribution during periods with and without supplemental oxygen. When receiving supplemental oxygen, there was a shift towards normoxaemia during A-FiO2, which is more apparent than the small difference in SpO2 medians (table 3). As expected during periods without supplemental oxygen, both histograms shifted to the right with similar and limited hypoxaemia exposure.

Figure 2
Figure 2

Histogram SpO2 with and without supplement O2. Histogram constructed with the mean of the subjects’ SpO2 in each bin. The grey lines are M-FiO2 and the black A-FiO2. The dashed lines are during periods without supplemental oxygen and the solid with supplemental oxygen. FiO2, fraction of inspired oxygen.

Table 3
|
Overview of intervention

There were four adverse events reported during the study period, two in each treatment arm. A detailed description can be found in online supplemental material. In addition, four subjects spent more time in normoxaemia during manual control.

Discussion

We conducted a multicentre pragmatic cross-over study to evaluate the performance of the PRICO A-FiO2 system. We found that this A-FiO2 system markedly increased time in normoxaemia which was mainly driven by a reduction in time spent in hyperoxaemia.

The difference in the mean time in the target range (23% (95% CI 16% to 30%)) is slightly higher than reported in two systematic reviews which included 14 studies investigating eight different A-FiO2 systems.9 10 The subgroup meta-analysis in one review, including only those studies using a target range of 90%–95% as we used, showed a difference of 18% (95% CI 5% to 32%).9 Our finding is within these confidence limits. In addition to differences in the A-FiO2 algorithms, the larger treatment effect might be related to manual oxygen titration practices, staffing levels and inclusion of bigger infants.

To date, three other studies have investigated the efficacy of the PRICO system in infants. These studies12 13 17 reported a median improvement of the time within the target range between 6% and 16%, again considerably lower than this study. This larger difference between A-FiO2 and M-FiO2 seems to be mediated on the one hand by a higher percentage of time within the target range during A-FiO2 with a lower percentage of time within the target range in the M-FiO2. While all studies, including ours, tended to reduce time above and below the target range, there was considerable variation in the reduction at SpO2 extremes in the different studies. Two suggested a more prominent effect in reducing the time in severe hypoxaemia13 17 while our study showed better performance of A-FiO2 in avoiding severe hyperoxaemia. One study reported a decrease in severe hypoxaemia but did not report exposure to severe hyperoxaemia.12 Although it is unclear what caused these differences in performance, it is important to emphasise, that there were important differences between the studies which may have impacted the efficacy of both automated and manual FiO2 control. One study by Dijkman et al, only included infants on HFNC, which allows for a large leak at the nasal interface.13 Two of the previous studies used a broader SpO2 target range (88%–95%)13 17 and a shorter measurement period (8 hours),17 which may also have impacted performance. Finally, there were differences in the patient characteristics, such as gestational age, respiratory support modalities and the postnatal age at inclusion. Overall, these findings suggest that the relative effectiveness of an A-FiO2 system is not fixed but will vary between different patients, target ranges and modes of respiratory support. As an example, two recent reports suggest the relative effectiveness of A-FiO2 compared with M-FiO2 is most marked in unstable infants.17 18

Although most of the focus when evaluating the performance of an A-FiO2 system has been on the time in the intended SpO2 target range, what happens with oxygenation during those episodes outside the target range is also of clinical importance. Especially, prolonged periods in extreme hyperoxaemia and hypoxaemia have been associated with neonatal morbidity in the short term and long term.5 8 We found that the differences between A-FiO2 and M-FiO2 in per cent time at SpO2 extremes were consistent with the number of prolonged episodes. This finding is consistent with other reports,5 19 further supporting the premise that reducing prolonged episodes outside target range is associated with improved time in the target range and leads to potential long-term benefits of A-FiO2.

Besides, the benefit of improved SpO2 control, A-FiO2 also promises to reduce nursing stress and labour. Only one of the previous studies of PRICO reported on the reduction in manual FiO2 adjustments.12 They reported a modest decrease of nine adjustments per day. In contrast we found a decrease of 22 per day (30–8). Clearly, the need for manual adjustments is dependent on the stability of the infant, which probably also explains the large IQR during manual control (upper quartile 106 adjustments/day) compared with automated control (upper quartile 12 adjustments/day).

We found no difference in the response to SpO2 dropouts between A-FiO2 and M-FiO2, suggesting PRICO’s fall-back to a clinician’s set FiO2 level is as effective as nursing response. Others have suggested the importance of the response to drop-outs,20 but we are unaware of published evaluations of the effectiveness of other A-FiO2 fall-back strategies for periods without reliable SpO2 readings.

Our report has some strengths and limitations. First of all, it enrolled more subjects than most evaluations of A-FiO2 systems. Also, the enrolment was not restricted to extremely preterm infants. In addition, it reflects a multicentre experience. While its cross-over design is the typical approach for evaluating the performance of an A-FiO2 system, the differences identified in this type study might not be projectable to routine care. This was a pragmatic study which generally provides a better basis for projection to the impact in routine use, however, the impact of different approaches to respiratory care and settings are confounding. While another strength of the study was the use of A-FiO2 in different respiratory support modes, the study was only powered to detect an overall difference in performance. Therefore, it was inadequate to evaluate the relative effect of different respiratory support modes. Particularly, additional studies during high-frequency oscillatory ventilation are warranted. Eight of the subjects were being managed by A-FiO2 when randomised, but a previous study suggests that this probably does not affect our findings.21 Finally, the study centres were all experienced with both the ventilator and the use of A-FiO2, and the finding should be projected to naïve centres cautiously.

Conclusions

The study demonstrates the ability of the PRICO automated oxygen control algorithm to improve the maintenance of SpO2 in normoxaemia and to avoid hyperoxaemia while reducing the need for nursing intervention, in a wide range of preterm infants supported with various respiratory support modalities.