Background

Continuous of monitoring of peripheral oxygen saturation (SpO2) is the standard of care for preterm infants requiring supplemental oxygenation. Keeping the SpO2 within an acceptable range and minimizing exposure to extremes is associated with better neonatal and long-term outcomes.

Titrating the fraction of inspired oxygen (FiO2) to maintain the SpO2 within a prescribed target range is a challenging task in daily care, resulting a mere 50% compliance within the intended target range [1]. As a result of disordered breathing and apnea, these infants frequently desaturate and caregivers sometimes respond with a transient increase in FiO2 to offset the potential hypoxemia. However, the response with an increase in FiO2 is often slow and inappropriate resulting in extended episodes of hypoxemia as well as swings to hyperoxemia. In addition, a failure to quickly reduce the FiO2 back to baseline levels once the desaturation has resolved further increases the risk of hyperoxemia [2]. The cumulative time in both extreme saturations is impacted by both the frequency and duration of these episodes. Previous studies have shown that the duration in extreme saturations is probably the key determinant for the risk of adverse outcomes [3, 4].

Given the limitations of manual FiO2 titration, automated FiO2 control algorithms for SpO2 have become available and studied. These studies have shown that automated control improves the time within the intended target range, and reduces time spend at the SpO2 extremes [5, 6]. It is unclear how automated FiO2 control impacts the frequency and duration of episodes of extreme hypoxemia and hyperoxemia. Some studies have reported a reduced frequency of one duration (1 min) associated with automated control [5]. More information is needed.

The aim of this post-hoc analysis was to provide additional detailed characterization of the frequency and duration of severe hypoxemic and hyperoxemic episodes, and specifically how they differ between automated and manual FiO2 control.

Methods

This was a prospectively defined analysis of retrospective data from randomized cross-over studies investigating automated versus manual FiO2 control in preterm infants 32 weeks gestational age or less in need of invasive or non-invasive respiratory support. The study was carried out in accordance with relevant guidelines and requirements.

Design

This is a post-hoc analysis of a cohort from prospective multicenter randomized cross-over studies [7,8,9,10].

Setting

The 7 study sites were tertiary care neonatal intensive care units in 6 different countries.

Subjects

Eligible subjects who were born less than 33 weeks gestational age were considered from four cross-over studies investigating manual versus automated FiO2 control (AVEA-CLiO2, Vyaire Mettawa IL, USA) in which individual subject data was available [7,8,9,10]. These studies used either two 24 or 12 hour intervention periods. To avoid potential bias, the two selection criteria were made prospectively, that is, without reviewing the outcomes. Since these individual studies used different SpO2 targets, and we did not want to confound the results, we selected only studies using a SpO2 mid-point of 90% and a target range width of 4%. This excluded two studies [9, 10]. Finally, subjects in the two remaining studies [7, 8] that did not spend at least 75% of time on supplemental oxygen were excluded to permit better characterization of hyperoxemic episodes. All these cases are from clinical trials with ethics approval, and are de-identified. All subjects were enrolled with written informed consent of their guardian(s).

Outcome measures

We defined the primary SpO2 exposure metrics for SpO2 extremes as < 80% (hypoxemia) and > 98% (hyperoxemia). Episodes of 600 seconds or longer were excluded as being likely related to procedures and not representative of routine SpO2 management. The primary endpoint was the percent time of SpO2 episodes in each of 5 episode-duration categories (< 5 s, 5 to < 30s, 30 to < 60s, 60 to < 120 s, 120 to < 600 s). We also determined the frequency of episodes within each duration epoch. This categorization was based on data collected with a resolution of every 5 seconds, and two consecutive data points were defined as 5 seconds.

Statistical analyses

With the sample of 58 subjects, we determined that we would have a > 80% chance of detecting an absolute difference of 10%-time, assuming an absolute variance of 25%-time, with an uncertainty of p < 0.05.

Extraction of the endpoints from the 5-second data points in the database was accomplished with purpose-built software (MatLab, Mathworks, Natick MA USA). Differences among the episode length categories and mode of FiO2 control were determined with the Kruskal-Wallis test with Dunn’s procedure for pairwise comparisons. A two-tailed p < 0.05 was considered statistically significant. Statistical tests were conducted with XLSTAT v19.03 software (Addinsoft, Paris, France).

Results

From the initial 179 potential subjects, using the prospective selection criteria, we made the following exclusions: two studies (59 subjects) were excluded because of their target range [9, 10], 40 subjects were excluded from one study that evaluated high and low target range cohorts [7] and finally, due to higher exposure to room air, in the remaining two studies [7, 8], 22 subjects were excluded (12, 10 respectively). Thus, we evaluated the SpO2 control of 58 preterm infants receiving respiratory support and supplemental oxygen. Fifty-one percent of the cases came from one study [7], and the rest from the second study [8]. The inspired oxygen in these studies was controlled manually (M-FiO2) for 1 day and automatically (A-FiO2) on the other, in random order. All the infants were on the same mode of respiratory support (noninvasive or intubated) during the days of manual and automated control. Most were managed noninvasively (83%). The demographics of the subjects are shown in Table 1, they were mostly extremely preterm and were studied weeks after birth. The subjects’ baseline oxygen needs were relatively low (median FiO2 0.28, IQR 0.25-0.32). A majority of the subjects spent no time on room air (59%). Those who did have periods without supplemental oxygen, spent a nominal amount of time without (median 4%, IQR 1-9%). The median SpO2 for both the automated and manual control methods were nearly identical (91%). However, there was a difference between the groups in the time spend outside normoxemia; the SpO2 was < 87% for 13% of the time during A-FiO2, and 16% during M-FiO2, whereas 11% of the time the SpO2 was > 95% during A-FiO2, and 16% during M-FiO2. These differences were statistically significant (p < 0.001).

Table 1 Subject demographics
Full size table

There were frequent episodes at SpO2 extremes with differences between the two methods of control as shown in the Table 2. Episodes at the SpO2 extremes (< 80 and > 98%), tended to be less frequent during A-FiO2 within each of the 5 episode-duration categories. Most of these extreme episodes were shorter than 30 seconds.

Table 2 Frequency of extreme hypoxemic and hyperoxemic episodes
Full size table

Our primary endpoint is shown in the Fig. 1. It presents the total time at< 80% and at > 98% SpO2, along with the contribution towards the total from each of the 5 episode-duration categories. The difference in the total time for A-FiO2 and M-FiO2 at each of these two endpoints was significantly different (p < 0.001), favoring A-FiO2. The time at each of the episode length categories tended to be lower during A-FiO2, and nearly all differences were statistically significantly significant [all but 5 to < 30 seconds with SpO2 < 80%]. Both shorter and longer episodes made a relevant contribution to the total duration of extreme SpO2 exposure with about half of the duration of time in hypoxemia < 80% resulted from episodes with a duration of 30 seconds or shorter, whereas one third of the duration of time with hyperoxemia > 98% was a result of episodes 30 seconds or shorter. In contrast, the duration of time < 80% from episodes 60 seconds or longer were 19% for A-FiO2 and 49% for M-FiO2. More than a third of the duration of time > 98% were from episodes 60 seconds or longer (34% A-FiO2, 38% M-FiO2). Importantly, though infrequent during both modes of control, nearly two-thirds of the reduction of total time spent at the extremes between automated and manual control (3.8 to 2.1% for < 80% SpO2 and 3.0 to 1.6% for > 98% SpO2) was from a reduction of episodes of 1 min or longer.

Fig. 1
figure 1

Stacked Histogram of % Time at SpO2 Extremes. Differences in % Time between automated fraction of inspired oxygen (A-FiO2) control and manual (M-FiO2) control: Total %Time (A-FiO2-M-FiO2):> 98% p < 0.001, < 80% < 0.0O1, For each of the 5 duration categories (ns, * < 0.05, ** < 0.01, *** < 0.001)

Full size image

Discussion

In a population cohort of extremely preterm infants receiving supplemental oxygen and respiratory support, we investigated the frequency and duration of marked hypoxemic and hyperoxemic episodes during both A-FiO2 and M-FiO2. We confirmed that the infrequent longer episodes were the primary contributor to total time at SpO2 extremes. We found that, compared to M-FiO2, automated control of SpO2 reduced the number of episodes in all durations, but that most of the reduction in the total exposure to these extremes was from a reduction of episodes of 1 min or longer.

Though we provide more detail, our hypoxemic episode results are consistent with other studies. The extensive report of Poet et al., found that hypoxemic exposure with manual FiO2 control is comprised of episodes of desaturations across a range of durations, with prolonged episodes being infrequent, but nevertheless contributing markedly to the total exposure to SpO2 less than 80% [3]. Other studies using this automated FiO2 control algorithm have reported a reduced frequency of hypoxemic episodes longer than a minute, compared to manual FiO2 control [9, 11]. Our study identified that the duration of hypoxemic episodes is the primary factor in A-FiO2 reducing total time in hypoxemia. One study of A-FiO2 reported that the total number of shorter episodes below the target range was increased but that the longer episodes were decreased with A-FiO2, suggesting that potentially longer episodes were compressed [9]. This was not confirmed in our study. Rather we found that automated FiO2 control also reduced the frequency of shorter episodes of hypoxemia. We suggest that this different finding is a result of a higher median SpO2 during manual in the previous report. The impact of such a shift is consistent with another study [7].

We believe our data are the first report of the details of the frequency and duration of episodes of hyperoxemia. We found, consistent with desaturations, that it is comprised of episodes across a range of duration, with prolonged episodes being infrequent, but nevertheless contributing markedly to the total exposure to SpO2 greater than 98%. Studies of this automated FiO2 control system have also shown a reduction of the frequency of hyperoxemic episodes longer than a minute [8,9,10,11], but our study demonstrated that this is the primary factor in reducing total time in hyperoxemia with A-FiO2.

There is limited information on how these differences might impact outcome, that has come from rigorous post-hoc analyses of the large randomized trials. One report found that better control of the SpO2, regardless of the target range, improved long term outcomes but they did not report the associated exposure to SpO2 extremes [12]. Another reported an association between cumulative exposure to SpO2 < 80%, which correlated to the number of prolonged episodes, with an increased risk of late death or disability at 18 months and also speculated that therapies that reduced these prolonged events could improve long-term outcomes [3]. Another research team reported that an increased frequency of all hypoxemic events was associated with severe retinopathy of prematurity and bronchopulmonary dysplasia [13, 14]. We speculate that more frequent desaturations would also correlate with more prolonged hypoxemic episodes, as well as an increased risk of overshoot to hyperoxemia. The impact of hyperoxemia has not been so carefully evaluated, but a landmark study published nearly 20 years ago confirmed that high levels of SpO2 are associated with severe retinopathy of prematurity and bronchopulmonary dysplasia, without a difference in hypoxemia [4]. Regardless, it is not clear whether these studies report a cause-effect relationship, or rather a marker of adverse outcome.

Alarm setting strategies ought to be studied much more. They should strive to reduce the time at SpO2 extremes, by specifically balancing the risk of missing important events because of false negatives versus not responding to important events due to alarm fatigue. These results may help in formulating strategies to reduce oximeter alarm fatigue by considering the trade-off between false-positive and false-negative alarms that is associated with the alarm delay. Our data show that during manual FiO2 control, most of the episodes of extreme SpO2 levels in this group of infants resolved on their own in less than 30 seconds. One might question the clinical relevance of the settings that trigger highly frequent alarms not needing attention. Nevertheless, these frequent alarms might be useful in alerting nurses to instability. In contrast, it seems reasonable that an A-FiO2 control system, which continuously makes FiO2 adjustments as often as many times per minute, might need different alarm delays. It is important to note that in the case of A-FiO2 control, a saturation alarm indicates that changes in the FiO2 have not adequately mitigated the alarm condition, and thus personal attention is needed. Our data suggest that alarms delay during A-FiO2 might be set at 60 seconds or longer in order to reduce false alarms. One study supports these considerations [15].

We evaluated one A-FiO2 system and the findings of this study should be generalized to other A-FiO2 systems cautiously. Alternative algorithms, that consider changes much less frequently than every second would not be expected to reduce episodes of shorter duration. Nevertheless, we also found that much of the reduction of total exposure was a result of reducing the duration of longer episodes, so this might not be clinically relevant. In contrast more frequent automatic adjustments might lead to more overshoot between extremes, which is a constant concern during manual control. Overshoot using the A-FiO2 algorithm that we studied has shown to be better than M-FiO2 [9, 16], but this needs to be evaluated in other A-FiO2 systems, regardless of their frequency of adjustment.

The primary limitation of our study is that it reflects a small population of infants and a single day of exposure. However, the results do seem to be consistent with other studies. The thresholds for extreme levels of SpO2 were prospectively defined, and have been shown to be associated with extreme levels of PaO2 [17]. Nevertheless, evaluation of other, less extreme, SpO2 thresholds might yield different conclusions and also impact outcomes. Nevertheless in a post hoc analysis we found the effects of using cut offs of < 87% and > 95% were similar, suggesting the findings would be insensitive to the exact cut off. Further we counted episodes > 98% even if the subject was temporarily not receiving supplemental oxygen. This would tend to increase the frequency of such episodes, but our analysis population, had a very limited amount of time without supplemental oxygen. There are also some other aspects of the study impacting the generalization of our work. First, our study reflects experience with a lower SpO2 target range resulting in a median SpO2 of 91%. A higher target range, if resulting in a higher median SpO2, would likely have a different profile. One study comparing two ranges (89-93% and 91-95% SpO2) found a shift to less hypoxemia and more hyperoxemia during both automated and manual FiO2control [8]. Studies also suggest that M-FiO2 control of SpO2 at higher target ranges is easier to manage than at lower target ranges [7, 18]. Finally, this study population was not large enough to explore relative differences associated with invasive and noninvasive ventilation or during periods of differing stability.

Conclusions

Our study characterizes the distribution of extreme hypoxemic and hyperoxemic episodes. Nearly all of these extreme episodes are of short duration. However, the relatively infrequent longer episodes were the main contributors to the total SpO2 extreme exposure. We demonstrated that A-FiO2 results in fewer episodes of all durations, but that a reduction in the infrequent longer episodes was the main factor in its improved effectiveness. We believe this information might be useful not only in refining SpO2 alarm practices, but also for refining and comparing the effectiveness of A-FiO2 control systems.