Preterm infants off positive pressure respiratory support have a higher incidence of occult cerebral hypoxia

Abstract

Background:

Impaired cerebral oxygen delivery is a known risk factor for preterm brain injury and may not be apparent using conventional monitoring. Our objective was to utilize cerebral near-infrared spectroscopy (NIRS) to quantify occult cerebral hypoxia across respiratory support modes in preterm infants.

Study Design:

In this prospective, longitudinal, observational study, infants ≤32 weeks gestation underwent serial pulse oximetry (SpO₂) and cerebral NIRS monitoring (4–6 hours per session) following a standardized recording schedule (daily for two weeks, every other day for two weeks, then weekly until 35 weeks corrected gestational age).

Four calculations were made: median cerebral saturation, median cerebral hypoxia burden (proportion of NIRS samples below hypoxia threshold, <67%), median systemic saturation, and median systemic hypoxia burden (proportion of SpO₂ samples below desaturation threshold, <85%). During each recording session, respiratory support mode was noted (room air (RA), low-flow nasal cannula (LFNC), high-flow nasal cannula, non-invasive positive pressure ventilation (NIPPV), CPAP, and invasive ventilation).

Results:

1013 recording sessions were made from 174 infants with a median length of 6.9 hours. Although systemic (SpO₂) hypoxia burden was significantly greater for infants on the highest respiratory support (invasive and NIPPV), the cerebral hypoxia burden was significantly greater during recording sessions made on the lowest respiratory support (8% for RA; 29% for LFNC).

Conclusion:

Premature infants on the highest levels of respiratory support have less cerebral hypoxia than those on lower respiratory support. These results raise concern about unrecognized cerebral hypoxia during lower acuity periods of NICU hospitalization and adverse outcomes.

Introduction

Neonatal cerebral hypoxia has been linked to preterm brain injury including intraventricular hemorrhage (IVH) and white matter injury (WMI) [1–5]. Pulse oximetry (SpO₂) is often used in conjunction with blood pressure and heart rate monitoring to evaluate the cardiorespiratory status of the infant and to guide clinical management including choice of respiratory support, the amount of supplemental oxygen, inotropes, and blood transfusion. However, there are significant limitations to conventional vital sign monitoring. For example, the optimal blood pressure in preterm infants remains unclear [6–10], pulse oximetry may underestimate oxygen saturation [11–13], and some infants may have significant cerebral hypoxia despite otherwise unremarkable vital signs [14,15].

Continuous pulse oximetry revolutionized neonatal monitoring, providing a real-time estimate of arterial oxygen saturation (SpO₂) far superior to visual observation of cyanosis [16,17]. Pulse oximetry was the backbone of pivotal clinical studies and meta-analyses including STOP-ROP [18], BOOST-II [19], SUPPORT [20], and NEOPROM MA [21] where targeted oxygen saturation limits were explored and refined to improve the odds of intact survival while limiting the adverse outcome of retinopathy of prematurity. Near-infrared spectroscopy (NIRS) measures tissue oxygenation using similar optical principles as pulse oximetry, and provides a measure of tissue oxygenation expressed as a ratio of oxygenated to total hemoglobin in the tissue underlying the sensor [22,23]. NIRS correlates with cardiac output [24,25] and serves as a more accurate predictor of vital organ failure as reflected by low tissue oxygenation values [26,27].

Cerebral NIRS monitoring has been extensively studied and utilized in preterm infants during the first week of life. Previous studies have described reference values [28], changes associated with brain injury such as intraventricular hemorrhage (IVH) [2,29–31], high severity of illness [32], and a hemodynamically significant PDA [33,34]. These observations led to phase II [35] and phase III [36] interventional clinical trials which utilized clinical NIRS monitoring in conjunction with a standardized treatment plan to reduce severe brain injury and/or death. While the preliminary results of interventional NIRS trials are promising, all observation and intervention studies have been focused on the 7-day period immediately following birth, despite an expected length of stay for an infant born before 28 weeks of more than 100 days.

The need for respiratory support in preterm infants is nearly universal and is required to treat diverse pathologic states that include respiratory distress syndrome and chronic lung disease, the maintenance of functional residual capacity to prevent atelectasis, or for treatment of apnea of prematurity. Respiratory failure is a defining medical diagnosis for most preterm infants. Respiratory support is principally managed through blood gas analysis, chest radiographs, physical examination (e.g., work of breathing), and non-specific measures of oxygenation (pulse oximetry). However, pulse oximetry has known limitations including motion artifact and disparities in accuracy, especially for infants on supplemental oxygen and of non-White race [11,37]. The pulse oximeter provides information only on oxygen delivery, not consumption, and thus is unable to identify impaired oxygen utilization at an organ-specific level (e.g., the brain). The lack of cerebral monitoring outside of the first days following birth and the pulmonary focus of respiratory monitoring leaves a potential gap of unrecognized poor cerebral oxygenation; previous reports have demonstrated the presence of critical cerebral desaturation, even in the presence of otherwise normal vital signs [14,15].

In this project, we performed frequent, longitudinal cerebral NIRS and continuous pulse oximetry monitoring of preterm infants to investigate the burden of systemic desaturation and cerebral hypoxia associated with different respiratory support modes, and to quantify the frequency of occult cerebral hypoxia not detected by pulse oximetry.

Methods

Study population and clinical practices

Infants were recruited from the neonatal intensive care unit (NICU) at Saint Louis Children’s Hospital, a 150-bed level IV NICU serving urban, suburban, and rural populations primarily from Missouri, Illinois, and Arkansas between 2012 and 2020. Infants were eligible if born ≤ 32 completed weeks of gestation and recruited within the first week of life. Infants were excluded if there was a known congenital anomaly or hemoglobinopathy (homozygous for HbS or HbC, α-thalassemia), informed consent could not be obtained, or were not receiving active life support measures. The study was reviewed and approved by the Institutional Review Board at Washington University. Written informed consent was obtained from the parents prior to the initiation of any study procedures.

Infants in this study received standard NICU care; NIRS monitoring was not used in clinical decision-making, and providers were blinded to quantitative NIRS metrics. Institutional practice included targeted oxygen saturation guidelines with a goal SpO₂ range between 90 and 95%. The SpO₂ target range was broadened to 90–100% once infants reached 35 weeks corrected gestational age or no longer required any form of respiratory support, whichever occurred sooner.

Data capture

After enrollment in the study, all neonates underwent cerebral NIRS monitoring using an FDA-approved NIRS monitor (ForeSight Elite, Edwards LifeSciences, Irvine CA) on a pre-determined schedule, with a goal of daily monitoring for the first two weeks after birth, every other day for the second two weeks after birth, and then weekly monitoring until reaching 35 weeks corrected age. Most recordings were made during the daytime, usually starting in the morning, and ending in the afternoon. For each monitoring session, a member of the study team placed a non-adhesive optode embedded in a soft, flexible neoprene band (ForeSight sensor kit small)on the fronto-parietal scalp which was secured using a Velcro clasp. The positioning of the sensor was altered slightly between recording sessions to avoid skin irritation or injury. Although the goal monitoring time was 4–6 hours per session, recordings were sometimes extended or shortened to coincide with scheduled patient care. In some cases, monitoring was not performed due to patient instability; the recording schedule was resumed on the next possible day. For each monitoring session, time synchronized NIRS and pulse oximetry data (Nellcor OxiMax, Medtronic, Minneapolis, MN) were captured with a sampling rate of 0.5 Hz. Recording files were archived and processed after hospital discharge.

Imaging

Following our institutional protocol, all infants underwent standard head ultrasound screening with at least one scan in the first 3 days after birth, a second scan between 7 and 10 days after birth, and one scan at 30 days of life. Posterior fossa imaging was performed using a mastoid window to identify cerebellar hemorrhage. All surviving infants underwent a non-sedated, non-contrast MRI performed between 36 and 42 weeks corrected gestational age. For those infants who were too unstable to safely travel to MRI, a bedside term-equivalent head ultrasound was performed. Given the mixed imaging modalities and the known challenge of identifying small lesions on head ultrasound, injury was identified at a macroscopic scale including intraventricular hemorrhage (IVH) graded according to the Papile scale [38], cerebellar hemorrhage (CH, present or absent), and white matter injury (WMI, present or absent).

Demographic data, clinical outcomes, and respiratory support

A standard set of demographic factors was collected for all study infants including gestational age at birth, birth weight, sex, and race (as reported on birth certificate). Clinical outcome data were collected including mortality before hospital discharge, BPD (defined as need for supplemental oxygen after 36 weeks corrected age), culture positive sepsis, stage IIIB necrotizing enterocolitis (NEC), and severe retinopathy of prematurity (ROP, defined by need for laser surgery or treatment with bevacizumab).

The mode of respiratory support and the average fraction of inspired oxygen (FiO₂) during each recording session were extracted from electronic medical records. Modes were classified as:

• Invasive (inclusive of conventional mechanical ventilation, high-frequency oscillatory ventilation [HFOV], and invasive neurally-adjusted ventilatory assist [NAVA])

• Non-invasive positive pressure ventilation (NIPPV, including non-invasive NAVA)

• Continuous positive airway pressure (CPAP, including bubble CPAP and ventilator delivered CPAP)

• High flow nasal cannula (HFNC, defined as ≥ 2 LPM)

• Low-flow nasal cannula (LFNC, defined as < 2 LPM)

• Room air (RA)

Data cleaning and calculation of saturation metrics

Each recording underwent simple error correction; measurements of 0 or missing values were replaced with NaN (not-a-number) to permit omission from statistical calculations. Although the goal recording length was 4–6 hours, in a previous analysis [39] we empirically determined that a minimum threshold of 1 hour is needed for accurate assessment of mean cerebral saturation. Thus, recordings were excluded if the error-corrected recording was less than 60 minutes in duration.

The mean cerebral saturation (cSat) and mean systemic saturation (SpO₂) were calculated for each recording as the arithmetic mean using the nanmean function in MATLAB. Cerebral and systemic hypoxia burdens were both calculated in the same fashion, defined as the proportion of the recording which fell below the hypoxia threshold. In group analysis, the median values of each calculated measure were used, as assumptions about the normality, variance, independence, and outliers are not met. For this project, the cerebral hypoxia threshold was defined as cSat <67%, consistent with the threshold used for ForeSight Elite monitors in the SafeBoosC trials [35,36,40]. Because our institutional practice is to maintain systemic saturations between 90 and 95% for preterm infants below 35 weeks corrected age, the systemic hypoxia threshold was defined as SpO₂ <85% and did not include mild, transient events. All calculations were made using a script developed in MATLAB (MathWorks, Natick MA).

Statistical approach

Descriptive statistics of demographic and clinical factors were calculated and reported using the appropriate statistical measure for variable type. As the recording and physiologic data were recorded in a real-world environment, normal distribution was not expected. The overall distribution of each physiologic factor was assessed using the median and the 25^th/75^th quantile, grouped by respiratory support subgroups.

Given the unbalanced, repeated measure design, linear-mixed-effects modeling which accommodates repeated measure, missing, and unbalanced data types was used to investigate the relationships between physiologic outcome measures (median SpO₂, median cSat, cerebral and systemic hypoxia burden) and respiratory support mode. CPAP was selected as the reference group. As IVH has been associated with lower cerebral saturations [2,29,41], IVH was included in the model. Within-subject variability was addressed by entering subject ID as a random effect in the model. All modeling was performed using the nlme package within R version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Demographics and clinical characteristics

In this study, 174 infants were included with mean gestational age at birth of 26.3 ± 2.1 weeks, a mean birth weight of 951 ± 306 grams, 52% female, 58% diagnosed with BPD, and 11% died during NICU hospitalization (Table 1). Other than exclusionary factors, the demographic characteristics of the recruited infants matched those of VLBW NICU population at St. Louis Children’s Hospital. A CONSORT diagram of patient enrollment is shown in Figure 1.

Table 1.

Patient characteristics

	N=174
Gestational age, mean (SD), weeks	26.3 (2.1)
Birth weight, mean (SD), grams	951 (306)
Female sex, n (%)	90 (49%)
Race
White, n (%)	99 (54)
Black, n (%)	82 (44)
Other, n (%)	4 (2)
Antenatal steroids, n (%)	137 (74)
Died, n (%)	20 (11)
BPD, n (%)	101 (55)
Severe ROP, n (%)	24 (13)
Intraventricular hemorrhage
Any grade, n (%)	54 (29)
Grade III/IV, n (%)	16 (9)
Cerebellar hemorrhage, n (%)	16 (9)
White matter injury, n (%)	30 (16)

Figure 1 (Online only):

CONSORT flow diagram of study screening, exclusion, and consent.

Recording characteristics and data quality

1015 recordings were made, with only two excluded for length less than one hour after error correction. The remaining 1013 recordings had a median length of 6.9 hours with an interquartile range of 3.3–8.8 hours. Error correction removed a median of 1% of each recording, with an interquartile range of 0–11% of the recording removed. Most recordings (884/1013, 87%) were made while infants received some form of respiratory support, with invasive ventilation and CPAP being the most common (26 and 25% of recordings, respectively). Infants received respiratory support with supplemental oxygen (defined as FiO₂ greater than 21%) during most recordings (628/1013, 62%), while 256/1013 (25%) were made on infants receiving respiratory support but without supplemental oxygen (FiO₂ = 21%), and 129/1013 (13%) were made on infants on room air (Table 2).

Table 2.

Recording characteristics

	Number of recordings (%), n=1013	FiO2 %, mean (SD)
Invasive	266 (26)	35.6 (16.4)
NIPPV	204 (20)	26.8 (8.4)
CPAP, bubble CPAP	255 (25)	24.7 (5.9)
High-flow NC (≥ 2 LPM)	118 (12)	24.0 (3.8)
Low-flow NC (< 2 LPM)	41 (4)	28.7 (17.5)
No respiratory support	129 (13)	21 (0)

Cerebral and systemic saturation data

The median SpO₂ varied somewhat by respiratory support mode but fell almost entirely within unit guidelines. Infants receiving invasive ventilation and NIPPV had statistically lower SpO₂ values in mixed-model regression (−2.0% p<0.01; −1.2% p<0.01, respectively) than those receiving CPAP (Table 3a). The median systemic hypoxia burden (SpO₂ <85%) was minimal (0–4% of each recording) across all respiratory support modes but was statistically greater for infants receiving invasive and NIPPV support (+3.1% p<−0.01; +1.4% p=0.04, respectively) than CPAP (Table 3a).

Table 3a.

Systemic saturation characteristics

	Median SpO2, (IQR)	Changea	P valuea	Median systemic hypoxia burden, (IQR)	Changea	P valuea
Invasive	92 (91, 94)	−1.9%	<0.01	4% (1–10)	+4.1%	<0.01
NIPPV	94 (92, 96)	−1.0%	<0.01	2% (0–5)	+1.7%	<0.01
CPAP, bubble CPAP	95 (93, 97)	ref		1% (0–2)	ref
High-flow NC (≥ 2 LPM)	94 (93, 96)	+0.1%	0.95	1% (0–4)	0	0.94
Low-flow NC (< 2 LPM)	95 (93, 97)	−0.1%	0.83	2% (0–5)	+1%	0.46
No respiratory support	96 (95, 98)	−0.24%	0.39	0% (0–1)	+1%	0.18

Footnotes: IQR=interquartile range.

^aChange and pvalues represent estimated departure from reference (CPAP in all cases) in mixed-model regression analysis, using subject ID as random effect and accounting for repeated measures.

Median cerebral saturation values ranged between 79% (CPAP) and 68% (LFNC). All were above the pre-defined hypoxia threshold of 67%. Using CPAP as the reference group, NIRS detected significantly lower cerebral saturation with room air and LFNC respiratory support in mixed-model regression (−4.5%, and −11.3% respectively) (Table 3b). In contrast to systemic hypoxia burden, cerebral hypoxia burden varied by respiratory support type between 29% (LFNC) and 1% (CPAP) and was significantly greater for LFNC and RA support types (+21.9% p<0.01; +6.9% p=0.01, respectively) (Table 3b).

Table 3b.

Cerebral saturation characteristics

	Median cSat, (IQR)	Changea	P valuea	Median cerebral hypoxia burden, (IQR)	Changea	P valuea
Invasive	73 (67, 78)	−2.9%	0.04	4% (0–29)	+3.9%	0.10
NIPPV	74 (68, 79)	−4.4%	0.01	5% (0–31)	+8.5%	0.16
CPAP, bubble CPAP	79 (73, 83)	ref		1% (0–12)	ref
High-flow NC (≥ 2 LPM)	77 (69, 82)	−1.7%	0.19	2% (0–15)	+1.6%	0.66
Low-flow NC (< 2 LPM)	68 (59, 81)	−8.8%	<0.01	29% (0–56)	+17.8%	<0.01
No respiratory support	73 (62, 82)	−6.0%	<0.01	8% (0–45)	+10.2%	<0.01

Footnotes: IQR=interquartile range.

^aChange and pvalues represent estimated departure from reference (CPAP in all cases) in mixed-model regression analysis, using subject ID as random effect and accounting for repeated measures.

The distribution and range of cerebral saturation and cerebral hypoxia burden are shown in Figure 2. Kernel density plots of all physiologic variables are shown in Figure 3.

Figure 2:

Box/violin plots are shown for cerebral saturation (left) and cerebral hypoxia burden (right). Each dot represents the median value for one recording, divided by respiratory support groups. The group median value is shown with a large circle, and the distribution of values is indicated by the box edges (IQR) and violin edges (frequency distribution).

Figure 3 (Online only):

Kernel density plots for each of the four physiologic measures are shown including SpO₂ (A), cSat (B), systemic hypoxia burden (C), and cerebral hypoxia burden (D). Kernel density is a type of smoothed histogram which graphically displays the probability of a given variable to have a certain value. The peak of each curve represents the value of the variable which is most likely to occur in a random sample. As with histograms, a sharp tall peak suggests that a particular value is very likely to occur, while a lower, flatter peak suggests a broader range of values with equal likelihood of occurrence. Group distribution is labeled by color with CPAP in orange, HHNC in yellow, invasive ventilation in green, LFNC in turquoise, and room air in magenta.

Imaging results

All 174 infants included in the study had at least one head ultrasound performed in the first week of life. For term-equivalent imaging, 119/174 (68%) of infants had MRIs, 27 (16%) had head ultrasounds, 20 (11%) died before term-equivalent age, and 8 (5%) did not have term-equivalent imaging. Overall, 31% of infants were noted to have IVH of any grade, 9% of infants had high-grade IVH, 9% had cerebellar hemorrhage, and 17% had white matter injury (Table 1). As expected, the presence of IVH was independently associated with lower cerebral saturations and an increased cerebral hypoxia burden, with mixed-model analysis indicating an expected decrease of 6.2% in cerebral saturation and an increase of 10.3% in the cerebral hypoxia burden.

Discussion

Preterm infants on the lowest form of respiratory support (LFNC) or with no respiratory support (RA) have lower, but within reference range, cerebral saturations measured by NIRS, but markedly longer periods of cerebral hypoxia, as much as 29% of the recorded time. Importantly, this hypoxia is isolated to the brain, with most (75%) of concurrent SpO₂ values falling within targeted oxygen saturation guidelines and very little time at critically low systemic oxygen saturations, typically 0–4% of the observed time during intermittent, longitudinal monitoring. These results raise concern about the frequency of occult cerebral hypoxia during lower acuity periods of NICU hospitalization and highlight a notable limitation of pulse oximetry. Given the frequent occurrence of neurodevelopmental delay in the absence of overt radiographic injury and lack of NICU co-morbidities [42], alternative monitoring strategies may need to be considered.

Many studies have examined the potential use of cerebral NIRS monitoring in the preterm infant population. Reference values for the first three days [28] and first week [43] following birth have been identified. Separate reports by El-Dib [44] and Alderliesten [29] have identified associations between early cerebral hyperoxia followed by subsequent desaturation and germinal matrix or intraventricular hemorrhage. A systematic review conducted by Kalteren et al. suggested that NIRS-based cerebral tissue oxygen saturation assessment may help to identify anemia and the need for red blood cell transfusion, potentially improving neurologic outcomes in preterm infants [45]. Two randomized trials of NIRS-guided vs. standard of care treatment for VLBW infants in the first three days following birth (SafeBoosC-II and III) have been published, demonstrating a marked reduction in hyper- and hypoxia in the NIRS-guided arm but no difference in mortality or severe brain injury by 36 weeks corrected gestational age [1,40]. In this context, the results of the present study are even more intriguing. While NIRS may not provide additional clinical value during the early postnatal transition, when there is already a high intensity of monitoring, most preterm infants have limited laboratory and physiologic monitoring in the late stages of the NICU stay. The potential value of NIRS monitoring to uncover “hidden” pathology during this quiescent period may be much greater.

For this analysis, we utilized a cerebral hypoxia threshold taken from the SafeBoosC clinical trials, which in turn is based on the Alderliesten reference values, adjusted for device and sensor differences [46,47]. As reference standards outside of the first week of life are not available, the use of this threshold represents the best available data. Other thresholds have been proposed; in the Early NIRS Study, Chock et al., identified a cSat of 50% (corresponding to a ForeSight saturation of 60%) as the optimal threshold for identification of clinically meaningful cerebral desaturation [43] and matches an earlier report by Verhagen [48]. Another recent study compared two threshold measures, cSat <60% (ForeSight monitor) for more than 5 minutes and a moving threshold over the first 7 days after birth, primarily between 65 and 70% [49]. While the selection of a threshold is important when building predictive models, as it influences sensitivity and specificity, it does not alter the proportional relationships of hypoxia burden, just absolute numbers.

The majority of cerebral NIRS research in preterm infants has been conducted within the first week of life when infants have a high acuity of illness, typically matched to more intensive monitoring and interventions. In contrast, infants near the end of their NICU hospitalization are on little or no respiratory support, have comparatively less vital sign monitoring, and minimal routine laboratory testing. Despite lower clinical acuity, these recovering infants may have frequent apnea-bradycardia-desaturation events due to immature control of breathing which often coincides with the nadir of anemia of prematurity. This confluence of risk factors may potentiate the risk for hypoxic ischemia. Although several transfusion trials [50–52] have failed to demonstrate differences in death or disability between high and low transfusion targets, a recently published secondary study of the TOP trial demonstrated significantly lower cerebral saturations for infants in the lower transfusion threshold group [53]. As noted previously, the striking paucity of longitudinal cerebral saturation data in preterm infants makes it difficult to separate changes due to differences in normal development from pathologic disease processes in which interventions might change outcomes. The data reported here raise the question of the contribution of occult cerebral hypoxia in the late convalescent course to adverse outcomes in preterm infants. Taken together, these data support the need for future study of the impact of late cerebral saturations on neuroimaging, neurodevelopment, and other adverse outcomes. These results would inform clinical care to ensure that infants are receiving the level of intervention, inclusive of transfusion and respiratory support, needed to maintain adequate cerebral perfusion.

Although we found lower cerebral saturations and an increased cerebral hypoxia burden for those infants with IVH, it is important to note that the identified differences in cerebral hypoxia burden by respiratory support type were independent of IVH status. IVH has been linked to a lower cerebral saturation not only in the period following birth [29] but also for a period of weeks or months later [2,41]. The reason for persistently low cerebral oxygenation after IVH is not clear but may result from increased oxygen extraction [54], decreased cerebral blood flow [55,56], or possibly interference from deoxygenated blood within the ventricles themselves. Regardless of the exact mechanism, occult cerebral hypoxia during periods of lower or no respiratory support may serve as a “second hit” and further increase the risk of overt or microstructural injury in the form of white matter injury or decreased brain growth. Further research using advanced imaging techniques is needed.

There are several important limitations to this study. First, while the optimal strategy from a data collection standpoint would be uninterrupted, continuous monitoring of the neonate from birth to discharge, the thin, fragile skin of a premature infant would not tolerate NIRS monitoring on the order of weeks or months without adverse effect. The shorter recording time frames used in this study sought to balance the importance of data collection with avoidance of injury. Monitoring schedules were also adjusted, skipped, or abbreviated based on the clinical factors outside of study control. For example, patients who were isolated for COVID exposure, who had undergone major procedures, or who were too clinically unstable for handling were not monitored during these significant periods. However, as noted earlier, as little as a single hour of high-quality NIRS data closely approximates the daily mean cerebral saturations, and the large sample size of the recording library is likely representative of the true trajectory of cerebral saturations.

A second important limitation is the use of a single NIRS probe as a proxy for whole brain saturation. The neonatal probe used in this study provides a penetration depth of approximately 25 mm in a regional field of view. Cerebral perfusion is not monolithic, and it is possible the observed saturations are higher or lower than other regions of the brain. Multi-sensor cerebral oxygenation devices have demonstrated differences in cerebral autoregulation between the left and right hemisphere in older children undergoing ECMO [57], although it is unclear that this difference is clinically significant. Similar study in preterm infants found very little regional variation except in the setting of the most severe cases of intracranial hemorrhage [58]. Additionally, recent data [59] suggests that head circumference may influence NIRS measurements, with lower saturations noted in infants with smaller head circumferences as the field of view increasingly captures non-oxygenated ventricular space. Decreased respiratory support is anticipated to be highly correlated with advancing postmenstrual age, pulmonary maturity, and increased head circumference, however we found very little difference in median cerebral saturation between the highest and lowest forms of support. The differences in observed cerebral hypoxia burden are driven by intermittent hypoxic events rather than persistently lower saturation and are not likely to have a basis in anatomic variance. Nevertheless, head circumference is an important factor which should be considered in future studies.

It is technically and logistically challenging to perform repeated NIRS monitoring sessions in the VLBW population. Although the recording plan was ambitious, not all recordings could be made as originally schedule due to patient death, clinical instability, or staff/equipment availability. Most missing recordings occurred due to patient death or instability, which predominantly occurred early in the hospital course when infants were on higher levels of respiratory support. Although these higher modes of support were well represented in the overall distribution (Table 2), it is important to validate these estimates of cerebral saturation and hypoxia burden in future cohorts. As NIRS sensor technology improves, becoming smaller and less obtrusive, it may allow easier access to prolonged monitoring, even in unstable patients with poor tolerance to handling.

Finally, the recording window of this study went through 35 weeks corrected gestational age, which meant that fewer recordings were made on lower forms of respiratory support. Future studies should include the entire NICU course to capture greater numbers of these recordings. Data were analyzed based on mode of respiratory support at the time of the recording and not corrected for gestational age. Although there is a strong correlation between increasing corrected gestational age and decreasing levels of respiratory support, this is not universally true. The independent effect of postnatal age should be considered in future studies. Long term neurodevelopmental follow up will be needed to determine if the differences in cerebral saturation impact patient outcomes.

Conclusion

The results from this study raise concerns about the risk of occult cerebral hypoxia in infants receiving lower levels of respiratory support. Infants on higher levels of respiratory support are exposed to higher effective FiO₂. However, our results suggest they have significantly less cerebral hypoxia compared to periods on lower respiratory support. Pulse oximeters are an important component of NICU monitoring but provide information only on arterial oxygen saturation—oxygen delivery, not consumption. At the level of a specific tissue (e.g., the brain), NIRS oximetry provides information about the balance between oxygen consumption and delivery. This single measure reflects the many components which might influence this balance including anemia, brain activity (for growth and function), the effects of medications, and cardiorespiratory support. These findings suggest an urgent need for a future clinical trial incorporating NIRS monitoring into longitudinal clinical decision making about respiratory support late in the NICU course.

Data sharing statement:

Patient privacy restrictions limit data sharing.