Introduction

Acquired brain injury in very preterm infants usually results from haemodynamic instability and cerebral dysautoregulation. Germinal matrix intra-ventricular haemorrhage (GM-IVH) and periventricular leucomalacia (PVL) account for the majority of cerebral injury in very preterm infants [1]. Neuro-monitoring strategies are limited in preterm infants. Cerebral oxygenation monitoring with near infra-red spectroscopy (NIRS) provides one potential method. Cerebral NIRS is non-invasive, portable, painless and easy to apply. This device can continuously measure regional cerebral oxygen saturations (rcSO2) and can indirectly estimate cerebral blood flow and cerebral tissue oxygen extraction (cTOE) [2]. rcSO2 or cTOE could be used to monitor the timing of GM-IVHs and PVL development in preterm infants [3, 4]. This information could be used to develop therapeutic interventions, which may prevent some causes of cerebral brain injury.

In the recent SafeboosC trial an adult NIRS sensor was used in preterm infants less than 28 weeks gestation. Normative regional cerebral oxygenation values of 55 to 85% [5] in the first 72 h of life were utilised based on previous unpublished data [6]. The Safeboosc trial showed that if cerebral saturation values were visible to the bedside clinician it was possible to obtain stable values between 55 and 85% in the first 72 h of life. More recent data from the group at Utrecht has been published. This group have pioneered the utilisation of NIRS in newborn infants [4, 7,8,9]. The data from this recent publication was obtained from a very large cohort of almost 1000 newborn infants, and predominantly utilised the adult sensor. This study provided normative data for various gestational age groups and showed gestation specific cerebral NIRS values over the first days of life [10]. They also evaluated the discrepancy between the adult sensor and the neonatal sensor in a small subgroup of infants, highlighting that neonatal sensors tend to overestimate rcSO2 values by 8–14% compared to the adult sensor [10]. Neonatal NIRS sensors, which are relatively smaller than standard adult sensors, are now available, are easy to apply and are used in preterm infants.

The aim of this study was to evaluate the temporal evolution of the rcSO2 over the first 48 h of life using a neonatal NIRS sensor. We also explored the relationship between rcSO2 and short-term neurologic outcome (STNO) in a cohort of preterm infants <32 weeks’ gestation. We hypothesised that there was a relationship between changes in postnatal transition rcSO2 values obtained in the first 48 h of life and STNO in preterm infants <32 weeks when using the NIRS device with the neonatal sensor.

Methods

This was a prospective observational study performed in a level three neonatal intensive care unit in Cork, Ireland. Data for this study was collected from two separate infant cohorts enroled between December 2009 and April 2015. All infants born <32 weeks of gestation were eligible for recruitment. Written informed parental consent was obtained prior to recruitment. Infants with known major chromosomal abnormalities or congenital abnormalities were excluded. The study was approved by the local Cork Research Ethics committee.

Materials

Cerebral oxygenation data was collected using the INVOS 5100 NIRS (Covidien, Mansfield, MA, USA) with a neonatal sensor, the OxyAlertTM NIRSensor (Covidien IIc, Mansfield, MA, USA). Recording began as soon after birth as feasible and continued for up to 48 h. rcSO2 data was stored on the INVOS device with a sampling period between 5 and 6 s and later exported for off-line analysis. Recordings of duration <6 h were excluded from further analysis.

Cranial ultrasound scans (CRUS)

CRUS were performed routinely as per unit protocol for all infants born <32 weeks of gestation. Scans were usually performed on day 1 to day 3, between day 10 and 14 and at ~1 month of age. All cranial ultrasound scans were formal clinical scans performed and reported by a Paediatric Radiologist who was not involved in the study and was blinded to NIRS data. The reports of the scans were used to define STNO. This was defined as the presence of GM-IVH, PVL, or death within 1 month of birth. Outcome was divided into three groups: severe, GM-IVH grade three or four, PVL, or death within 1 month after birth; moderate, GM-IVH grade one or two, absence of PVL, and survival by 1 month; and normal, absence of GM-IVH,PVL and survival to 1 month.

Analysis of rcSO2

A mixed-effects model was fitted to the rcSO2 data to describe the evolution of rcSO2 over time and to explore the effect of variables such as outcome and gestational age on rcSO2. The INVOS NIRS device measures rcSO2 within the range 15–95% where a value of 15% represents periods when skin-contact was lost with the sensor; accordingly, rcSO2 values of 15% were removed from further analysis. For the mixed model, the rcSO2 was low-pass filtered using a 1-h moving-average filter with 50% overlap to enable the model to capture the trend of rcSO2 over time.

A linear, quadratic, and cubic-time model was fitted to the data, with each infant’s rcSO2 recording representing a random effect. Gender, outcome, and gestational age (GA) groups were included as fixed effects. Outcome was defined in three groups, as previously described, and GA was dichotomised as either extremely premature, with GA < 28 weeks, or very premature, with GA ≥ 28 and <32 weeks. Model selection was achieved by a backwards elimination procedure, starting with all terms in the model. Terms were removed only when the model fit did not significantly improve, as tested by the log-likelihood ratio test, using a χ 2 distribution, with P < 0.05. Where appropriate, random effects of time, time-squared, and time-cubed were removed first, followed by the fixed effects of outcome, gender, GA groups, and the relevant time interactions. Lastly, time, time-squared, and time-cubed fixed-effects were removed if model fit did not significantly improve.

Two other features of the rcSO2, independent of the absolute rcSO2 values, were used to test possible association with outcome. These two features represented the total area below 55% and the total area above 85% and were previously developed to quantify the level of hypoxia (area < 55%) and level of hyperoxia (area > 85%) [11]. To incorporate expected differences in rcSO2 depending on GA, we increase these thresholds by 5% for the very preterm group, to 60 and 90%, while keeping the lower thresholds (55 and 85%) for the extremely preterm infants. These area measures were estimated on the raw rcSO2 signal. Because these unfiltered rcSO2 signals may contain artefacts, a 30 s collar was applied around the 15% rcSO2 data points which were then removed from further analysis. Area was measured using the trapezoidal method, a numerical approximation of integration. The area measures were normalised to produce a value between 0 and 1 and then log-transformed to offset the skewed (towards 0) nature of the measures [11]. The Kruskal–Wallis test was used to ascertain differences in the area values between the three outcome groups. If there was a significant difference, with P < 0.05, Mann–Whitney tests were used to test pair-wise associations. For this comparison the Holms procedure [12] was used to correct the P-value for multiple comparisons; both the adjusted P-value and the un-adjusted P-values were reported. We did not calculate cFTOE because of concerns about variability in synchronisation of our SpO2 data and rcSO2 data which would have made the calculation of this feature thus unreliable.

The rcSO2 data were processed using the Matlab software package (Release 2013a, The MathWorks, Inc., Natick, Massachusetts, US). The mixed-effect models were implemented in R (version 3.3.1, The R Foundation of Statistical Computing, http://www.r-project.org) using the lme4 package (version 1.1-10). Patient characteristics were compared in R using the Kruskal–Wallis and Fisher’s exact tests as appropriate.

Code availability: Matlab and R code used to generate the mixed-effect model and measures of hypoxia and hyperoxia is freely available (version 0.1.3, https://github.com/otoolej/rcSO2_temp_evolution). This code is distributed under a permission free software licence (BSD 3-clause licence).

Results

Patient characteristics

Figure 1 provides a breakdown of the patient enrolment. The principal reasons for non-enrolment were the unavailability of research staff and the lack of a device. A total of 134 preterm infants meeting recruitment criteria with NIRS recordings were considered for the analysis. Analysis was limited to the first 2 days of postnatal age (≤48 h). From this cohort, infants were excluded because the recording duration of NIRS was <6 h (n = 4), incorrect timing on NIRS machine clock (n = 5), and recording started after 48 h of age (n = 5). Thus, a total of 120 infants were included in the analysis. Median (IQR) GA was 29 (26.7–30.5) weeks, birth weight was 1.14 (0.88–1.45) kg, and there was 61 (50.8%) male infants. The majority of infants (83%) were born by C-section. Over two thirds of infants (69%) had a normal outcome, less than a quarter of infants (22%) had a moderate outcome, and 9% of infants had severe outcome. Patient characteristics are summarised in Table 1.

Fig. 1
figure 1

This Figure provides a breakdown of the patient enrolment for the study

Full size image
Table 1 Patient baseline characteristics represented either as median (IQR) or n (%)
Full size table

NIRS data analysis

The median (IQR) duration of NIRS recordings was 40.3 (24.7–46.6) hours and percentage of missing data, including removed artefacts, was 0.6% (0.1–4.3%). Figure 2, shows a plot of mean absolute rcSO2 values for these 2 GA groups and indicates that there is no obvious trend in rcSO2 over time, but that rcSO2 values are lower for extremely preterm infants comparative to very preterm infants. The following mixed-effect model analysis confirms these observations.

Fig. 2
figure 2

rcSO2 data summarised per recording (mean) over a 2-h epoch, showing that infants born extremely preterm (n = 41) have relatively lower ranges of rcSO2 values compared to the very preterm infants (n = 79). Circles and squares represent median values, thick bars represent inter-quartile range, and thin lines represent 95 percentiles

Full size image

Mixed effects model analysis

The mixed effects model included time, time-squared, time-cubed, and gestational age group as fixed effects and time, time-squared, and time-cubed as random effects. Coefficients of the fixed effects in Table 2 indicate that extremely preterm infants, comparative to the very preterm infants, have an rcSO2 that is 7.13% lower (CI: 4.18–10.07, P < 0.001). We found no significant GA group by time interactions thus indicating that the difference for GA groups is independent of time. Also, outcome and gender were not found to significantly improve model fit and were therefore not included in the mixed-effect model. Figure 3 plots the temporal evolution of the mixed effects model, showing a slow oscillation with a maximum deflection of ~5%. There does not appear to be any obvious rcSO2 increase or decrease in overall trend over the 48 h period.

Table 2 Estimates of fixed effects from a mixed-effects model of rcSO2 within the first 48 h after birth
Full size table
Fig. 3
figure 3

Mixed-effects model with cubic-time for rcSO2. Random effects represent the fitted trend for each infant’s rcSO2 and fixed effects represent an average trend over the groups

Full size image

Relationship with outcome

Table 3 shows the relationship between hypoxia and hyperoxia, as defined by the rcSO2 area < 55/60% and area > 85/90% measures, and the three outcome groups. Results indicate a difference in hypoxia for outcome (P = 0.009) but no difference in hyperoxia for outcome (P = 0.680). Pairwise comparison between the three different outcome groups show a greater degree of hypoxia for the moderate comparative to the normal group (P = 0.015) and moderate group comparative to the severe group (P = 0.042).

Table 3 Features of the rcSO2 relating to grades of outcome: normal, n = 83; moderate: n = 26; severe: n = 11
Full size table

Discussion

Our study describes trends in rcSO2 values in a large cohort of preterm infants born before 32 weeks of gestation monitored with the neonatal INVOS sensor over the first 48 h of life. The results show that cerebral oxygenation values are relatively stable over the 48 h period. Although the raw rcSO2 changes over time for each infant, over all infants we find no clear temporal trends in rcSO2 for either gestational age group (Fig. 3). Extreme preterm (<28 weeks) infants had a relatively lower rcSO2 values compared to those born very preterm (28–32 weeks) throughout the 48 h (Figs. 2, 3 and Table 2). The mixed model estimates this significant difference between groups as 7.13% (95% CI: 4.18–10.07, P < 0.001), and rcSO2 for both groups follow the same cubic trajectory over time. The intercept value, which represents the average rcSO2 values at time zero at the beginning of recording in both the gestational age groups, is 75.8% for the extreme preterm and 82.9% for the very preterm. This reflects higher rcSO2 average values for the neonatal sensor than existing data from the adult probe [13] and these values need to be considered when designing future trials of cerebral NIRS in preterm infants. The work of Alderstein and colleagues highlighted this difference in a 16 preterm infants with a mean gestational age of 30 weeks who had both the neonatal and the adult sensor in place simultaneously, showing a difference of about 10% between the devices [10]. Greisen and colleagues recently evaluated various devices on a phantom model and highlighted various inter-device differences [14,15,16].

Our results indicate that there was a statistically significant association between cerebral hypoxia and outcome (P = 0.009) but no difference between cerebral hyperoxia and outcome (P = 0.680). Though the association between cerebral hypoxia and poor neurologic outcome may not be surprising, this is important to note, as this suggests that it may be possible to use NIRS as a monitoring tool to predict and ultimately prevent brain injury in this group of preterm infants. In an exploratory analysis the Safeboosc group noted a higher incidence of brain injury for those infants in the lowest quartile for cerebral hypoxia [17].

Pairwise comparisons showed an increase in hypoxia for the group with a grade 1 or 2 haemorrhage compared to both the normal (P = 0.015) and severe groups (P = 0.042). This was an interesting finding because we expected the severe groups to have a greater degree of hypoxia when compared to the normal group. They were more immature, of a lower birth weight and had more respiratory support. We suspect that these infants were more likely to receive more interventions, primarily in the form of respiratory management. The SafeboosC trial highlighted that adjustment in fractionated oxygen support was the most common intervention performed in the sickest preterm infants [5]. However we can only postulate as we did not measure the number of interventions within groups. It is important to note that our study was observational and the NIRS data was collected but not acted upon by the bedside clinician. It is conceivable that NIRS values were similar as significantly increased interventions may have occurred in the more immature group. However these questions can only be answered with appropriately designed trials utilising appropriately chosen target ranges, incorporating quantitative measures of degree of hypoxia, or investigating different signal features which may be better predictors of risk of brain injury [18].

A recent study [4] assessing neurologic outcome and NIRS-measured rcSO2 showed that infants who developed GM-IVH had higher rcSO2 and lower cFTOE values before the development of GM-IVH, suggesting increased perfusion [4]. These findings contrasted with another [19], which reported that patients with GM-IVH tended to have lower left ventricular output and cerebral rSO2 and higher cFTOE during the first 12 h. In the present study, we looked at the cumulative rcSO2 data over the first 48 h, but we could not determine the exact timing of brain injury, as frequent ultrasound scanning in the first days of life was not routinely performed. A recent study that assessed long term neurologic outcome following NIRS monitoring found that rcSO2 values monitored during the first 2 weeks after birth had an association with longterm neurodevelopmental outcome at 2–3 years, specifically high and low rcSO2 on day one of life [20]. Identifying the specific timing of brain injury could potentially allow clinicians to understand in more detail the changes that occur to the infant’s cerebral haemodynamics, including rcSO2 values, prior to or during the event, and could in turn help develop specific treatment or preventative measures. Future work in this area is warranted.

There are a number of limitations. Filtering rcSO2 values for the mixed-effect model suppresses faster oscillations in the signal, thus there might be discriminatory information in these faster oscillations that we have not examined here. The neonatal sensor has known limitations in detecting hyperoxia in preterm infants, which we also confirmed in the present study. In our study the majority of infants who developed IVHs had mild to moderate grades (grade less than II). We categorised these infants as having poor STNO, because there is evidence suggesting that the presence of grade I–II GM-IVH, even with no documented white mater injury or late ultrasound abnormalities, may be associated with poor neurodevelopment outcome in extremely preterm infants [21]. Even though we had a larger cohort of infants monitored with the neonatal sensor compared to previous studies, the overall number of infants with a poor STNO was low.

Conclusion

Immature preterm infants have relatively lower rcSO2 values compared with their more mature counterparts but rcSO2 values are relatively stable over the first 48 h of life, showing no obvious increase or decrease in overall trends There was no statistical difference in the absolute values or temporal trajectories of rcSO2 between the three outcome groups. It remains difficult to quantify hyperoxia for infants >28 weeks GA using the neonatal sensor. Further trials are warranted to determine if actively intervening to maintain rcSO2 within certain limits results in improved outcome.