Main

Despite advances in neonatal intensive care that have led to a decline in morbidity, preterm birth is still associated with neurological sequelae (1). Brain injury in preterm infants is often caused by disturbances in cerebral blood flow (CBF) and oxygenation (2,3,4). Evidence is accumulating that monitoring blood pressure alone is not enough to ensure adequate (cerebral) perfusion and oxygenation (5,6).

Near-infrared spectroscopy (NIRS) is a technique that can be used to monitor regional cerebral oxygen saturation (rScO2), being both a measure of cerebral oxygenation as well as a surrogate of CBF. NIRS monitoring can be applied for prolonged periods of time, even in the most vulnerable infants (7). It uses multiple wavelengths of NIR light and relies on the distinct absorption spectra of oxygenated (O2Hb) and deoxygenated (HHb) hemoglobin to calculate relative concentrations of O2Hb and HHb, which are then used to calculate the rScO2 (O2Hb/(O2Hb + HHb)). Where pulse-oximetry only measures the oxygen saturation in arterial blood (SaO2), NIRS makes no distinction between different (cerebral) blood volume compartments; therefore, the rScO2 represents the oxygen saturation in a mixed arterial–capillary–venous compartment in an approximate 20:5:75 distribution (8).

NIRS is increasingly being used as a trend monitor of cerebral oxygen supply in neonates admitted to the neonatal intensive care unit (NICU). Readily interpretable reference values could provide another way of using NIRS in neonates by identifying neonates at risk. In other words, to identify neonates whose rScO2 resides at the outskirts (high or low) of what is considered “normal.” Furthermore, reliable reference values could benefit NIRS research by suggesting thresholds that should be explored in relation to interventions and (neurodevelopmental) outcome. However, current literature is quite heterogeneous, with different age groups, small sample sizes, different onsets and durations of measurement, and the use of different devices and sensors (9,10,11,12,13,14,15). Therefore, the aim of this study was to construct gestational age (GA)–specific reference curves during the first 72 h of life for rScO2 and its derived cerebral fractional tissue oxygen extraction (cFTOE (SaO2 − rScO2)/SaO2) in a large group of neonates who were measured with a small adult NIRS sensor (SomaSensor SAFB-SM using INVOS 4100 or 5100c monitors) (16). The second aim was to provide a conversion model to convert rScO2 values obtained by a neonatal sensor (OxyAlert CNN cerebral NIRsensor; Covidien) to the adult sensor equivalent.

Results

Out of the 1,059 participating infants, 41 were excluded because of technical problems during data collection (e.g., data corruption, missing cable connections, or electrical interference). An additional 19 infants were excluded for having cardiac malformations (n = 8), chromosomal or severe genetic abnormalities (n = 6), or severe congenital malformations (n = 5). Table 1 summarizes the clinical characteristics of the study population.

Table 1 Baseline characteristics of the study population

A total of 59,135 1-h periods (median: 873; interquartile range: 817–905 per 1-h period) were available for analysis. Although 1-h periods were used for modeling, Figure 1 displays the raw rScO2 and cFTOE data in 6-h averages for the first 24 h and 12-h averages for the 48 h thereafter to yield an easily interpretable figure.

Figure 1
figure 1

Boxplots of the raw data are displayed for four gestational age groups: white boxes indicate 24–25 wk, light gray boxes indicate 26–27 wk, dark gray boxes indicate 28–29 wk, and black boxes indicate 30–31 wk for (a) regional cerebral oxygen saturation (rScO2) and (b) cerebral fractional tissue oxygen extraction (cFTOE). Data are displayed in 6-h periods for 0–24 h after birth and in 12-h periods for 24–72 h after birth.

PowerPoint slide

Figures 2 and 3 display the main results: the rScO2 and cFTOE reference curves for four different GA groups. Depending on GA and postnatal age (PA), the mean rScO2 during the first 72 h of life ranges from 62 to 71%, with a positive association between rScO2 and GA. A single SD in rScO2 is ~7%, which provides a ±2 SD bandwidth of ~30%. Likewise, mean cFTOE ranges from 0.25 to 0.34 during the first 72 h of life, and a single SD is 0.08, which provides a ±2 SD bandwidth of 0.32.

Figure 2
figure 2

rScO2 reference value curves for neonates of (a) 24–25 wk GA, (b) 26–27 wk GA, (c) 28–29 wk GA, and (d) 30–31 wk GA. The line patterns depict different percentiles: dotted lines indicate p2.3 and p97.7, dash dot dot dash lines indicate p5 and p95, dashed lines indicate p10 and p90, dash dot dash lines indicate p20 and p80, and solid lines indicate p50. GA, gestational age.

PowerPoint slide

Figure 3
figure 3

cFTOE reference value curves for neonates of (a) 24–25 wk GA, (b) 26–27 wk GA, (c) 28–29 wk GA, and (d) 30–31 wk GA. The line patterns depict different percentiles: dotted lines indicate p2.3 and p97.7, dash dot dot dash lines indicate p5 and p95, dashed lines indicate p10 and p90, dash dot dash lines indicate p20 and p80, and solid lines indicate p50. GA, gestational age.

PowerPoint slide

Table 2 lists the coefficients of the rScO2 and cFTOE models. For example, mean rScO2 increases 0.9% per week of GA. In the models, PA is also represented by the square of PA (PA-sq) to model the parabolic relationship that rScO2 and cFTOE have with PA. The PA, PA-sq, and interactions with PA and PA-sq determine the location of the vertex (i.e., maximum for rScO2 and minimum for cFTOE) and the curvature of the parabola. Figure 4 is a graphical representation of the main effects and interactions of being born with a birth weight (BW) <−1 SD (i.e., small for gestational age, SGA) and having a hemodynamically significant patent ductus arteriosus (hsPDA) as reported in Table 2 . Infants who developed a hsPDA <84 h of life had a lower rScO2 and demonstrated a sharper decline after ~24 h. Infants born SGA started off with a higher rScO2 and had slightly lower values at 72 h compared to those at 1 h PA, whereas infants born appropriate for gestational age (AGA) had higher values at 72 h than at 1 h PA. This difference between infants born SGA and those born AGA diminishes over time (i.e., SGA and AGA lines in Figure 4 converge) but is still present at 72 h PA. No significant differences were found between data obtained before or after June 2012 (i.e., Poly 5 software + INVOS 4100 vs. BedBase software + INVOS 5100c).

Table 2 Final model coefficients for the rScO2 and cFTOE
Figure 4
figure 4

Graphic representation of the interactions of hsPDA and SGA with the representatives of postnatal age in the model for (a) rScO2 and (b) cFTOE. Black solid lines indicate AGA with hsPDA, gray solid lines indicate AGA without hsPDA, black dashed lines indicate SGA with hsPDA, and gray dashed lines indicate SGA without hsPDA. AGA, appropriate for gestational age; cFTOE, cerebral fractional tissue oxygen extraction; hsPDA, hemodynamically significant patent ductus arteriosus; rScO2, regional cerebral oxygen saturation; SGA, small for gestational age.

PowerPoint slide

Conversion Diagrams

A strict linear model provided the best fit to convert data obtained by the SAFB-SM (adult) sensor to the CNN (neonatal) sensor: rScO2-neo = 0.8481 * rScO2-adult + 19.11, R2 = 0.65. Figure 5 is a conversion of Figure 2 by using this equation.

Figure 5
figure 5

rScO2 reference value curves obtained with the neonatal sensor for neonates of (a) 24–25 wk GA, (b) 26–27 wk GA, (c) 28–29 wk GA, and (d) 30–31 wk GA. The line patterns depict different percentiles: dotted lines indicate p2.3 and p97.7, dash dot dot dash lines indicate p5 and p95, dashed lines indicate p10 and p90, dash dot dash lines indicate p20 and p80, and solid lines indicate p50. Note: the neonatal probe (CNN) rScO2 values were obtained by using a conversion from the adult (SAFB-SM) probe values: rScO2-neo = 0.8481 rScO2-adult + 19.11. GA, gestational age; rScO2, regional cerebral oxygen saturation.

PowerPoint slide

Discussion

This is the first study to report reference values of rScO2 and cFTOE obtained using NIRS during the first 72 h of life in a large cohort of preterm neonates born at a GA <32 wk.

Four factors should be taken into account when comparing the work reported here to work of others: (i) GA, (ii) PA, (iii) sample size, and (iv) the sensor and device that were used (see discussion below). Values found in the literature agree quite well with the values reported here ( Table 3 , mean difference: −0.9%). The differences are likely explained by the characteristics of the reported populations (e.g., GA, PA, specific morbidity), duration of measurements, and small sample sizes (9,10,11,17,18,19,20). It seems likely that the rScO2 will either stabilize or may even increase again after 72 h (12,13,15,21). Note that van Hoften et al. collected data with a pediatric sensor and Pocivalnik et al. and Pichler et al. with a neonatal sensor (15,20,21).

Table 3 rScO2 values obtained from the literature compared to rScO2 reference values established in this study

It is noteworthy how close the −2 SD bands (i.e., p2.3) are to the rScO2 threshold (i.e., 33–44%) reported to be associated with functional impairment of the brain (22,23). A lower CBF, either regional or global, in infants with a lower GA is the most plausible explanation for the positive association between GA and rScO2. Roche-Labarbe et al., while using a frequency domain NIRS system, also demonstrated lower levels of cerebral oxygenation during the first 7 wk of life in infants with a GA <31 wk compared to infants with a GA >31 wk (24). Furthermore, their data show that infants with a GA of 24–27 wk have the lowest blood flow index, supporting lower CBF as an explanation for lower cerebral oxygenation in younger infants. No associations were found between head circumference and rScO2, and SaO2 and GA. This makes the influence of head circumference (i.e., different curvature of the head influencing NIRS from a technical point of view) or SaO2 unlikely. Furthermore, a similar (inverse) association was found between GA and cFTOE. An increased metabolic demand in neonates of lower GA seems unlikely as cerebral activity increases with GA (25).

Female neonates had lower rScO2 as compared to male neonates. This gender difference was also observed by the Pichler et al. during transition from fetal to neonatal life (personal communication, data not published). Again, this could not be explained by a difference in SaO2 or head circumference. Therefore, possible explanations are a higher (regional) CBF or lower metabolic demand. A hsPDA can cause a ductal steal phenomenon with a surplus of pulmonary flow at the cost of systemic perfusion and thus CBF (26). Although notably increasing with PA, the effect of a hsPDA seems rather limited during the first 3 d of life. The most plausible explanation for this is the fact that most hsPDAs become clinically apparent from day 3 onward. In addition, an objectively present hsPDA (i.e., confirmed by cardiac ultrasound) does not necessarily decrease CBF, and thus rScO2, as the magnitude of systemic steal depends on shunt volume and left ventricular output. Moreover, in this study, the PA at diagnosis was dichotomized (i.e., ≤84 h); therefore, the exact PA of the individual at diagnosis and start of treatment was not taken into account. In previous publications, we took a different approach with case–control designs and the start of indomethacin or surgery as time reference, at median postnatal days 2 and 7, respectively (26,27).

Higher rScO2 values in infants born SGA demonstrate the brain-sparing effect with a compensatory higher CBF. This has been demonstrated previously with other techniques (28). The difference in rScO2 between SGA and AGA infants diminishes over time, suggesting that the CBF returns to normal after day 3. Interestingly, unlike in AGA infants, rScO2 values in SGA infants were slightly lower at 72-h PA compared to rScO2 values shortly after birth. This suggests downregulation of compensatory mechanisms instead of a relative lack of hemodynamic development in SGA infants as an explanation for values converging toward AGA values. The limited number of severe SGA cases (i.e., BW z-score <−2, n = 18) prevented statistical significance of the effect that being severely SGA has on rScO2 and cFTOE.

We preferred to present “overall” reference curves over presenting numerous curves according to different morbidities. Therefore, the reference curves are valid for a population of preterm infants admitted to the NICU, with inherent morbidity. The generalized additive models for location, scale, and shape (GAMLSS) results were similar to the mixed-model approach. We chose to report the mixed-model procedure because results are easier to interpret and there is more extensive expertise in mixed-model procedures in our institution.

As mentioned before, substantial differences exist between different NIRS sensors (14,15). The correlation between values obtained with the adult and neonatal sensor is not perfect (R2 = 0.65), which is probably caused by the in vivo nature of our experiment. However, the relation is clearly linear, which has also been demonstrated in vitro (29). Although the difference between the neonatal and adult sensor can be as high as 15% (mean: 10%), trend monitoring is still possible in a way similar to adult sensors. Moreover, the ±2 SD limits provide a “bandwidth” of ~30% which makes the maximum 15% difference less stressing. However, awareness of a possible offset between sensors is crucial when comparing data between patients, institutions, or devices. For example, a rScO2 of 55% seems low but acceptable when using an adult sensor ( Figure 2 ) but is below the −2 SD threshold in all GA groups when using a neonatal sensor ( Figure 5 ). A rScO2 of 55% with the neonatal sensor converts to an adult sensor value of ~40%, which is close to or below the thresholds (33–50%) reported to be associated with neuronal damage and adverse neurodevelopmental outcome (5,22,23). Two of these thresholds were established in piglet studies by using devices that are not commercially available (22,23). Therefore, these thresholds should be used with caution, especially in preterm neonates. Also, higher values pose pitfalls, as most devices have an upper detection limit of 95%. Any value >85% obtained by an adult sensor would register as 95% (i.e., 85 + 10%) with a neonatal sensor, losing the ability to monitor the rScO2 trend. This is particularly relevant for the prognostic value of the rScO2, for example, in asphyxiated neonates (30,31).

A possible limitation of this study is the generalizability of the results as participants were admitted to a single level III NICU. However, neonatal intensive care in the Netherlands takes place in 10 NICUs and admissions are purely based on geography. Equipment availability was the only factor to prohibit data collection. As unavailability was supposedly random, this influence should be minimal. Moreover, the number of recording setups substantially increased over the years, now ensuring round-the-clock availability. The second limitation is the restriction to the first 72 h of life. This choice was made to encompass the most vulnerable period of life and to limit strain on nursing staff at the same time. The final limitation is clinical practice in our unit regarding the used SaO2 thresholds (i.e., 85–92%), which might differ from other institutions. The rScO2 results were not corrected for SaO2 to avoid overly complicating results and because model coefficients ( Table 2 ) changed less than 5% when correcting to a SaO2 of 90%. Moreover, the cFTOE curves already provide a form of SaO2 correction. Altogether, we feel confident that the current results are generalizable to other populations of preterm neonates with a GA <32 wk during the first 72 h of life.

Currently, the core application of NIRS on the NICU lies in trend monitoring. For the inexperienced user, we suggest plotting an infants’ rScO2 in the appropriate GA-specific reference curve. In case of sudden changes in rScO2 ≥7% (i.e., 1 SD), we recommend evaluation of clinical parameters (e.g., ventilator settings, hemoglobin levels, presence of a hsPDA, medication, perform a cranial ultrasound), but only after ensuring that the measurement setup has not changed (e.g., sensor displacement). Likewise, absolute rScO2 values close to or outside the ±2 SD bands, should trigger an evaluation, similar to what has been done during the SafeBoosC trial (Safeguarding the Brains of our smallest Children) (32). For a more detailed discussion on which parameters to evaluate in case of either low or high levels or rScO2, we refer to the treatment guideline published by the “SafeBoosC” research group (33).

Conclusion and Future Research Directives

This study provides reference values for rScO2 and cFTOE measured by NIRS during the first 72 h of life in premature infants. Both rScO2 and cFTOE are influenced by GA, PA, hsPDA, gender, and being born SGA. Furthermore, an equation is provided to extend the results to rScO2 values obtained with a neonatal NIRS sensor. These data provide an additional way for applying NIRS on the NICU, on top of trend monitoring. Furthermore, reliable reference data can be useful in future studies. Indices of cerebral oxygenation have only been suggested to be related to outcome (5,22). Future research should focus on developing robust indices of cerebral oxygenation that are related to (long-term) outcome and can be used to guide interventions. Our results suggest that GA- and PA-specific thresholds are worth exploring in this regard. The SafeBoosC research group already reported that NIRS can be used to stabilize the cerebral oxygenation in preterm infants (32).

Methods

Patients

This study is part of an ongoing prospective observational cohort study which aims to record physiological parameters during at least the first 72 h of life in all infants born with a GA <32 wk who are admitted to the NICU of the Wilhelmina Children’s Hospital, Utrecht, The Netherlands. The medical ethical board of the University Medical Center Utrecht approved this study. Informed parental consent was obtained in all cases. Data collection was attempted in 1,059 infants between January 2005 and September 2013.

Data Collection

Obstetrical, intrapartum, and neonatal data were collected from the hospital records. Peri- and intra-ventricular hemorrhages (PIVH) were graded according to the classification of Papile et al. (34). The presence of a hsPDA was defined as a PDA confirmed to be hemodynamically significant on cardiac ultrasound and either treated with indomethacin or surgically closed (2).

Standard physiological parameters were monitored using a patient monitor (IntelliVue MP70, Philips, Best, The Netherlands): SaO2 using a pulse-oximetry probe, arterial blood pressure by means of an indwelling catheter (e.g., umbilical, radial, or tibial artery), and heart rate by gel electrodes. In general, the pulse-oximetry probe was placed on one of the lower limbs. In case of a hsPDA, the probe was placed on the right hand (i.e., pre-ductal). rScO2 was monitored by using a two-wavelength (i.e., 730 and 810 nm) NIRS monitor (INVOS 4100 or 5100(c)); Covidien, Mansfield, MA) in combination with a small adult sensor (SomaSensor SAFB-SM, Covidien). An elastic bandage was used for sensor fixation. Until June 2012, data were recorded with Poly 5 (Inspector Research Systems, Amsterdam, The Netherlands) at a sample rate of 1 Hz using an INVOS 4100 monitor. Thereafter, in-house developed software (BedBase, University Medical Center Utrecht, Utrecht, The Netherlands) was used to record data from the patient monitor and an INVOS 5100c NIRS monitor at a sample rate of 0.4 Hz.

Data Processing

Data were analyzed with the offline version of the BedBase software (SignalBase, University Medical Center Utrecht, Utrecht, The Netherlands). Before analysis, artifacts were removed manually. Artifacts were defined as: changes in rScO2 that could not be physiologically explained (e.g., a 30% step change between two subsequent data points) or changes that were accompanied by severe distortion in the other parameters suggesting infant movement or handling. Thereafter, 1-h periods were selected during the first 72 h of life and counted in reference to a patient’s birth date and time (i.e., PA). Periods with short drops in SaO2 (i.e., <85%) were not included in the analysis. In case of SaO2 drops where additional O2 was given to assist recovery, the duration of the associated increase in SaO2 and rScO2 over baseline conditions was also excluded from analysis (35).

Statistical Analysis

Before statistical analysis, 1-h periods containing less than 10 min of data were rejected. Mean values of the 1-h periods were used for analysis. A mixed-model approach was performed using R for Windows 64-bit, version 3.1.1 (R Core Team, R Foundation for Statistical Computing, Vienna, Austria) with the nlme package. This approach can handle missing data and obsoletes correcting for multiple comparisons.

The decision was made, a priori, to investigate four variables: GA (in weeks), BW z-score, gender, and the presence of a hsPDA. The BW z-score was based on recently published Dutch reference curves and explored both as a continuous variable and dichotomized at −1 SD and −2 SD (36).

The time of diagnosis of a hsPDA was expressed as PA at time of the cardiac ultrasound. This PA at diagnosis was dichotomized at different cutoffs, ranging from 60 to 132 h with 12-h increments.

Linear, squared, and polynomial models of time (i.e., PA) were explored to find the best fit to the data. Either the rScO2 or cFTOE was selected as the dependent variable, with the individual subject as a random factor. Both main effects and interactions with PA were explored. It was decided a priori that interactions with PA would be included into the final model when the interaction was statistically significant and caused at least a 10% coefficient change with respect to the coefficient for PA in the model with only main effects.

The final model was used to create rScO2 and cFTOE reference curves by generating predictions based on a new set of predictors (i.e., PA, GA, hsPDA, gender, and BW). This new set was generated as follows: PA ranging 1–72 h with 0.2-h increments (e.g., 1.0, 1.2, 1.4 h, etc.), GA ranging 24–32 with 1-wk increments, hsPDA yes/no, female gender yes/no, and BW <−1 SD yes/no. The SEs of these predictions were calculated and used to create percentile plots based on a normal distribution.

To assess the robustness of the results, analysis was also performed using GAMLSS in R with the GAMLSS package (37).

Unless specified otherwise, data are presented as mean with SD for parametric data, median with interquartile range for nonparametric data, and counts (%) for categorical data. A P value < 0.05 was considered statistically significant.

Conversion Models for Neonatal NIRS Sensor

In a subset of infants (n = 16, GA 30 ± 3 wk), rScO2 was recorded bilateral by using the neonatal (OxyAlert CNN cerebral NIRsensor) and small adult sensor (SAFB-SM, as described above), as reported previously (14). Both sensors use a single LED light source, two distant detectors (30 and 40 mm), and two wavelengths (i.e., 730 and 810 nm). After at least 1 h of stable recordings (e.g., free of care, feedings, and interventions), the sensors were switched to the contralateral side, and data collection continued for another hour. Data analysis was performed in MATLAB (vR2011b; The Mathworks, Natick, MA), and artifacts were removed manually. A linear model, polynomial models, and a piecewise linear model were examined to find the best fit (i.e., lowest residuals and highest R2) in order to convert data obtained by the neonatal sensor to small adult sensor–equivalent values and vice versa.

Statement of Financial Support

A.C. received a postdoctoral fellowship from the “Research Foundation Flanders” (FWO, Brussels, Belgium) and S.V.H. received funding from the European Research Council (ERC, Brussels, Belgium) under the European Union’s Seventh Framework Programme (FP7/2007–2013) as well as an ERC (ERC, Brussels, Belgium) advanced grant (no. 339804): Biomedical Data Fusion using Tensor based Blind Source Separation (BIOTENSORS). There are no further financial support acknowledgments to report.

Disclosure

The authors have no conflicts of interest to disclose.