Sage Journals: Discover world-class research

Abstract

A characteristic feature of complex healthy biological systems is the ability to react and adapt to minute changes in the environment. This ‘complexity’ manifests itself in highly irregular patterns of various physiological measurements. Here, we apply Multiscale Entropy (MSE) analysis to assess the complexity of systemic and cerebral near-infrared spectroscopy (NIRS) signals in a cohort of 61 critically ill preterm infants born at median (range) gestational age of 26 (23–31) weeks, before 24 h of life. We further correlate the complexity of these parameters with brain injury and mortality. Lower complexity index (CoI) of oxygenated haemoglobin (HbO₂), deoxygenated haemoglobin (Hb) and tissue oxygenation index (TOI) were observed in those infants who developed intraventricular haemorrhage (IVH) compared to those who did not (P = 0.002, P = 0.010 and P = 0.038, respectively). Mean CoI of HbO₂, Hb and total haemoglobin index (THI) were lower in those infants who died compared to those who survived (P = 0.012, P = 0.004 and P = 0.003, respectively). CoI-HbO₂ was an independent predictor of IVH (P = 0.010). Decreased complexity of brain signals was associated with mortality and brain injury. Measurement of brain signal complexity in preterm infants is feasible and could represent a significant advance in the brain-oriented care.

Keywords

Brain injury brain monitoring complexity entropy preterm

Introduction

Brain injury in the preterm infant is associated with an increased risk of poor developmental outcome and death.^1,2 The preterm brain is particularly susceptible to intraventricular haemorrhage and haemorrhagic parenchymal infarction, usually occurring within the first 3 days of life. This results from bleeding in the germinal layer surrounding the lateral ventricles; this is an area of high cellular proliferation in the developing foetal brain, where neuron and glial cells precursors arise. The rich vasculature of the germinal matrix is formed by thin vessels, which are very sensitive to hypoxia and changes in cerebral perfusion pressure.² Several studies have shown that fluctuations in systemic and cerebral blood flow have been associated with the pathophysiology of ischemic and haemorrhagic brain lesions in preterm infants.^3–5 In clinical practice, strategies for preventing brain injury in the preterm population have emphasized the importance of measurements of systemic blood flow and cardiac function; more recently, early monitoring of cerebral blood flow and oxygenation has been investigated, but has yet to be established as the standard of care.^6,7 Near-infrared spectroscopy (NIRS) has been applied as a non-invasive monitor to measure cerebral oxygenation. Conventional NIRS measures the changes in the concentration of oxygenated haemoglobin (HbO₂) and deoxygenated haemoglobin (Hb). Spatially resolved NIRS is able to provide a quantitative index of cerebral oxygen saturation – the tissue oxygenation index (TOI) and a semi-quantitative index of cerebral blood volume – the tissue haemoglobin index (THI).⁸ Conventional and spatially resolved NIRS have been used to investigate cerebral haemodynamics in both adults and newborn infants.^9–11 Assuming a low and constant cerebral metabolic rate for oxygen (CMRO₂), TOI can also be interpreted as a surrogate measurement of cerebral blood flow and has been extensively used in neonatal research to investigate brain perfusion or cerebral autoregulation.^12–16 More recently, studies have demonstrated that it is possible to stabilize cerebral oxygenation and monitor cerebrovascular reactivity using NIRS, supporting the idea that early monitoring of cerebral blood flow could have a positive impact on outcome of preterm infants.^6,17

Most studies using cerebral NIRS or systemic measurements of blood flow use linear analysis. However, a complex biological system, such as the human brain, includes many complex regulatory mechanisms that interact in a non-linear way, resulting in effects that cannot be understood wholly through the analysis of its individual constituents. Therefore, non-linear methods to assess physiological signals may reveal more accurately the magnitude of changes over time. A complex biological system has the ability to react and adapt to minute changes in its environment. When the complexity of a system is reduced or lost, a pathological status is potentially present. Loss of complexity has been observed in several diseases and ageing.^18,19 It has also been associated with poor outcome in severely sick adults and children.^20,21 A suggested method of assessment of the complexity of biological signals in healthy and pathological systems is the Analysis of Entropy, a non-linear measurement of system randomness and unpredictability. In the early 1990s, Approximate Entropy was used first to assess the irregularity of biological time series and has been extensively used in physiological and medical studies.²² A decade later, Sample Entropy was proposed as a modified method, which was less dependent on the length of the time series. In neonates, Sample Entropy analysis of heart rate variability (HRV) was used to predict the onset of neonatal sepsis 24 h before clinical diagnosis.²³ More recently, Multiscale Entropy (MSE) has been used to provide a more meaningful measure of complexity, exploring calculations of Sample Entropy over multiple time scales. MSE shows that correlated random signals are more complex than uncorrelated random signals (white noise). Moreover, MSE describes pathological physiology better than simple entropy measures, doing a better differentiation between healthy, complex systems and deranged or random interactions.²⁴ MSE has been used to assess HRV and electroencephalogram (EEG) changes in patients with different neurological conditions.²⁵ Complexity of intracranial pressure (ICP) measured using MSE has been correlated with clinical outcome. In adults with traumatic brain injury, reduced complexity of mean ICP was associated with poor outcome and increased mortality.²⁶ In infants, MSE was significantly reduced in infants with neonatal seizures and later diagnosis of epilepsy when compared to control infants.²⁷

To date, there are no previous studies using MSE to assess complexity of NIRS signals in preterm infants. Here, we hypothesize that MSE can be applied to study the complexity of cerebral NIRS signals and the complexity of systemic physiological signals in a cohort of preterm infants undergoing intensive care. We further hypothesize that the complexity index (CoI) of cerebral NIRS signals and systemic signals, measured using MSE, correlate with intraventricular haemorrhage and mortality in this population.

Methods

Ethics

This prospective observational study was authorized by The Research and Development Department of Cambridge University Hospitals NHS Foundation Trust and approved by The East of England Research Ethics Committee (12/EE/0524), in accordance with the declaration of Helsinki. The data included in this study were collected at the Neonatal Intensive Care Unit (NICU) at The Rosie Hospital, a busy tertiary unit that receives the most complex antenatal and postnatal cases from across the east of England region. Data were collected from September 2010 to May 2015. All infants were studied following signed informed parental consent.

Population

Preterm infants born at less than 32 weeks’ gestational age, with birth weight <1500 g, who had indwelling arterial catheters inserted for clinical reasons, were enrolled. The recruitment protocol was the same applied for a previous study on monitoring of cerebrovascular reactivity recently published by our group, therefore, this cohort included sick preterm infants who required invasive monitoring of continuous blood pressure.¹⁷ Exclusion criteria included the presence of major malformations. Sixty-one preterm infants were studied (Table 1) within the first 24 h of age. The mean age at the start of study was 8 h (standard deviation ±5 h). Fifty-seven (93%) infants had antenatal steroids and all infants were intubated and received surfactant within 10–20 min of birth. Thirty-four (55.7%) had a further one to three doses of surfactant after admission. All infants were receiving mechanical ventilation support at the start of the study.

Table 1.

Demographic data.

Variables	Total (n = 61)	No-IVH (n = 41)	IVH (n = 20)	P	Survived (n = 53)	Died (n = 8)	P
Gestational age (weeks + days)	26 + 6 (23 + 3 – 31)	26 + 6 (23 + 4 – 31)	25 + 5 (23 + 3 – 29 + 2)	0.12	26 (23 + 3 – 30 + 1)	26 (24 + 1 – 31)	0.88
Birth weight (g)	805 (525–1355)	805 (525–1355)	802 (550–1320)	0.98	810 (525–1355)	767 (575–1180)	0.72
Gender (male/female)	26/35	20/21	6/14	0.18	23/30	3/5	1.0
CRIB II	11 (5–16)	10 (5–16)	12 (9–16)	0.025*	11 (5–16)	12 (8–16)	0.33
Inotrope medication	27 (44.3%)	18 (43.9%)	9 (45%)	1.0	21 (39.6%)	6 (75%)	0.12
Sepsis	37 (60.7%)	21 (51.2%)	16 (80%)	0.023*	30 (56.6%)	7 (87.5%)	0.24

Characteristics of enrolled infants and differences between no-IVH (intraventricular haemorrhage) and IVH groups and between infants who survived and died. Gestational age, birth weight and CRIBII (clinical risk index for babies II) are presented as median (range). Gender is presented as absolute numbers. Remaining variables are presented as frequencies.

P < 0.05.

Measurements

Following enrolment, a NIRS sensor from a NIRO 200NX near-infrared spectrophotometer (Hamamatsu Photonics, KK, Japan) was placed on one side of the infant’s head in the temporoparietal area. The NIRO 200NX is a bedside NIRS monitor that measures changes in the concentration of HbO₂ and Hb using conventional NIRS, calculated using the modified Beer–Lambert Law, as well as providing measurements of TOI and THI using spatially resolved NIRS, calculated using an approximation of the light diffusion equation. The NIRO 200NX uses three light-emitting diodes with wavelengths of 735, 810 and 850 nm, respectively, and two detecting photodiodes, to measure light attenuation at different distances from the source. A sensor holder was used to fix the light source at 3 cm away from the receiver diodes and adhesive paper secured the sensor to the infant’s skin. A light-proof cover was also used. The sensor was changed to the opposite temporo-parietal side of the infant’s head every 6 to 8 h to avoid skin marks.

The following systemic physiological signals were recorded simultaneously and continuously from the neonatal intensive care monitors (Solar 8000; Carescape B850, GE Healthcare, Milwaukee, Wisconsin): full resolution of arterial blood pressure waveform (120 Hz, Solar, or 100 Hz, Carescape) peripheral oxygen saturation (SaO₂) and heart rate (HR) (1 Hz). NIRS signals were sampled at the rate of 1 Hz. Both NIRS and physiological signals were collected, synchronized and stored using ICM+ software (Cambridge Enterprise Ltd, Cambridge, UK).²⁸

Clinical data within the study period and data on outcome were collected from the medical notes. Clinical decision-making was at the sole discretion of the attending neonatal consultant and included the management of blood pressure and treatment of sepsis. All infants were screened for sepsis and antibiotics started within the first hour after admission to the neonatal unit. Those with no clinical signs of sepsis, normal c-protein reactive (CRP), negative blood culture or antibiotics given only for the first 48 h of life were considered to have a negative sepsis screen. Those with presumed sepsis (CRP moderately high, negative culture and antibiotics for 5–7 days), clinical sepsis (high CRP, clinically unwell, negative culture and antibiotics for 7–14 days) or confirmed sepsis (clinical sepsis with positive culture) were considered to have a positive screen. CRIB II (Clinical Risk Index for Babies II) score of mortality and morbidity was calculated for all infants in the study.²⁹ Cranial ultrasonography scans were performed at the start of the study and repeated every 12–24 h until third of life and then every 1–3 weeks until corrected gestational age at term. IVH was defined according to Papile et al. and the greatest grade of haemorrhage during the admission period was used for analysis.³⁰

Data pre-processing

Data collected at the full frequency resolution supported by respective monitors was then retrospectively analysed using ICM+.²⁸ Artefacts had been identified and manually removed. Most of the artefacts were the result of umbilical arterial line sampling, infants being handled or movements. Nurses used a separate study chart to record date and time of events like cares, examination, procedures, changes on drug infusions or any other intervention. These annotations allowed experienced investigators to exclude manually periods of data with invalid measurements or with transients induced by external interventions. To date there is no universal, validated method for automatic robust removal of data artefacts.

Subsequently, all the recorded physiological variables were treated with the moving average filter of 10-s length and downsampled to frequency of 0.1 Hz. This has effectively removed all the components associated with pulse and respiratory activities while retaining waves in the low and very low frequency range (<0.02 Hz), thus focusing further analysis entirely on the cerebrovascular and systemic regulatory responses. These time series were then entered into the MSE analysis in order to produce complexity indices of the studied variables.

MSE analysis

MSE analysis was performed using algorithm published originally by Costa et al.^24,25 Briefly, the analysis can be divided into three steps: constructing coarse-grained time series, computing sample entropy for each of them and calculating overall complexity.³¹ Coarse-grained time series for scale τ is derived from original time series by averaging its τ samples in a moving (non-overlapping) window. Thus, time series for scale τ is τ times shorter than original time series. Particularly, for scale 1, we get simply original time series. For each coarse-grained time series, sample entropy was computed. Sample entropy estimates the probability that similar sequences of m consecutive points in time series will also be similar when sequences of the length m + 1 are considered. To be precise, two sequences containing the same number of points are considered similar if the absolute differences between their corresponding points are smaller than a specified tolerance value r. Sample entropy is calculated as the negative natural logarithm of the ratio of the total number of (m + 1)-point template matches to the total number of corresponding m-point template matches.³² Higher values of sample entropy indicates that time series is more irregular and less predictable. We assumed m = 2 and r = 0.15 SD, where SD denotes standard deviation of analysed time series, as originally proposed by Costa et al.²⁴

MSE was calculated up to scale 20.^25,26 It can be presented as a function of scale factor (Figure 2). The area under MSE curve, or in other words, the sum of sample entropy values for all scales, was defined as the CoI.^26,31

Statistical analysis

Data were assessed for normality using the Shapiro–Wilk test and analysed using parametric or non-parametric statistical methods as appropriate. Comparisons between groups of infants with IVH and no-IVH, and between infants who died or survived, were made using the Student t-test, Mann–Whitney or Chi-square test, depending on data distribution. Binary logistic regression analysis was used to ascertain independent effects of the complexity of NIRS signals on the likelihood of the infants developing IVH. Spearman’s rank correlation coefficient (r_s) was used to correlate the complexity of brain and systemic signals. A statistical test was considered significant if P value was < 0.05 (two-tailed). Statistical analysis was performed using the SPSS software package version 23 (SPSS, Inc., Chicago, IL, USA).

Results

Demographic data and differences between groups are shown in Table 1. The median (range) gestational age was 26 weeks + 5 days (23 + 3 – 31). In all, 75% were extremely preterm infants (born at less than 28 weeks). Twenty infants (32.8%) developed IVH; six of them had IVH diagnosed within the first 24 h of life. Of those who developed IVH, 10 infants had IVH grade II, 4 had IVH grade III and 6 had IVH grade IV. Only one infant had minimal bleeding in the germinal matrix (small grade I IVH) and, therefore, this infant was included in the non-IVH group. Eight infants (13.1%) died before term corrected age of 40 weeks. Causes and age at death are described in Table 2. Seven of them were born before 28 weeks of gestational age (extremely preterm infants) and one was born at 30 weeks + 2 days, who was a severe in-utero growth restricted (IUGR) infant and birth weight of 700 g. All infants who died had signs of antenatal or perinatal hypoxia, as severe IUGR, twin–twin transfusion, cord prolapse or profound hypotension and hypoplastic lungs. Four of those infants who died (50%) also had IVH diagnosed during the admission in the neonatal unit. There was no statistical correlation between mortality and IVH (P = 0.42).

Table 2.

Age at death and causes related to death.

Infant	GA at birth (weeks + days)	CGA at death (weeks + days)	Cause of death
01	23 + 5	24 + 1	Immature lungs, pulmonary haemorrhage and IVH
02	26 + 2	28 + 3	Twin–twin transfusion, severe antenatal brain and heart hypoxia, multi-organ failure
03	24 + 4	25 + 2	Severe antenatal hypoxia, IVH and multi-organ failure
04	24 + 1	25 + 5	NEC (antenatal history of drug abuse, severe RDS and pneumothorax within the first 48 h of life)
05	25 + 5	29	NEC (severe hypotension, pulmonary haemorrhage within the first 24 h of life)
06	24 + 2	28 + 4	Sepsis (perinatal history of PPROM and cord prolapse)
07	26 + 3	31 + 2	Sepsis (lung hypoplasia, PPHN and severe hypotension and IVH within the first 24 h of life)
08	30 + 2	37 + 2	NEC (perinatal history of IUGR, reversal end-diastolic flow on antenatal Dopplers, severe hypotension within first 24 h of life)

IVH: intraventricular haemorrhage; NEC: necrotizing enterocolitis; PPROM: preterm prolonged rupture of membranes; PPHN: persistent pulmonary hypertension of the newborn.

Examples of monitoring of brain and systemic signals are presented in Figure 1(a) (infant who died) and Figure 1(b) (infant who survived).

Figure 1.

Example of brain and systemic signal monitoring for an infant who died (a) and survived (b).

Figure 2 shows average MSE plots of HbO₂ and Hb for the whole cohort of infants. Sample Entropy values were averaged individually for each scale and plotted against the scale. This was done separately in groups of those infants who had IVH and did not have IVH, as well as those infants who died and survived. It can be observed through the lower value of sample entropy at each scale factor, resulting in a decrease in overall CoI, that infants who developed IVH or died had consistently lower entropy in the NIRS signals.

Figure 2.

Multiscale entropy. Results are mean ± 1.96 standard error of mean. Oxygenated haemoglobin (HbO₂) and deoxygenated haemoglobin (Hb).

Association between brain and systemic signals with outcome

Lower mean CoI-HbO₂, CoI-Hb and CoI-TOI were observed in those infants who developed IVH compared to those who did not (Table 3). However, there was no difference in mean CoI-THI between the two groups (Table 3). Mean TOI was not significantly different between infants who had IVH and those who did not have IVH (P = 0.38).

Table 3.

Differences in complexity of brain and systemic signals between outcome groups.

	IVH	No-IVH	P	Died	Survived	P
CoI-HbO₂	6.24 (5.02–7.45)	11.69 (9.43–13.96)	0.002	5.30 (3.26–7.35)	10.60 (8.74–12.45)	0.012
CoI-Hb	6.16 (7.54–10.71)	9.03 (7.54–10.71)	0.010	4.09 (2.77–5.36)	8.69 (7.28–10.15)	0.004
CoI-TOI	10.09 (7.62–12.93)	13.9 (11.94–16.25)	0.038	8.96 (5.48–13.21)	13.36 (11.43–15.26)	0.057
CoI-THI	4.44 (2.96–6.72)	6.23 (4.64–8.60)	0.242	2.38 (1.48–3.35)	6.16 (4.69–7.97)	0.003
CoI-MABP	17.85 (14.33–21.37)	21.68 (19.12–24.25)	0.82	14.98 (10.305–19.722)	21.25 (19.24–23.41)	0.04
CoI-SaO₂	17.43 (14.92–19.94)	18.07 (16.21–19.93)	0.35	16.01 (12.54–19.48)	18.14 (16.52–19.76)	0.16
CoI-HR	18.65 (15.31–21.99)	20.72 (18.34–23.10)	0.31	17.16 (12.58–21.74)	20.47 (18.38–22.57)	0.24

Mean and 95% confidence interval for complexity index (CoI) of NIRS and systemic physiological signals.

Decreased complexity of brain signals was observed in infants who died. Mean CoI-HbO₂, CoI-Hb and CoI-THI were lower in those infants who died compared to those who survived (Table 3). CoI-TOI just failed to reach statistical significance (P = 0.057). There was no difference in mean TOI between these two groups (P = 0.52).

There was no difference between mean CoI of mean arterial blood pressure (MABP), mean CoI-HR and mean CoI-SaO₂ between those infants who developed IVH and those who did not (Table 3). Mean MABP, mean SaO₂ and mean HR were not different between these two groups (P = 0.52, P = 0.46 and P = 0.91, respectively). In contrast, mean CoI-MABP was lower in those infants who died compared to those who survived. However, there was no difference in mean CoI-SaO₂ or mean CoI-HR between these two groups (Table 3). There was no difference between MABP, SaO₂ and HR between those infants who survived or died (P = 0.79, P = 0.38 and P = 0.20, respectively).

Complexity of brain signals as an independent predictor of IVH

Out of all the studied parameters and demographic data only CRIB II (P = 0.031), sepsis (P = 0.025) and CoI-HbO₂ (P = 0.004) were identified by binary logistic regression analysis as predictors of IVH when individual analysis was performed for each of these variables (Table 4). Subsequently, these variables were entered as co-variants in a binary logistic regression model. The continuous variables in the model met the assumption of linearity and there were no studentized residuals. Of the three-predictor variables, only CoI-HbO₂ remained statistically significant, showing that decreased complexity was independently associated with an increased likelihood of having IVH (Table 4).

Table 4.

Binary logistic regression for CoI-HbO₂, sepsis and CRIB II as predictors of IVH.

	B	S.E.	Wald	df	P	Odds ratio	95% C.I. for Odds ratio
	B	S.E.	Wald	df	P	Odds ratio	Lower	Upper
Independent univariate models
CoI-HbO₂	−.236	.082	8.286	1	.004	.790	.637	.928
Constant	1.247	0.667	3.496	1	.062	3.481
Sepsis	−1.574	.704	4.993	1	.025	.207	.0.52	.824
Constant	−0.272	.332	.672	1	.413	.413
CRIB II	.234	.108	4.663	1	.031	1.264	1.022	1.563
Constant	−3.371	1.303	6.692	1	.010	.034
Multivariate model
CoI-HbO₂	−.216	.083	6.706	1	.010	.806	.684	.949
CRIBII	.249	.144	2.979	1	.084	1.282	.967	1.701
Sepsis	−1.395	.799	3.049	1	.081	.248	.052	1.186

Sepsis is for presence of sepsis compared to no sepsis.

CoI-HbO₂: complex index of oxygenated haemoglobin; CRIBII: clinical risk index for babies II.

CoI-HbO₂ had no correlation with sepsis (P = 0.074) or CRIB II (P = 0.138). Table 5 shows the matrix of univariate correlations between CoIs of systemic and brain signals.

Table 5.

Correlation between complexity index of brain and systemic signals.

		CoI-HbO2	CoI-Hb	CoI-TOI	CoI-THI	CoI-SaO₂	CoI-MABP	CoI-HR
CoI-HbO₂	rs	1.000	.696*	.442*	.411*	.301*	.428*	.330*
CoI-HbO₂	P		.000	.000	.001	.018	.001	.009
CoI-Hb	rs	.696*	1.000	.358*	.412*	.242	.412*	.290*
CoI-Hb	P	.000		.005	.001	.060	.001	.024
CoI-TOI	rs	.442*	.358*	1.000	.330*	.238	.336*	.307*
CoI-TOI	P	.000	.005		.012	.065	.008	.016
CoI-THI	rs	.411*	.412*	.330*	1.000	.262*	.279*	.223
CoI-THI	P	.001	.001	.012		.049	.035	.096
CoI-SaO₂	rs	.301*	.242	.238	.262*	1.000	.351*	.270*
CoI-SaO₂	P	.018	.060	.065	.049		.006	.035
CoI-MABP	rs	.428*	.412*	.336*	.279*	.351*	1.000	.681*
CoI-MABP	P	.001	.001	.008	.035	.006		.000
CoI-HR	rs	.330*	.290*	.307*	.223	.270*	.681*	1.000
CoI-HR	P	.009	.024	.016	.096	.035	.000

P values are not adjusted for multiple comparisons.

P < 0.05.

r_s: Spearman’s rank correlation coefficient.

Discussion

This study is the first to assess the complexity of cerebral NIRS signals in critically ill preterm infants. We applied MSE to study the complexity of brain and physiological systemic signals in a cohort of preterm infants in the first 24 h of life. Our results revealed an association between complexity of brain signals and outcome.

Several studies have suggested that decreased variability in physiological signals is a sign of failure in complex regulatory system. In perinatal medicine, the ‘silent pattern’ in foetal ECG is a well-known example of loss of ‘natural’ variability of vital signs and has been associated with poor neonatal outcome.³³ However, in clinical practice, there is no consensus on which method of intrapartum monitoring is the most effective.³⁴ Moreover, the continuous monitoring of neonatal vital signs does not take in consideration the non-stationarity characteristic of the biological signals. Entropy analysis has been extensively studied to investigate physiological systems. According to the ‘decomplexification’ theory of illness, loss of complexity is usually associated with pathological states like diseases or ageing. However, despite being an old concept, assessment of complexity of physiological signals in the neonatal unit is not performed routinely. Moreover, clinicians persist to correlate lack of variability as a sign of clinical stability. In our study, decreased complexity of brain signals was observed in those infants who developed IVH and who died, supporting the ‘decomplexification’ theory.^22,35 MSE of HbO₂ and Hb were consistently lower in all scale factors for groups of infants with IVH compared to those who did not have IVH, and for groups of infants who died compared to those who survived. These results are similar to recently published work using MSE to investigate the association between complexity of EEG and neurological outcome in a cohort of term infants with refractory neonatal seizures.²⁶ Lower values of MSE were observed in most scales in the group of infants who later developed epilepsy, compared to normal controls, and to those who did not develop epilepsy. In adults with traumatic brain injury, a decrease in entropy of ICP signals across all MSE scales was observed during the development of intracranial hypertension and loss of cerebrovascular reactivity. In this cohort, the CoI was significantly lower in those patients who had moderate and severe disability compared to those who had good outcome.²⁶

High complexity reflects strong, non-linear interactions between different healthy control systems responding to transient stressors or adapting to changes in the environment. The impairment of those regulatory mechanisms or the breakdown in their interactions is characterized by low complexity of various relevant physiological measurements. This may manifest itself in the total breakdown of (auto and cross) correlations of the measured signals (e.g. instantaneous change in heart rate in atrial fibrillation) or patterns that are overly regular and deterministic in nature (e.g. flat heart rate dynamic seen in chronic heart failure).^19,36 Assuming that the impairment of certain physiological control systems occurs before the onset of clinical symptoms of the developing pathology, or even assuming that such impairment could be the potential cause of diseases, a decrease in complexity could serve as an alert of complications like sepsis or poor outcome.^26,37 In our study, complexity of HbO₂ was an independent predictor of IVH and decreased complexity of brain signals within the first 24 h of life was observed in those infants who died later during admission, suggesting that the loss of interaction between control systems may be present even before the occurrence of brain injury and death. This could potentially, from early hours of life, divide those infants who are at more risk of poor outcome to those who are more likely to have less morbidity. While the ultimate goal of early analysis of physiological signals would be to identify the ‘high risk’ infants and develop therapies to prevent IVH and death, identifying those who are ‘healthy’ could help clinicians to be less invasive and differentiate between normal physiological to pathological adaptation.

The lower complexity of HbO₂, Hb and TOI in those infants who developed IVH may reflect, at least partially, the underlining loss of cerebrovascular reactivity that precedes the occurrence of IVH, as most infants developed IVH after the first 24 h of life. Changes in cerebrovascular reactivity and loss of autoregulation within the first hours of life have been associated with the incidence of IVH in preterm infants.^15,38,39 Several studies have shown that cerebral hypo-perfusion is associated with the pathophysiology of IVH and death.^3,40 In our study, most infants who died were extremely preterm and all of them had some degree of either antenatal or early postnatal hypo-perfusion or hypoxia. Even those infants who died due to NEC or sepsis were severely sick within the first days of life. Decreased oxygenation has been associated with death as shown in a large cohort of infants included in the BOOST-II trial. In this study, infants with lower oxygen saturation target had higher rate of NEC requiring surgery or causing death.⁴¹ In our cohort, target SaO₂ was kept the same (between 91%–95%) in all groups and there was no difference between mean SaO₂ and CoI-SaO₂ between those infants who died or survived or between those who developed IVH and who did not. However, cerebral hypo-perfusion and loss of cerebrovascular reactivity and may be associated with reduced complexity of cerebral NIRS signals. This hypothesis should be tested in a larger cohort where other factors related with changes in cerebral blood flow, like PaCO₂, are also taken into consideration.

In adults, early assessment of complexity of HR and HRV measured by MSE has been used to predict outcome in patients with traumatic brain injury and acute stroke.^21,42 In neonates, early monitoring of systemic blood flow and cerebral oxygenation has been used to understand the pathophysiology of brain injury in preterm infants.^3–5,43 Although the incidence of severe IVH and mortality in this population has decreased in the last 20 years, with improved survival, the prevalence remains high, as does the degree of disability and morbidity in survivors.⁴⁴ Preterm infants who develop IVH are at risk of higher morbidity during the neonatal period and later in childhood, including the development of post-haemorrhagic ventriculomegaly, recurrent hospital admissions and mortality. In addition, they may present long-term neurodevelopment and cognitive delays, resulting in increased workload for the health and educational services.^1,45 Studies using cerebral NIRS have attempted to define limits of cerebral autoregulation and thresholds of cerebral oxygenation, in order to prevent the occurrence of IVH and improve outcome.^7,46,47 However, there is no established threshold of cerebral perfusion and the continuous assessment of cerebrovascular reactivity and oxygenation has not yet been proved to change the outcome of preterm infants. In this context, the analysis of complexity of cerebral NIRS signals may offer a new approach – perhaps an alternative if not a complementary one to the cerebrovascular reactivity assessment – in order to understand the physiological mechanisms involved in the developing brain and to identify those infants who are more likely to be severely sick or even die during the neonatal period.

Complexity of brain signals demonstrated stronger association with outcome than complexity of systemic signals. In our study, the complexity of MABP was the only complexity of systemic physiological signals that correlated with outcome. Decreased complexity of MABP was associated with mortality but not with IVH. In clinical practice, the assessment of organ perfusion in preterm infants still relies on indirect measures of systemic blood flow, such as MABP. In adults, the regulation of cerebral perfusion is independent of systemic perfusion. The healthy adult brain has mechanisms that protect its tissue against ischaemic damage. The complexity of cerebral NIRS signals may represent a more direct measure of cerebral perfusion and regulation of blood flow than the complexity of systemic signals. On the other hand, the positive correlation between complexity of systemic and brain signals may suggest that factors associated with loss of cerebral and systemic regulatory mechanisms are related to each other. Further studies are needed to explore the significance of this relationship. Nevertheless, our findings support the idea that the monitoring of brain signals in addition to systemic signals would provide more comprehensive grounds for understanding the physiological mechanisms related to mortality and brain injury in preterm infants.

Limitations

MSE analysis requires a large data set to be able to estimate complexity in a plausible way. The minimum number of samples from time series that allows reasonable MSE calculation rises with the number of scales. It can be estimated that at least 2000 data samples are required to perform analysis up to scale 20.⁴⁸ In our study, the analysed time series had one sample every 10 s, so at least a 6-h long monitoring period was required to produce meaningful values. Furthermore, MSE analysis is relatively sensitive to outliers and artefacts. Removing a small percentage of them is allowed, however, if the structure of time series is grossly altered due to artefact removal, the value of entropy may also change.²⁴ The amount of data containing artefacts from our cohort was small, but the artefacts were manually and retrospectively removed. Real-time bedside analysis would require appropriate automatic artefacts detection and removal as well as a criterion for automatic rejection of data segments with too high artefacts content.

Our results of the mortality analysis have to be treated with some caution. Although the association between complexity and mortality was significant, we acknowledge that our sample was small and unbalanced, as only eight infants died. Moreover, most infants in our cohort died later during the neonatal period due to factors, which were not necessarily directly related to changes in cerebral perfusion within the first days of life, as sepsis and NEC. However, all infants who died had signs of antenatal or perinatal hypoxia and remained very sick throughout the neonatal admission. Further studies with a larger sample size should be conducted, in order to understand the factors associated with decreased complexity of brain signals in sick preterm infants. However, early recruitment should be taken in consideration to better understand the pathophysiology of brain injury. We aimed to recruit infants before 12 h of age, but this may have contributed to our limited our sample size, in view of the challenges of getting informed consent from parents soon after the delivery of a sick preterm infant.

Conclusion

We have demonstrated that the assessment of complexity of brain and systemic signals using MSE in preterm infants undergoing intensive care is feasible. Decreased complexity of cerebral signals was more strongly associated with outcome of IVH and mortality than the complexity of systemic signals. Furthermore, the complexity of brain signals was an independent predictor of brain lesions in this preterm population. Real-time bedside complexity signal analysis could represent a significant advance in the brain oriented care of critically ill newborn infants.

Footnotes

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: SPARKS charity (11CUH02); Cambridge Trust and Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) scholarship to da Costa CS; National Science Centre, Poland (no. DEC-2013/10/E/ST7/00117) scholarship to Placek MM and Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPQ) (203792/2014-9) post-doctoral scholarship for Cabella B.

Declaration of conflicting interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: The ICM+ software used for data monitoring and analysis is licensed by Cambridge Enterprise Limited (University of Cambridge); P.S. and M.C. have interest in a fraction of the licensing fee. The other authors declare no conflicts of interest.

Authors’ contributions

da Costa CS co-conceptualized and designed the study, was responsible for identifying infants who met the inclusion criteria and for getting parental consent, collected the data, collaborated with data analysis, drafted the initial manuscript and approved the final submitted manuscript. Placek MM co-conceptualized the study, was responsible for the Multiscale Entropy analysis of the data and collaborated with statistical analysis. Czosnyka M co-conceptualized and designed the study, run exploratory data analysis, supervised data analysis. Cabella B supervised data analysis. Kasprowicz M supervised the data analysis. Austin T co-conceptualized and designed the study, supervised data collection and data analysis, revised the results. Smielewski P designed the study, supervised data analysis and revised statistical analysis. All authors contributed to writing the manuscript and approved the submitted manuscript.

References

Mukerji

Shah

. Periventricular/intraventricular hemorrhage and neurodevelopmental outcomes: A meta-analysis. Pediatrics 2015; 136: 1132–1143.

Volpe

. Intracranial hemorrhage: Germinal matrix-intraventricular hemorrhage of the prmenature infant. In: Neurology of the newborn, 5th ed. Philadelphia: WB Saunders Co, 2008, pp. 517–573.

Kluckow

Evans

. Low superior vena cava flow and intraventricular haemorrhage in preterm infants. Arch Dis Child Fetal Neonatal Ed 2000; 82: F188–F194.

Alderliesten

Lemmers

Smarius

et al.

Cerebral oxygenation, extraction, and autoregulation in very preterm infants who develop peri-intraventricular hemorrhage. J Pediatr 2013; 162: 698–704 e2.

Noori

McCoy

Anderson

et al.

Changes in cardiac function and cerebral blood flow in relation to peri/intraventricular hemorrhage in extremely preterm infants. J Pediatr 2014; 164: 264–270 e1–e3.

Hyttel-Sorensen

Pellicer

Alderliesten

et al.

Cerebral near infrared spectroscopy oximetry in extremely preterm infants: phase II randomised clinical trial. BMJ 2015; 350: g7635.

Alderliesten

Dix

Baerts

et al.

Reference values of regional cerebral oxygen saturation during the first 3 days of life in preterm neonates. Pediatr Res 2016; 79: 55–64.

Susumu Suzuki

Ozaki

Kobayash

. A tissue oxygenation monitor using NIR spatially resolved spectroscopy. SPIE 1998; Vol. 3597.

Weerakkody

Czosnyka

Zweifel

et al.

Slow vasogenic fluctuations of intracranial pressure and cerebral near infrared spectroscopy–an observational study. Acta Neurochir (Wien) 2010; 152: 1763–1769.

10.

Meek

Tyszczuk

Elwell

et al.

Cerebral blood flow increases over the first three days of life in extremely preterm neonates. Archiv Dis Childhood Fetal Neonatal Ed 1998; 78: F33–F37.

11.

Munro

Walker

Barfield

. Hypotensive extremely low birth weight infants have reduced cerebral blood flow. Pediatrics 2004; 114: 1591–1596.

12.

Wong

Nakamura

Alexiou

et al.

Tissue oxygenation index measured using spatially resolved spectroscopy correlates with changes in cerebral blood flow in newborn lambs. Intensive Care Med 2009; 35: 1464–1470.

13.

Boudes

Gilbert

Leppert

et al.

Measurement of brain perfusion in newborns: pulsed arterial spin labeling (PASL) versus pseudo-continuous arterial spin labeling (pCASL). Neuroimage Clin 2014; 6: 126–133.

14.

Taussky

O'Neal

Daugherty

et al.

Validation of frontal near-infrared spectroscopy as noninvasive bedside monitoring for regional cerebral blood flow in brain-injured patients. Neurosurg Focus 2012; 32: E2.

15.

Wong

Leung

Austin

et al.

Impaired autoregulation in preterm infants identified by using spatially resolved spectroscopy. Pediatrics 2008; 121: e604–e611.

16.

Brady

Lee

Kibler

et al.

Continuous time-domain analysis of cerebrovascular autoregulation using near-infrared spectroscopy. Stroke 2007; 38: 2818–2825.

17.

da Costa

Czosnyka

Smielewski

et al.

Monitoring of cerebrovascular reactivity for determination of optimal blood pressure in preterm infants. J Pediatr 2015; 167: 86–91.

18.

Goldstein

Fiser

Kelly

et al.

Decomplexification in critical illness and injury: relationship between heart rate variability, severity of illness, and outcome. Crit Care Med 1998; 26: 352–357.

19.

Goldberger

Amaral

Hausdorff

et al.

Fractal dynamics in physiology: alterations with disease and aging. Proc Natl Acad Sci USA 2002; 99 Suppl 1: 2466–2472.

20.

Norris

Anderson

Jenkins

et al.

Heart rate multiscale entropy at three hours predicts hospital mortality in 3,154 trauma patients. Shock 2008; 30: 17–22.

21.

Riordan

Jr Norris

Jenkins

et al.

Early loss of heart rate complexity predicts mortality regardless of mechanism, anatomic location, or severity of injury in 2178 trauma patients. J Surg Res 2009; 156: 283–289.

22.

Pincus

. Approximate entropy as a measure of system complexity. Proc Natl Acad Sci USA 1991; 88: 2297–2301.

23.

Lake

Richman

Griffin

et al.

Sample entropy analysis of neonatal heart rate variability. Am J Physiol Regul Integr Comp Physiol 2002; 283: R789–R797.

24.

Costa

Goldberger

Peng

. Multiscale entropy analysis of biological signals. Phys Rev E Stat Nonlin Soft Matter Phys 2005; 71: 021906.

25.

Costa

Goldberger

Peng

. Multiscale entropy analysis of complex physiologic time series. Phys Rev Lett 2002; 89: 068102.

26.

Czosnyka

Shieh

et al.

Complexity of intracranial pressure correlates with outcome after traumatic brain injury. Brain 2012; 135: 2399–2408.

27.

Chen

Chang

et al.

Multiscale entropy of electroencephalogram as a potential predictor for the prognosis of neonatal seizures. PLoS One 2015; 10: e0144732.

28.

Smielewski

Czosnyka

Steiner

et al.

ICM+: software for on-line analysis of bedside monitoring data after severe head trauma. Acta Neurochir Suppl 2005; 95: 43–49.

29.

Parry

Tucker

Tarnow-Mordi

et al.

CRIB II: an update of the clinical risk index for babies score. Lancet 2003; 361: 1789–1791.

30.

Papile

Burstein

et al.

Incidence and evolution of subependymal and intraventricular hemorrhage: a study of infants with birth weights less than 1,500 gm. J Pediatr 1978; 92: 529–534.

31.

Kang

Costa

Priplata

et al.

Frailty and the degradation of complex balance dynamics during a dual-task protocol. J Gerontol A Biol Sci Med Sci 2009; 64: 1304–1311.

32.

Richman

Moorman

. Physiological time-series analysis using approximate entropy and sample entropy. Am J Physiol Heart Circ Physiol 2000; 278: H2039–H2049.

33.

Noren

Amer-Wahlin

Hagberg

et al.

Fetal electrocardiography in labor and neonatal outcome: data from the Swedish randomized controlled trial on intrapartum fetal monitoring. Am J Obstet Gynecol 2003; 188: 183–192.

34.

Bloom

Belfort

Saade

et al.

What we have learned about intrapartum fetal monitoring trials in the MFMU Network. Semin Perinatol 2016; 40: 307–317.

35.

Buchman

. Nonlinear dynamics, complex systems, and the pathobiology of critical illness. Curr Opin Crit Care 2004; 10: 378–382.

36.

Costa

Schnettler

Amorim-Costa

et al.

Complexity-loss in fetal heart rate dynamics during labor as a potential biomarker of acidemia. Early Hum Dev 2014; 90: 67–71.

37.

Moorman

Delos

Flower

et al.

Cardiovascular oscillations at the bedside: early diagnosis of neonatal sepsis using heart rate characteristics monitoring. Physiol Meas 2011; 32: 1821–1832.

38.

Pryds

Greisen

Lou

et al.

Heterogeneity of cerebral vasoreactivity in preterm infants supported by mechanical ventilation. J Pediatr 1989; 115: 638–645.

39.

Tsuji

Saul

du Plessis

et al.

Cerebral intravascular oxygenation correlates with mean arterial pressure in critically ill premature infants. Pediatrics 2000; 106: 625–632.

40.

Evans

Kluckow

. Early determinants of right and left ventricular output in ventilated preterm infants. Archiv Dis Childhood Fetal Neonatal Ed 1996; 74: F88–F94.

41.

Group

BIUKC

Group

BIAC

Group

BINZC

et al.

Oxygen saturation and outcomes in preterm infants. N Engl J Med 2013; 368: 2094–2104.

42.

Tang

Jen

Lin

et al.

Complexity of heart rate variability predicts outcome in intensive care unit admitted patients with acute stroke. J Neurol Neurosurg Psychiatry 2015; 86: 95–100.

43.

Kluckow

Evans

. Superior vena cava flow in newborn infants: a novel marker of systemic blood flow. Archiv Dis Childhood Fetal Neonatal Ed 2000; 82: F182–F187.

44.

Moore

Hennessy

Myles

et al.

Neurological and developmental outcome in extremely preterm children born in England in 1995 and 2006: the EPICure studies. BMJ 2012; 345: e7961.

45.

Costeloe

Hennessy

Haider

et al.

Short term outcomes after extreme preterm birth in England: comparison of two birth cohorts in 1995 and 2006 (the EPICure studies). BMJ 2012; 345: e7976.

46.

Gilmore

Stone

Shepard

et al.

Relationship between cerebrovascular dysautoregulation and arterial blood pressure in the premature infant. J Perinatol 2011; 31: 722–729.

47.

Mitra

Czosnyka

Smielewski

et al.

Heart rate passivity of cerebral tissue oxygenation is associated with predictors of poor outcome in preterm infants. Acta Paediatr 2014; 103: e374–e382.

48.

Angelini

Maestri

Marinazzo

et al.

Multiscale analysis of short term heart beat interval, arterial blood pressure, and instantaneous lung volume time series. Artif Intell Med 2007; 41: 237–250.

Complexity of brain signals is associated with outcome in preterm infants

Abstract

Keywords

Introduction

Methods

Ethics

Population

Measurements

Data pre-processing

MSE analysis

Statistical analysis

Results

Association between brain and systemic signals with outcome

Complexity of brain signals as an independent predictor of IVH

Discussion

Limitations

Conclusion

Footnotes

Funding

Declaration of conflicting interests

Authors’ contributions

References