EEG Spectral Power and Neurovascular Coupling as Early Predictors of Neurodevelopmental Outcome in Neonatal Hypoxic-Ischemic Encephalopathy

Abstract

Hypoxic ischemic encephalopathy (HIE) remains one of the leading causes of morbidity and mortality in newborns. There is a strong need to predict their neurodevelopmental impairment (NDI) within early hours of life, tailoring treatment strategies accordingly. This study aims to explore the discriminatory capabilities of electroencephalogram (EEG) delta power (DP) and total power (TP), along with neurovascular coupling (NVC) to predict NDI. The study evaluates the relationships of single biomarkers (DP, TP, and NVC) with NDI using univariate logistic regression models. The predictive accuracy of single (DP, TP, and NVC) and combination of biomarkers (DP+NVC, TP+NVC) on NDI is further assessed through the receiver operating characteristic (ROC) curve, with the area under the ROC curve (AUC). Utilizing EEG and near infrared spectroscopy (NIRS) data from 35 newborns with mild and moderate HIE, we found that a one-unit increase in DP or TP significantly lowered the odds of NDI. The combination of DP or TP and NVC is most effective in distinguishing newborns who may develop NDI. These findings suggest that continuous multimodal real-time neuromonitoring could offer valuable insights into HIE severity, aiding in predicting brain injury and NDI.

I. Introduction

Hypoxic ischemic encephalopathy (HIE) is a type of brain injury that occurs when the fetal brain lacks sufficient oxygen or blood supply, either before or during labor [1]. This condition affects an estimated 1.5 out of every 1000 live births globally [2] and stands as one of the primary causes of death worldwide in children under the age of five [3]. The long-term outcomes are dependent on the severity of encephalopathy [4], emphasizing the potential for improved outcome through an early diagnosis and intervention.

The Total Sarnat Score (TSS), derived from a modified neurological exam within the initial six hours of life or through serial exams within the first day of life, serves as a crucial metric for determining encephalopathy severity [5]. This assessment aids in tailoring treatment options such as therapeutic hypothermia (TH) for newborns with moderate to severe encephalopathy [6]. Additionally, a TSS of ≥ 5 predicts a likelihood of two-year neurodevelopmental impairment (NDI) [5]. However, the dynamic nature of injury within the early hours of life, the difficulty of clinically classifying mild encephalopathy, and the heterogeneity of encephalopathy severity grading systems necessitate advanced continuous neuromonitoring techniques, such as electroencephalogram (EEG) and cerebral tissue oxygen saturation (SctO2) using near-infrared spectroscopy (NIRS) at the bedside to assess severity of encephalopathy accurately for timely neuroprotective treatments. EEG spectral power [7] and neurovascular coupling (NVC) [8] obtained through amplitude integrated EEG (aEEG) and SctO2 predicted encephalopathy severity within the first day of life, as well as brain injury on magnetic resonance imaging (MRI).

While existing research has predominantly focused on individual biomarkers from a single modality for NDI, a notable gap exists in the literature concerning the utilization of biomarkers calculated from different modalities to predict NDI. Addressing this gap, this study aims to establish an early-life quantitative biomarker panel. We hypothesize that lower EEG spectral power and reduced NVC will predict an elevated risk of NDI, and the combination of these biomarker will yield higher predictive power. Additionally, biomarker panel may offer valuable insights for the development of crucial early intervention strategies for a better long-term neurodevelopmental outcome.

II. Methods

A. Study Participants

This prospective cohort study enrolled newborns (≥36 weeks’ gestation) at Parkland Hospital in Dallas, TX, between 2017 and 2019, who met the following criteria: 1) a history of an acute perinatal event (e.g., placental abruption, cord prolapse, decreased fetal heart rate), 2) umbilical cord arterial pH or arterial blood gas pH of ≤ 7.0 or base deficit ≥ 15 mmol/L within first hour after birth, and 3) signs of encephalopathy. Newborns with genetic or congenital condition, birth weight < 1800 grams, and/or head circumference < 30 cm were excluded as these factors could influence the study’s primary outcome to predict NDI. The study received an approval from the Institutional Review Board at University of Texas Southwestern Medical Center. Informed written consent was obtained from a parent of each newborn prior to enrollment in the research study.

The newborns were evaluated within 6 h after birth using a modified Sarnat exam by trained clinicians to determine the severity of encephalopathy. The exam assessed six categories: 1) level of consciousness, 2) spontaneous activity, 3) posture, 4) tone, 5) primitive reflexes (suck, moro), and 6) autonomic system (pupils, heart rate, respirations), with scores of normal (0), mild (1), moderate (2), or severe (3). The TSS was calculated by adding the scores for each category, ranging from 0 to 18, where 0 represents normal in all six categories, and 18 represents severe encephalopathy in all six categories [5]. The clinical grade of encephalopathy was determined by the number of Sarnat abnormalities, with classifications ranging from mild, to moderate or severe. In cases with an equal number of abnormalities, the grade was determined by the degree of reduced level of consciousness. TSS scores were obtained from electronics medical records.

For newborns with moderate and severe encephalopathy, whole-body therapeutic hypothermia (TH) was initiated within 6 h after birth following the National Institute of Child Health and Human Development (NICHD) protocol [9]. A servo-controlled blanket (Blanketrol II, Cincinnati Sub-Zero Products LLC, OH, USA) maintained their core body temperature at 33.5 °C for 72 hours, followed by gradual rewarming at 0.5°C per one to two hours over the next six hours. Newborns with mild encephalopathy received normothermia as per the standard of care. However, if their condition worsened to more severe encephalopathy or they experienced seizures within the first day of life, TH was initiated per the NICHD late hypothermia protocol, maintaining their core body temperature at 33.5°C for 96 hours [6].

EEG (Nihon Kohden America Inc., Irvine, CA) signals were acquired at a sampling of 256 Hz from eight scalp electrodes (Fz, C3, Cz, C4, P3, P4, O1, O2) that are referenced to Pz, placed according to the modified 10–20 montage for newborns [10]. EEG data acquisition was initiated as soon as newborns were admitted after standard clinical care. Cerebral tissue oxygen saturation (SctO2) was measured from each newborn forehead using the INVOS^™ 4100–5100 oximeter (Somanetics, Troy, MI) at a sampling rate of 0.209 Hz. The Component Neuromonitoring System (CNS) Monitor (Moberg Research, Inc., Ambler, PA, USA) served as the bedside interface for EEG, SctO₂, other physiological signals including core body temperature and blanket temperature from Blanketrol cooling device, and electrocardiography (EKG), mean arterial pressure (MAP), and peripheral capillary oxygen saturation (SpO₂) from the Philips IntelliVue MP70 (Philips Medical Systems). Acquisition of multimodal signal acquisition was initiated using CNS monitor as soon as informed written consent was obtained. These physiological signals were converted into MATLAB (MathWorks Inc., Natick, MA, USA) format using CNS Envision software (Moberg Research, Inc., Ambler, PA, USA) for offline data analysis.

Central and parietal region interhemispheric (C3–4, P3–P4) bipolar EEGs were chosen for analysis because they reflect watershed injury patterns on MRI in HIE [11–13]. NDI at two years of age was defined by a cognitive score < 85 on the Bayley Scales of Infant Toddler Development, Third Edition (BSID-III) [5, 14–16] administered by certified professionals during follow up clinic visit at 18–24 months of age.

B. EEG Spectral Power

Continuous EEG signals were bandpass filtered (0.3 to 20 Hz) with zero phase 4th order Butterworth filter. EEG data for cross cerebral parietal bipolar electrode (P3–P4) was obtained by taking difference between P3 and P4 for further analysis because they reflect watershed injury patterns on MRI in HIE [7, 17]. Bipolar EEG was segmented into one second segment to identify artifacts. Segments flagged as artifacts were excluded from further analysis based on two criteria: exceeding an absolute maximum of 300 μV (indicating high-amplitude noise) or exhibiting a standard deviation exceeding 50 μV or falling below 0.01 μV (suggesting noise artefacts or flat signals). Subsequently, the EEG was segmented into 10-minute epochs, and only artifact-free epochs were eligible for spectral power analysis. Using Welch’s periodogram method with a Hamming window of 10 seconds (providing a frequency resolution of 0.1 Hz) and no overlap between segments, the power spectral density (PSD) was calculated for each artifact-free 10-minute epoch. Each PSD underwent visual inspection to ensure the absence of spurious components like line noise subharmonics or other periodic artifacts. Delta power (0.5–4 Hz) and total power (0.5–20 Hz) were then quantified by integrating the respective areas under the PSD curve. For further analysis, the median values of delta (DP) and total power (TP) were calculated across all artifact-free non overlapping 10-minute segments within the first 3 hours of recording. This spectral power calculation procedure employed here, aligns with our previously published work [7].

C. Neurovascular Coupling

The bipolar cross-cerebral EEG was obtained by taking the difference between two cross cerebral central region electrodes (C3–C4) and then bandpass filtered (2–15 Hz) using a Parks-McClellan linear-phase FIR filter of order 800. The Washington University Neonatal EEG analysis toolbox [18] (WU-NEAT) was used to convert continuous bipolar EEG to aEEG. Filtered signal was rectified, followed by envelop extraction using a first-order low pass Butterworth zero phase filter with a cut-off frequency of 0.021. A gain of 1.161 was applied on an envelope, followed by segmenting envelope into 4.78 seconds epochs to match sampling rate of SctO2. In each epoch, the 90th and 10th percentile were defined as the upper terminal point (UTP) and lower terminal point (ULTP), respectively, to represent each 4.78 s epoch. Next, a simple overlapping moving average with a step of 4.78 s was performed over every 3 upper and lower adjacent points to obtain smoothed envelope. Signal margins were obtained by temporal concatenation of all UTP and LTP, respectively. A linear scale was used from 0–10 μV, and a semi-log scale was used from 10–100 μV, following the conventional aEEG presentation format. The time-varying tracings of the differences between upper and lower margin amplitudes were used for aEEG. Outliers in the aEEG and SctO₂ were identified and interpolated using neighboring data points and visually inspected for accuracy as previously published methods [8, 19]. Artifact-free aEEG and SctO₂ were used to calculate wavelet transform coherence analysis (WTC) to quantify neurovascular coupling using a MATLAB-based software package ‘wtc’ (https://www.mathworks.com/matlabcentral/fileexchange/47985-cross-wavelet-and-wavelet-coherence) [20]. WTC is a time-frequency metric that characterizes the squared cross-wavelet coherence (R² ranges from 0 to 1) and the relative phase (ΔØ ranges from -pi to pi) between two time series signals at multiple time scales/frequencies, without any prior assumptions of linearity and stationarity through the entire time duration. Statistically significant R² and ΔØ pixels were identified in time-frequency map against simulated background noise (p value < 0.05) described in ‘wtc’ package. For each frequency component, the NVC between aEEG and SctO₂ was calculated as the percentage of significant pixels over total pixels across the entire time duration, and this process was repeated for all phase ranges. The pixels influenced by the edge effects of the cone of influence were not considered for the analysis.

D. Statistical Analysis

The demographic and clinical characteristics of newborns were summarized using descriptive statistics, stratified according to neurodevelopmental outcome, specifically classified as either normal or abnormal (NDI). Continuous variables were presented as means with standard deviations or medians with interquartile ranges (IQRs), while categorical variables were presented with counts with percentages. Statistical comparisons between normal and abnormal groups were conducted using Student’s t-test or Wilcoxon rank-sum test for continuous variables, and χ² test or Fisher’s exact test for categorical variables.

The relationships of single biomarkers (DP, TP, and NVC) with NDI were assessed using univariate logistic regression models. The receiver operating characteristic (ROC) curve, with the area under the ROC curve (AUC) was used to evaluate the prediction ability of single (DP, TP, and NVC) and combination of biomarkers (DP+NVC, TP+NVC) on NDI. The optimal cut-off values for biomarkers were obtained using the Youden method [21] to distinguish outcome. The cut-off values for single biomarkers are based on raw values, while those for the combined biomarkers are based on fitted values. Results were reported as odds ratios (ORs) with 95% confidence intervals in logistic regression models. A 2-tailed p value less than 0.05 was considered the threshold for statistical significance. All statistical analyses were performed using R version 4.2.2.

III. Results

This study enrolled 46 newborns exhibiting encephalopathy symptoms within the first day of life. According to the modified Sarnat exam, 26 were classified as mild, 18 as moderate, and 2 as severe. Standard whole-body TH was initiated to newborns in the moderate and severe categories within the first six hours of life. Seven out of 26 mild newborns developed worsening symptoms within 24 hours. These newborns were reclassified as mild-moderate and received TH per the NICHD late hypothermia protocol due to worsening neurological exam or seizures.

Eight infants (6 mild, 1 mild-moderate, 1 moderate) were lost to the two-year follow-up. Additionally, two infants in the severe group died within the first week of life due to redirection of the care and one infant did not have EEG and SctO2 data on the first day of life. This resulted in a final distribution of 12 newborns in the mild group and 23 in the non-mild group (mild-moderate and moderate). Among the remaining 35 newborns, 16 (6 mild, 10 non-mild) had NDI and 19 (6 mild, 13 non-mild) had a favorable outcome.

Table 1 provides neonatal and maternal characteristics for the overall cohort and for infants with and without NDI. None of the maternal or neonatal characteristics were significantly different from infants with or without NDI. Univariate logistic regression analysis revealed that higher DP and (OR 0.991, CI: 0.982–1.00, P-value = 0.039) and TP (OR 0.991, CI: 0.983–0.999, P-value = 0.037) were associated with decreased odds of NDI. In contrast, NVC did not show any statistically significant association on NDI (OR 0.996, CI: 0.911–1.025, P=0.253).

Table 1:

Neonatal and maternal characteristics of the cohort

Characteristics	Overall	Cognitive outcome
		Normal	Abnormal	P value
Total N	35	19	16	----
Male: N (%)	20 (57)	12 (63)	8 (50)	0.659
Gestational Age (weeks), median [IQR]	39 [38, 40]	39 [37, 40]	39 [39, 40]	0.396
Birth Weight (kg), mean (SD)	3.3 (0.7)	3.4 (0.8)	3.2 (0.6)	0.384
Apgar 1 minute, median [IQR]	2 [1, 4]	2 [1, 4]	2 [1, 3]	0.768
Apgar 5 minute, median [IQR]	6 [4, 7]	4 [3, 7]	6 [5, 7]	0.193
Umbilical Cord Gas pH, mean (SD)	7 (0.1)	7(0.2)	7 (0.1)	0.889
Base Deficit, median [IQR]	16.3 [13.3, 18.6]	15.5 [14.7, 17.6]	17.0 [12.4, 19.6]	0.828
Abnormal MRI (Global): N (%)	9 (26)	6 (32)	3 (19)	0.461
Maternal Race/Ethnicity: N (%)
Caucasian non-Hispanic	1 (3)	1 (5)	0 (0)	0.271
Black non-Hispanic	6 (17)	5 (27)	1 (6)
Hispanic	26 (74)	12 (63)	14 (88)
Other non-Hispanic	2 (6)	1 (5)	1 (6)
Delivery Mode: N (%)
C/S	18 (51)	13 (68)	5 (31)	0.064
Vaginal	17 (49)	6 (32)	11 (69)
Maternal Risk Factors: N (%)
Hypertension	7 (20)	4 (21)	3 (19)	> 0.999
Diabetes	4 (11)	2 (11)	2 (13)	> 0.999
Pre-eclampsia	11 (31)	8 (42)	3 (19)	0.167
Labor Complications: N (%)
Meconium	10 (29)	5 (26)	5 (31)	> 0.999
Umbilical Cord Prolapse	1 (3)	1 (5)	0 (0)	> 0.999
Placental Abruption	2 (6)	2 (11)	0 (0)	0.489
Uterine Rupture	3 (9)	3 (16)	0 (0)	0.234
Maternal Chorioamnionitis	12 (34)	5 (26)	7 (44)	0.468
Placental Chorioamnionitis	20 (57)	10 (53)	10 (63)	0.807
DOL at discharge, median [IQR]	11 [8, 20]	11 [8, 20]	12 [8, 19]	0.855

Abbreviations: IQR = Interquartile range, SD = Standard deviation, DOL = days of life

When the normality assumption was violated, and the Wilcoxon rank sum test was used to assess statistical significance, median and interquartile range (IQR) were reported. * Indicates statistical significance (p < 0.05)

Figure 1 shows the AUC and ROC results for NDI. DP (N=33) cut-off value of 101, with an AUC of 0.713 (95% CI: 0.534–0.893), sensitivity of 0.688, specificity of 0.706, positive predictive value (PPV) of 0.688, and negative predictive value (NPV) of 0.706 best distinguished NDI. Similarly, TP (N=33) of 119, an AUC of 0.721 (95% CI: 0.543–0.898), sensitivity of 0.688, specificity of 0.706, PPV of 0.688, and NPV of 0.706 distinguished between with and without NDI. NVC (N=31) demonstrated an optimal cut-off value of 12, yielding an AUC of 0.579 (95% CI: 0.368–0.791), sensitivity of 0.563, specificity of 0.667, PPV of 0.643, and NPV of 0.588 to predict NDI. When considering combined biomarkers, the combination of DP and NVC (N=27) at an optimal fitted cut-off value of 0.479 achieved an AUC of 0.720 (95% CI: 0.513–0.926), sensitivity of 0.857, specificity of 0.615, PPV of 0.706, and NPV of 0.800. Similarly, the combination of TP and NVC (N=27) with an optimal fitted cut-off value of 0.540 resulted in an AUC of 0.725 (95% CI: 0.520–0.930), sensitivity of 0.786, specificity of 0.692, PPV of 0.733, and NPV of 0.750 predicted NDI. These findings provide insights into the discriminatory abilities of individual and combined biomarkers in the studied cohort to predict NDI.

Figure 1:

Receiver operating characteristic (ROC) curves for neurodevelopmental impairment (NDI) for a single biomarker (A) and combined biomarkers (B). NDI at two years of age was defined by a cognitive score < 85 on the Bayley Scales of Infant Toddler Development, Third Edition (BSID-III). EEG delta power (DP) and total power (TP) were calculated at the cross-cerebral parietal bipolar electrode (P3–P4). Neurovascular coupling was quantified as wavelet coherence between amplitude-integrated EEG of cross-cerebral central region electrodes (C3–C4), and cerebral tissue oxygen saturation (SctO2). The gray diagonal line corresponded to random chance. The area under the curve (AUC) and its 95% confidence interval are shown in the legend. The optimal cut-off values for the biomarkers, sensitivity, and specificity are shown with dashed lines.

IV. Discussion

In this study, we extend our previous research, emphasizing on the effectiveness of EEG spectral power and NVC as biomarkers for predicting HIE severity and brain injury. Our current objective is to validate whether these biomarkers, either individually or in combination, can predict 2-year neurodevelopmental outcomes based on metrics calculated within early hour of life.

For the first time, our study provided evidence that the combined use of EEG spectral power and NVC within the first day of life surpasses the predictive capabilities of each biomarker alone predicting NDI. The clinical implementation of this study is especially pertinent to clinicians to distinguish between mild and non-mild HIE in newborns to initiate targeted neuroprotective therapies. Our specific focus is newborns with mild HIE, as they are typically not considered candidates for TH in most centers. Interestingly, we found that approximately 40% of newborns with mild HIE who did not receive TH developed NDI. Some centers provide TH for newborns with mild HIE even though they may not all require TH. Through our exploratory analysis, we demonstrate that the combination of biomarkers can effectively differentiate between newborns who may or may not develop NDI. DP and TP have been utilized to predict encephalopathy severity (mild vs. combined mild-moderate, moderate, severe categories) using first three hours of EEG recording [7]. DP has also been used to predict severity of brain injury on MRI (normal or mild vs. death or severe) as early as 9 hours of life in newborns with moderate or severe HIE [17]. TP was able to identify infants who are at risk for seizures within the first hour of EEG recording [13]. Normalized DP, among the several EEG features, led to an area under the ROC curve of 75.1%. Additionally, combination of nine EEG features and one clinical feature resulted in an area under the ROC of 87% and an accuracy of 84% for predicting neurodevelopmental outcome [22], which highlights the importance of multimodal prediction model. However, in their study, biomarkers derived from heart rate did not contribute to the best performance.

The WTC analysis between aEEG and SctO2 to quantify NVC demonstrated an AUC of 0.808, using a 10% NVC cutoff was able to predict brain injury on MRI [8]. The sensitivity, specificity, PPV, and NPV for predicting brain injury on MRI were 69%, 90%, 94%, and 52%, respectively. Deriving aEEG from EEG varies across algorithms used, but it does not influence the NVC calculation [19]. It is evident from previous studies that DP, TP, and NVC have been used to predict encephalopathy severity and brain injury.

In this current study, the PPV of the combined biomarker model (DP+NVC, TP+NVC) was marginally higher than that of individual biomarkers (DP, TP, NVC), suggesting a greater likelihood that a positive prediction of NDI is accurate. Similarly, the NPV also showed a slight improvement compared to individual biomarkers, indicating an increased probability of accurate negative predictions. This elevated PPV is crucial for the identification of newborns at risk for NDI, particularly for those with mild HIE who may require TH within the first 6 hours of life when its effectiveness is optimal. A higher NPV of combined model underscores the model’s ability to accurately rule out negative cases, minimizing the occurrence of false negatives, and aiding in the identification of mild newborns who may not require TH. The combined biomarker model exhibits a higher sensitivity compared to single biomarkers, suggesting its potential to identify newborns at risk for NDI more effectively. While specificity values are slightly higher for single biomarkers, indicating their utility in ruling out NDI in newborns with negative test results, the comprehensive performance metrics highlight the strengths of the combined biomarker approach in predicting and identifying newborns at risk for NDI. The AUC results for DP and TP were almost similar. This could be attributed to DP being predominant in newborns, as Kota et al. reported that DP constitutes 87% ± 11% of TP in their earlier study [7]. The cut-off values derived from the AUC analysis may assist clinicians at the bed side for timely neuroprotective interventions.

While the potential of combination of EEG spectral power and NVC in predicting NDI is clear, several limitations necessitate future advancements. Especially for the combined biomarker model with 27 subjects, there is a lack of adequate statistical power due to the issue of overfitting. The exploratory nature of this small cohort study necessitates larger confirmatory trials with an expanded cohort, potentially including COOLPRIME trial (https://clinicaltrials.gov/ct2/show/NCT04621279). Additionally, the temporal mismatch between EEG power’s three-hour window and NVC’s 20-hour window requires addressing. Developing a new method to calculate NVC for shorter intervals could overcome the disadvantage of WTC’s longer data acquisition and enhance its clinical utility. Therefore, future research should prioritize transitioning from offline analysis to real-time monitoring systems utilizing EEG spectral power and NVC. This shift holds immense promise for transformative advancements in HIE management, enabling:

• Enhanced, continuous detection of subtle brain function changes, especially in mild cases.

• Real-time seizure prediction algorithms for improved patient safety and intervention timing.

• Data-driven optimization of early intervention strategies based on continuous feedback from these biomarkers.

By addressing the limitations and capitalizing on real-time monitoring, we could unlock the full potential of these promising biomarkers and revolutionize the management of HIE in early hours of life for better neurodevelopmental outcome.

Conclusion

This study highlights the importance of implementing multimodal neuromonitoring and integrating quantitative biomarkers at the bedside during the early hours of life. By combining different modalities, these approaches provide complementary objective metrics regarding brain health. When these modalities are integrated, they contribute to a more accurate prediction of the severity of encephalopathy, brain injury on MRI, and neurodevelopmental outcomes within the early hours of life. Given the challenges associated with clinical classification of mild HIE, dynamic nature of injury evolution within early hours of life, and the variability in clinical exams, having quantitative multimodal biomarkers at the bedside can assist clinicians in promptly and accurately classifying the severity of HIE and predicting neurodevelopmental outcomes. This, in turn, facilitates timely decisions on neuroprotective strategies, ultimately aiming to improve long-term neurodevelopmental outcomes.