A Pilot Study on Dynamic Multimodal Neuromonitoring for Predicting Neurodevelopmental Outcomes in Neonatal Hypoxic-Ischemic Encephalopathy

Neonatal hypoxic-ischemic encephalopathy (HIE) is one of the leading causes of mortality and morbidity worldwide with an estimate of one to two million cases annually [1, 2]. The severity of HIE is assessed through neurological exams to determine the eligibility for therapeutic hypothermia (TH), which is effective if initiated within narrow window of 6 h of life [3]. The short therapeutic window exacerbates the challenge of classifying insult severity using neurological exams, which are further complicated by variability across exams [4], and the fact that it may not be practical to repeat them to monitor the dynamic progression of encephalopathy. The full impact on neurodevelopmental outcomes may remain unclear until later in childhood [5], as we currently lack sensitive biomarkers readily available at the bedside within the early hours of life to identify newborns at risk for adverse outcomes.

Physiological signals such as electroencephalogram (EEG) and regional cerebral tissue oxygenation (SctO₂) measured via near-infrared spectroscopy (NIRS) are commonly used for monitoring dynamic evolution of encephalopathy [6]. EEG is employed to assess brain function and detect electrographic seizures, while SctO₂ evaluates cerebral perfusion. Previous studies have demonstrated that quantitative metrics of EEG, such as delta power (DP) or total power (TP) measured within the first day of life, are associated with the severity of HIE [7] and can predict short-term outcomes, such as brain injury observed on magnetic resonance imaging (MRI) [8]. Similarly, SctO₂ values have been found to predict the severity and extent of brain injury [9]. Furthermore, neurovascular coupling (NVC), measured as wavelet coherence between amplitude-integrated EEG (aEEG) and SctO₂ also predict the severity of HIE and the extent of brain injury [10, 11]. However, existing research has predominantly focused on single modality biomarkers to predict severity of HIE, short-term and long-term outcomes. There remains a critical need to explore multimodal physiological biomarkers to enhance predictive capability of neurodevelopmental outcome.

The goal of this study was to establish quantitative neuromonitoring biomarkers during the first day of life to predict neurodevelopmental impairment (NDI) at 2 years of age, defined as death or a cognitive score below 85 on the Bayley scales of Infant Toddler Development III (BSID). We hypothesize that combining multimodal biomarkers (DP and NVC) within the first day of life, along with the Total Sarnat Score (TSS) [12] obtained within the first 6 h, will enhance the accuracy of predicting NDI compared to using individual biomarkers alone. These biomarkers could be utilized at the bedside to identify newborns at risk for NDI, enabling timely individualized neuroprotective interventions for those at higher risk while sparing those at lower risk from potential adverse side effects of interventions.

Full-term newborns (≥36 weeks of gestational age) diagnosed with HIE and admitted to the neonatal intensive care unit at Parkland Health, Dallas, between 2017 and 2019 were recruited for this prospective cohort study. The study (STU 022015-104) was approved by the institutional review board (IRB) at the University of Texas Southwestern Medical Center, and parental consent was obtained before enrollment. The inclusion criteria were as follows: (1) a history of an acute perinatal event (e.g., placental abruption, cord prolapse, or decreased fetal heart rate), (2) umbilical cord arterial pH or arterial blood gas pH at <1 h postnatal age of ≤7.0 or a base deficit ≥15 mmol/L, and (3) clinical signs of encephalopathy. Exclusion criteria included any genetic or congenital conditions, a head circumference <30 cm, or a birth weight <1,800 g, as these factors could influence study outcomes.

A modified Sarnat exam [12] was conducted by trained and certified attending physicians to assess the severity of HIE using the NICHD criteria [13]. The exam evaluated six categories: (1) level of consciousness, (2) spontaneous activity, (3) posture, (4) muscle tone, (5) primitive reflexes (such as suck and Moro reflexes), and (6) autonomic function (pupils, heart rate, and respirations). Each category was scored as normal (0), mild (1), moderate (2), or severe (3). The TSS was calculated by adding the scores from all six categories, which ranges from 0 (normal in all categories) to 18 (severe in all categories). The clinical grade of HIE was classified as mild, moderate, or severe based on the number of abnormalities in the Sarnat exam. If there was an equal number of abnormalities, the grade was determined by the level of consciousness. The TSS for enrolled study participants was obtained from electronic health records [12, 14].

Whole-body TH was initiated within 6 h of life for newborns diagnosed with moderate to severe HIE as a standard neuroprotective treatment [3]. The servo-controlled blanket (Blanketrol II, Cincinnati Sub-Zero, OH, USA) was used for TH, maintaining a target core body temperature of 33.5°C for 72 h, followed by rewarming at a rate of 0.5°C per hour until normothermia is achieved. Newborns with mild HIE did not receive TH. However, those initially classified as mild within the first 6 h of life who later developed seizures or progressed to more severe encephalopathy within the first day of life were treated with TH following the NICHD late hypothermia protocol [13]. NDI was defined as a death or cognitive score <85 [15] on BSID at age 18–22 months performed by certified professionals.

Noninvasive EEG electrodes (Fz, C3, Cz, C4, P3, P4, O1, and O2) were placed on newborn’s scalp according to the modified 10–20 system. EEG signals were referenced to the mid parietal electrode (Pz) during acquisition and sampled at 256 Hz (Nihon Kohden, CA, USA). SctO₂ was monitored on the newborns’ forehead using the INVOS™ 4,100–5,100 oximeter (Somanetics, Troy, MI) at a sampling rate of 0.21 Hz. A Moberg Component Neuromonitoring System monitor (Moberg Research, Inc., Ambler, PA, USA) was used to interface all physiological signals, and the data were subsequently downloaded for offline processing. The analysis in this study was limited to cross-cerebral central and parietal region EEG electrodes, as they represent watershed injury patterns [7, 8, 16, 17].

The continuous EEG signals were band-pass-filtered (0.3–20 Hz) using a fourth-order zero-phase Butterworth filter. Cross-cerebral parietal bipolar EEG signal was derived by calculating the difference between electrodes P3 and P4. The bipolar EEG data were segmented into nonoverlapping one-second epochs. Artifacts were identified based on absolute amplitude (>300 µV) or standard deviation (<0.01 µV or >50 µV) criteria and were excluded from the analysis. Power spectral density (PSD) was calculated for each artifact-free nonoverlapping 10-min segments using the Welch periodogram method with a 10-s Hamming window and no overlap. PSD curves were manually inspected to exclude segments with subharmonics of line noise or other periodic noise. DP was defined as the area under the PSD curve within the 0.5–4 Hz frequency range. The median DP across artifact-free 10-min segments within the first 3 h of EEG recording was calculated and used for statistical analysis to predict NDI. This methodology has been described in detail [7]. Data processing was performed using custom scripts in MATLAB (MathWorks, Natick, MA).

Cross-cerebral central region bipolar EEG was obtained by calculating the difference between electrodes C3 and C4. The bipolar EEG signal was band-pass-filtered (2–15 Hz) using an asymmetric Park-McClellan linear phase FIR filter and then converted to aEEG using the Washington University Neonatal EEG Analysis Toolbox (WU-NEAT) [18]. Artifact spikes in the aEEG data were identified and interpolated using neighboring data points. Both aEEG and SctO₂ signals were first detrended and then processed using a threshold denoising method to reduce speckle noise. Artifacts were identified and removed using linear interpolation, followed by a second-order polynomial detrending to eliminate slow drifts in both signals.

Wavelet coherence analysis between aEEG and SctO₂ signals was performed using a MATLAB-based software package, as described in our prior studies [11, 19, 20]. This analysis is conducted in the time-frequency domain, calculating the squared cross-wavelet coherence (R²) and relative phase (Δϕ) between aEEG and SctO₂ signals across multiple time scales, without assuming linearity or stationarity for the first 20 h of recording. R² is a localized correlation coefficient (ranging from 0 to 1), reflecting the coherence between aEEG and SctO₂ signals in the time-frequency domain. The statistical significance of R² was determined using Monte Carlo simulation with a 95% confidence interval, identifying statistically significant coherent pixels (p < 0.05). Pixels influenced by the edge effects of the cone of influence were not considered for the analysis. NVC was quantified as the percentage of significant pixels relative to total pixels within a wavelet scale of 64–250 min, corresponding to the very low-frequency range of 0.06–0.2 mHz, as reported in prior study [11]. Figure 1 presents the multimodal neuromonitoring approach, incorporating the TSS from the Sarnat exam, NVC, and EEG DP within the first day of life to predict neurodevelopmental outcomes.

For descriptive statistics of maternal and neonatal characteristics, continuous variables with a normal distribution are presented as mean and standard deviation, with significance tested using one-way analysis of variance (ANOVA). For non-normally distributed continuous variables, median and interquartile ranges (IQR, 25th and 75th percentiles) are reported, and significance was assessed using the Kruskal-Wallis test. The normality assumption was evaluated using the Shapiro-Wilk test. Categorical variables are expressed as counts and percentages, with statistical significance analyzed using either the chi-square test or Fisher’s exact test, as appropriate. A two-sided p value of less than 0.05 was considered statistically significant for all tests.

Logistic regression models were employed to assess associations between individual biomarkers (DP, TSS, and NVC) and NDI, and results were reported as odds ratios (ORs) with 95% confidence intervals (CIs). Receiver operating characteristic (ROC) curve analysis was used to determine the predictive accuracy of individual biomarkers (TSS, DP, and NVC) and combined biomarkers (TSS+DP, TSS+NVC, DP+NVC, TSS+DP+NVC) by calculating the area under the curve (AUC). Optimal cutoff values for the biomarkers were determined using the Youden’s index to differentiate outcomes. Cutoff values for individual biomarkers were based on their raw measurements, while those for combined biomarkers were derived from the fitted values, which represent a weighted combination of individual biomarkers calculated through a logistic regression model.

A Net Reclassification Index (NRI) analysis was conducted to assess the predictive performance of three baseline models TSS, DP, and NVC. Each of these baseline models was compared to models that incorporated additional biomarkers, specifically TSS+DP, TSS+NVC, and TSS+DP+NVC. To ensure a data-driven approach to classification, two cutoff strategies were applied: (1) Youden’s index, which determines the optimal threshold based on ROC analysis, and (2) quantile-based cutoffs, using the 25th and 75th percentiles of predicted probabilities to define risk categories. All statistical analyses were conducted using R version 4.4.1.

Forty-six newborns with symptoms of HIE were enrolled during the 3-year study period and after informed consent was obtained. Based on the Sarnat exam within the first 6 h of life, 26 were classified as mild, 18 as moderate, and 2 as severe. Newborns with moderate and severe HIE received TH. Additionally, 7 of the 26 newborns initially classified with mild HIE progressed in severity or developed seizures within the first day of life and subsequently received TH following the NICHD protocol [13]. This resulted in a final distribution of 19 mild, 25 moderate, and 2 severe grades. Maternal and neonatal characteristics for the overall cohort and each HIE grade are presented in Table 1. The majority of the cohort was male (63%) [21], with a higher proportion in the mild (74%) and severe (100%) groups. The median gestational age was 39 weeks overall, except in the severe group, where it was 37 weeks. The mean birth weight was consistent across groups at approximately 3.3 kg, with slightly higher values observed in the mild and severe groups. The median 1-min and 5-min Apgar scores were significantly lower in the severe group. Umbilical cord gas values were similar across all three groups, while the base deficit was significantly higher in the severe group. The maternal race was predominantly Hispanic (67%), with a similar distribution across all groups. Cesarean delivery was the most common mode of birth, occurring in 63% of cases. Maternal risk factors included hypertension (22%), pre-eclampsia (28%), and chorioamnionitis (28–52%), with placental chorioamnionitis being the most prevalent. Brain MRI was performed at an average of 4.8 ± 2.3 days of life across the entire cohort. Specifically, newborns in the mild group underwent MRI at 3.5 ± 1 days, while those in the moderate group had MRI at 5.8 ± 2.7 days of life. Abnormal MRI findings consisting of watershed and or basal ganglia injury according to Barkovich classification [22] were observed in 26% of the neonates, with a similar prevalence in the mild (21%) and moderate (24%) groups. The median day of life (DOL) at discharge was 9 days [IQR: 6, 17] for the overall cohort, with a median of 6 days [IQR: 5, 7] for the mild group, 6 days [IQR: 5, 6] for the severe group, and a significantly longer duration of 14 days [IQR: 10, 20] for the moderate group. Both neonates with severe HIE died following redirection of care.

Neuromonitoring was initiated at the 12 ± 6 h of life for all newborns, continuing until 40 ± 11 h of life in the mild group and until 94 ± 29 h of life in newborns who received TH. For multimodal biomarker analysis, the first 3 h of EEG data collected within the first day of life were used to calculate DP, while the first 20 h of aEEG and SctO₂ data were utilized to assess NVC. No seizures or administration of seizure medications were observed during this analysis period. On average, 4.75% of the time series data per subject was interpolated during artifact removal, with all interpolated segments visually inspected to ensure quality.

Eight infants were lost to follow-up (6 mild, 2 moderate) at 2 years of age, and one died from unrelated cause to asphyxia. Both neonates with severe encephalopathy died following redirection of care. Eighteen out of the remaining 37 infants experienced death or disability at 22 ± 2 months of age. The composite outcome of death or disability occurred in 18 infants (49%) overall, including 6 (50%) in the mild group, 10 (44%) in the moderate group, and 2 (100%) in the severe group, with no significant differences among groups (p = 0.379). There was no significant difference in BSID cognitive scores between the mild and moderate cohorts (median [IQR]: 83 [75–95] for mild vs. 85 [80–90] for moderate, p = 0.874). Figure 2 compares NVC in two newborns with moderate HIE at birth: one with NDI and another without NDI. Figure 2a shows an example of simultaneously recorded aEEG, SctO₂, and NVC for a newborn without NDI, demonstrating 14% significant coherence across a wavelet scale of 64–250 min. In contrast, Figure 2b illustrates data for a newborn with NDI, with only 4% significant coherence.

Univariate logistic regression analysis showed that higher DP was associated with a reduced odds of NDI (OR 0.991, 95% CI: 0.983–0.999, p = 0.034). In contrast, TSS (OR 1.113, 95% CI: 0.949–1.306, p = 0.188) and NVC (OR 0.958, 95% CI: 0.902–1.018, p = 0.164) did not demonstrate a statistically significant association with NDI. Figure 3 illustrates the predictive performance of single and combined biomarkers using ROC curves, as well as sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Among the single biomarkers, TSS showed an AUC of 0.618 (95% CI: 0.440–0.797) at a threshold of 4, demonstrating high sensitivity (0.950) but low specificity (0.263). DP had a higher AUC of 0.708 (95% CI: 0.534–0.884) at a threshold of 101, with improved specificity (0.706) but moderate sensitivity (0.667). NVC showed an AUC of 0.622 (95% CI: 0.423–0.822) at a threshold of 12, with sensitivity and specificity of 0.611 and 0.667, respectively.

Combining biomarkers enhanced predictive performance compared to individual biomarkers. The combination of TSS and DP resulted in an AUC of 0.709 (95% CI: 0.533–0.886), with high sensitivity (0.944) and improved specificity (0.471). The combination of TSS and NVC reached an AUC of 0.653 (95% CI: 0.429–0.840) achieving sensitivity of 1,000 and specificity of 0.286. The combination of DP and NVC produced an AUC of 0.712 (95% CI: 0.508–0.915), with sensitivity of 0.875, specificity of 0.615, PPV of 0.737, and NPV of 0.800. Finally, the combination of all three biomarkers (TSS, DP, and NVC) yielded the highest AUC of 0.755 (95% CI: 0.569–0.941), with sensitivity of 0.750, specificity of 0.769, PPV of 0.800, and NPV of 0.714, indicating improved predictive accuracy for outcomes.

The NRI analysis showed that the combined (TSS + DP + NVC) model yielded the highest NRI value (0.558), indicating the strongest improvement in risk classification. When using DP as the baseline model, the DP + TSS + NVC combination achieved an NRI estimate of 0.418. Similarly, for the NVC baseline model, the NVC + DP combination resulted in an NRI estimate of 0.355. These findings highlight the superior performance of combined models, with TSS + DP + NVC being the most effective. Furthermore, quantile-based cutoffs resulted in more meaningful reclassification effects, while Youden’s index did not produce significant NRI changes. These results highlight the benefit of integrating multimodal biomarkers for improved risk prediction.

The pursuit of multimodal quantitative physiological signal biomarkers at the bedside is essential for effectively monitoring encephalopathy progression in newborns with HIE. This approach enables early identification of newborns at risk of NDI and facilitates timely neuroprotective interventions. Our capacity to differentiate mild from moderate HIE based solely on the early neurological examination within 6 h of life is very limited. The primary finding of this study is that our clinical assessment of encephalopathy as mild or moderate is empirical at best, despite training, fidelity, and certified examiners, it results in same neurodevelopmental outcomes at 2 years. This is due to the dynamic evolving nature of the clinical encephalopathy necessitating frequent examinations around the clock, which are not always feasible in busy intensive care units. Integrating the TSS using the worse score in 6 h with dynamic physiological biomarkers such as DP and NVC provides superior predictive accuracy for NDI compared to individual biomarkers or two-marker combinations. These results emphasize the importance of a quantitative approach in understanding neurophysiological mechanisms related to brain function and perfusion, allowing for a more accurate monitoring the dynamic evolution of encephalopathy, rather than relying solely on neurological exams during the early hours of life. This study aimed to explore multimodal physiological biomarkers to address the clinical challenge of accurately distinguishing between newborns at risk for NDI and those likely to have normal outcomes, enabling more targeted interventions.

Chalak et al. [12] reported neurodevelopmental outcomes at 18–22 months of age for infants enrolled in the Prospective Research for Infants with Mild Encephalopathy (PRIME) study that a TSS of ≥5 is associated with a higher encephalopathy burden and increased risk of disability. We observed similar results in this current study. The Magnetic Resonance Biomarkers in Neonatal Encephalopathy (MARBLE) study validated the full range of the scoring system in mild, moderate, and severe HIE [23]. TSS can range from 1 to 10 in mild, 6–14 in moderate, and 9–18 in severe, demonstrating overlap [4]. Additionally, heterogeneity across various encephalopathy grading systems, such as the Sarnat, modified Sarnat exam, NICHD, and SIBEN, along with challenges in frequently repeating the neurological exam, highlight the need for multimodal neuromonitoring at the bedside [4].

EEG oscillations represent brain function and provide insights into brain health during development. Slower frequencies (delta 0.5–3 Hz or theta 4–7 Hz) are generated by the thalamus and layers II to VI, while faster frequencies (alpha 8–12 Hz) originate from cells in layers IV and V of the cortex. In our earlier study [7], we reported that DP and TP are sensitive real-time biomarkers for monitoring dynamic evolution of encephalopathy severity in the first day of life. DP value of 147 best distinguished mild vs. combined mild-moderate, moderate, and severe and TSS was significantly correlated with DP. DP was also used predict death or severe brain injury on MRI as early as 9 h of life in newborns with moderate and severe HIE who received TH [8]. Temko et al. reported that nine EEG features, two heart rate variability features, and Apgar score at 24 h of life predicted (AUC under the ROC curve of 87%) neurodevelopmental outcomes in newborns with mild to severe grades of HIE who did not receive TH, highlighting the importance of a multimodal prediction model [24]. Earlier studies have shown that TP is effective for monitoring EEG background impairment [25], predicting seizures [26] and assessing the severity of HIE [7]. However, while TP can be affected by subharmonics of line noise and other periodic noise in the neonatal intensive care unit, DP is predominant in developing brain [27]. Consequently, this study focused on DP as the primary EEG biomarker.

The neurovascular unit, comprising neurons, astrocytes, blood-brain barrier endothelial cells, myocytes, pericytes, and extracellular matrix components, plays a crucial role in maintaining cerebral circulation homeostasis and ensuring the delivery of oxygen and nutrients to the brain [28‒30]. Dynamic wavelet coherence between aEEG and SctO₂ representing NVC predicted neurodevelopmental outcome in newborns with HIE who received TH versus non-encephalopathic controls [15]. In another study, an NVC of 10% on the first day of life was the best predictor of brain injury on MRI and was superior to TSS in predicting brain injury. NVC was also employed as a physiological biomarker to monitor the real-time response to erythropoietin therapy in newborns with moderate and severe HIE [31]. Despite significant heterogeneities in deriving aEEG tracings from EEG, coherence-based quantification of NVC remains unaffected [19]. Similarly, wavelet coherence between SctO₂-processed EEG (band-passed, rectified, and down-sampled) and SctO₂-aEEG showed a significantly high correlation, confirming that both processed EEG and aEEG can be reliably used with SctO₂ to assess NVC [20].

Previous studies [7, 8, 11, 12] have established that DP and NVC are strong predictors of encephalopathy severity and brain injury, while TSS is a reliable predictor of NDI [12]. However, while individual biomarkers offer valuable insights, their predictive accuracy can be limited when used in isolation. In this study, for instance, TSS demonstrated high sensitivity (0.950) but low specificity (0.263), whereas DP and NVC showed moderate sensitivity and specificity. In contrast, the use of combined biomarkers significantly improved predictive performance. The combination of TSS and DP increased sensitivity to 0.944 and NPV to 0.889, effectively minimizing false negatives. The combination of TSS and NVC achieved perfect sensitivity (1.000), making it ideal for ruling out impairment. Meanwhile, combining DP and NVC provided a better balance, with sensitivity of 0.875 and specificity of 0.615. The integration of all three biomarkers (TSS, DP, and NVC) yielded the best overall performance, achieving sensitivity of 0.750, specificity of 0.769, and PPV of 0.800. These findings highlight the potential of a multimodal biomarker approach to enhance early clinical stratification and facilitate tailored interventions.

This study has several strengths, including the simultaneous acquisition of EEG and SctO₂ data from newborns with mild to severe HIE and their 2-year neurodevelopmental outcomes. The use of automated methods to quantify DP and NVC enhances the reliability of data interpretation and supports the development of a multimodal prediction system to improve diagnostic accuracy. These data-driven approaches promote precision medicine by enabling a more tailored assessment of dynamic evolution of HIE severity for implementing targeted therapies. The potential for real-time bedside implementation further strengthens its clinical relevance. By focusing data analysis on the first day of life, we captured a critical window for clinical decision-making regarding the initiation of TH, especially for newborns with mild HIE who may progress within the first day of life.

This study has limitations including need of validation with larger cohorts, and our enrollment was from a large county hospital single inborn hospital. As a result, the reported findings, including AUC, sensitivity, and specificity, should be interpreted with caution. While this study focused on a single-center cohort, there is potential to expand the participant pool by including collaborating sites from the Comparative Effectiveness for Cooling Prospectively Infants with Mild Encephalopathy (COOLPRIME) trial (NCT04621279) [32]. Although DP and NVC show promise in predicting encephalopathy severity and MRI outcomes, and their combination with TSS predicts neurodevelopmental outcomes, they have not been integrated into commercial tools and require validation through multicenter studies such as COOLPRIME. Standardizing data acquisition, analysis methods, and clinical thresholds is essential for broader adoption. Data processing was conducted offline and the current method for calculating NVC requires at least 20 h of recording. Developing a method for shorter intervals could enhance the clinical utility of this approach for real-time monitoring at the bedside. Another limitation is that neuromonitoring was initiated at 12 ± 6 h of life, primarily due to the consenting process, as EEG is standard care while NIRS is not. Future studies should refine monitoring protocols to enable earlier and more effective outcome prediction. Addressing these limitations may improve the identification of newborns at risk for seizures or worsening encephalopathy symptoms, particularly among those with mild HIE. Early identification could facilitate the initiation of TH within the first 6 h of life, the period when treatment is most effective. Additionally, it may improve the detection of newborns at risk for seizures during the rewarming phase in those already undergoing TH [33].

Our prediction model does not include heart rate variability measures because ECG data are available for only a subset of newborns. While heart rate variability at 24 and 48 h correlates with 2-year neurodevelopmental outcomes [34], our study focuses on predicting outcomes within the critical early hours of initiating TH. Instead, we prioritized DP and NVC, which were previously validated in our studies [7, 11, 15, 31, 35]. Future work will focus on developing a comprehensive multimodal monitoring system that integrates maternal and neonatal characteristics, autoregulation [36], heart rate variability, DP, and NVC. This approach aims to enable real-time prediction of encephalopathy severity, early detection of brain injury, and long-term neurodevelopmental outcomes by adopting automated cloud service tools [37, 38]. A multimodal biomarker approach could enhance precision medicine by facilitating the identification of targeted therapies tailored to the specific needs of each individual, ultimately improving outcomes for newborns.

The combination of TSS from a neurological exam within the first 6 h of life, alongside EEG DP and NVC measured within the first day, shows significant promise as a predictor of neurodevelopmental outcomes at 2 years of age. This study highlights the critical need for dynamic multimodal neuromonitoring to identify newborns at risk for adverse neurodevelopmental outcomes. These novel multimodal physiological biomarkers could significantly improve early clinical stratification, as well as the prediction of brain abnormalities and neurodevelopmental outcomes in newborns with HIE. The capacity for real-time, prospective brain health monitoring could drive a paradigm shift in neonatal neurocritical care by establishing precise criteria to identify newborns most likely to benefit from neuroprotective interventions, ultimately enhancing their long-term outcomes.

We sincerely thank Dr. Yu-Lun Liu for her valuable assistance with the statistical analysis in this study.

The study (STU 022015-104) was approved by the Institutional Review Board at University of Texas Southwestern Medical Center, Dallas, TX, and written informed consent was obtained from a parent of each newborn prior to enrollment.

The authors have no conflicts or any competing interest to disclose.

This study is supported by Award Number R01 NS102617 from NIH National Institute of Neurological Disorders and Stroke to Dr. Chalak.

Drs. Lina F. Chalak and Srinivas Kota have full access to all study data and responsible for the integrity of the data and the accuracy of the data analysis. Concept and design: Srinivas Kota and Lina F. Chalak. Acquisition, analysis, or interpretation of data and critical revision of the manuscript for important intellectual content: Srinivas Kota, Lynn Bitar, Pollieanna Sepulveda, Soheila Norasteh, Hanli Liu, and Lina F. Chalak. Drafting of the manuscript: Srinivas Kota, Lynn Bitar, and Lina F. Chalak. Obtained funding; administrative, technical, or material support; and supervision: Lina F. Chalak.

Requests for data access should be directed to the corresponding author, accompanied by a reasonable justification. Data will be shared in compliance with relevant data protection and confidentiality guidelines.