Introduction

Neonatal neurobehavior is an indicator of long-term developmental outcomes and may be particularly critical for infants at risk for poor outcomes due to adverse prenatal conditions or preterm birth.1,2,3 Identifying infants with atypical neurobehavioral patterns could be one way to screen infants who could benefit from early intervention and prevention services. Additionally, early identification of infants with atypical neurobehavioral patterns could help researchers and clinicians pinpoint which perinatal exposures impact very early neurobehavior.

The NICU Network Neurobehavioral Scale (NNNS) is a standardized neonatal neurobehavioral exam that assesses neurologic integrity, motor and behavioral functioning, and signs of stress and abstinence.4 The NNNS exam has been used in prior research with at-risk infants, including infants born premature,5,6,7,8 infants with prenatal substance exposure,1,9,10,11,12,13,14,15,16,17 and infants without any known risk factors.18,19,20 These prior studies have identified distinct neurobehavioral profiles, including two different dysregulated profiles (i.e., hyper- and hypo-aroused profiles). Both the hypo- and hyper-aroused profiles have been shown to be associated with long-term cognitive and behavioral outcomes.1,19,21

Prior studies in single cohorts have identified pre- and perinatal factors associated with dysregulated neurobehavior as measured by the NNNS. These factors include measures of neurological integrity (e.g., abnormal cranial ultrasound,22 brain volume20), prenatal medical complications,5 prenatal mood disorders,5 newborn medical morbidities (e.g., sepsis),6 demographic risk factors (e.g., low socioeconomic status [SES], minority race/ethnicity),6 and type and dose of substance exposure.10,12 However, nearly all prior studies have investigated these associations in single cohorts of infants with similar characteristics, such as full-term, healthy infants or infants born preterm. As such, it is unknown whether the same set of NNNS profiles would be observed in a large sample representing infants across the gestational age (GA) spectrum or whether pre- and perinatal factors are associated with neurobehavioral profiles across a heterogeneous cohort of infants. Expanding on prior single-cohort findings in a large, heterogeneous cohort allows for replication of prior work and an assessment of the stability of previously reported neurobehavioral patterns.

The current study describes NNNS profiles and examines socioeconomic, medical, and substance use variables as potential predictors of these profiles in a sample of newborns spanning the GA spectrum (i.e., 22–43 weeks). We leveraged data from the Environmental influences on Child Health Outcomes (ECHO) Program, a multi-cohort study funded by the National Institutes of Health that brings together diverse cohorts of infants with varying characteristics and exposures from around the United States (U.S.). Specifically, we analyzed data from five cohorts who collected neonatal NNNS data, including two cohorts of infants born very premature (<33 weeks GA).

Methods

Study population

The ECHO Program was launched in 2016 to investigate the influence of environmental exposures on child health and development.23,24,25 Of the 69 extant cohorts within ECHO (https://www.nih.gov/echo/pediatric-cohorts), five collected data on neonatal neurobehavior using the NNNS exam (Table 1). The current analyses include data from these five cohorts, which represent 10 recruitment sites and 1112 mother–infant pairs. Of the five cohorts, two were comprised of infants born very preterm (<33 weeks GA), and three were comprised of infants mostly born at term (≥37 weeks GA). This resulted in a sample of infants with wide-ranging GA (mean = 29.9 weeks, standard deviation [SD] = 5.26 weeks, range = 22–43 weeks). Overall, 67% were born at <30 weeks; 12% were born between 30 and 37 weeks; and 21% were born at 37 weeks or later. For each cohort contributing data, the local Institutional Review Board reviewed and approved all data collection procedures, and cohort participants provided written informed consent.

Table 1 ECHO cohort description.
Full size table

Measures

NICU Network Neurobehavioral Scale

The NNNS is a standardized newborn neurobehavioral assessment that examines neurological integrity, motor and behavioral functioning, and signs of stress and abstinence. It has previously been used in studies of at-risk and healthy infants1,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20 with documented internal consistency and validity.26 The NNNS results in 12 summary scores indexing attention, handling, self-regulation, arousal, excitability, lethargy, hypertonicity, hypotonicity, non-optimal reflexes, asymmetric reflexes, quality of movement, and signs of stress and abstinence. Prior studies have used these summary scores to create discrete, mutually exclusive profiles of infants who share similar patterns of neurobehavior.1,6,7,18,19,20,27,28,29 We utilized latent profile analysis (LPA) to estimate these neurobehavioral profiles for infants included in this ECHO-wide analysis.

Maternal and neonatal variables

We examined associations between infants’ NNNS profile membership and several pre- and perinatal variables that were available and harmonized at the time of the analysis. These variables included maternal demographic, medical, and substance use characteristics and neonatal demographic and medical characteristics (Table 2). Maternal and neonatal variables were collected via standardized maternal interviews and medical record abstraction according to each cohort’s protocol. We utilized harmonized data from cohort-specific protocols along with data collected via the standard ECHO-wide protocol.24,30 Variable definitions are provided in Supplementary Table 1. Some variables were not included in final models due to multicollinearity (e.g., infant NICU status was redundant with cohort and GA) or other estimation issues (e.g., the unique effect of being large for gestational age was unable to be estimated due to low prevalence).

Table 2 Characteristics of the study sample.
Full size table

Statistical analysis

Statistical analyses proceeded in three steps. First, we used LPA to classify infants into mutually exclusive groups based on the 12 NNNS summary scores. We fit LPA models with different numbers of profiles and compared the fit of each model. To determine the best-fitting model, we used the following criteria: (1) the majority of solutions met statistical convergence criteria; (2) lower Bayesian Information Criteria (BIC) values indicated goodness of fit alongside model parsimony; (3) higher entropy and higher average class probabilities indicated high classification accuracy; (4) non-significant Lo–Mendell–Rubin (LMRT) and bootstrapped likelihood ratio (BLRT) tests indicated that a model with k profiles fit significantly better than a model with k−1 profiles; and (5) all profiles contained 5% or more of the sample. All LPA models were run in Mplus 8.8. Correlated residuals were allowed for summary scales that had items in common. Using the best-fitting LPA solution, we described the mean pattern of NNNS summary scores in each profile. A plot of mean scores by profile was generated to aid interpretation.

After identifying the latent NNNS profiles, we conducted univariate analyses using generalized estimating equations (GEEs) to compare individual neonatal and maternal characteristics across the profiles. Our GEE models included a logit link function to account for a categorical outcome. We accounted for nesting of children within the 10 recruitment sites using an exchangeable correlation matrix at the cohort-site level.

Finally, we conducted multivariate, adjusted models to determine the strongest predictors of NNNS profile membership. We performed multiple imputation (n = 20 datasets) to account for missing data in maternal and neonatal variables. Using the imputed datasets, we then ran a series of logistic GEE models to predict latent profile membership from the list of pre- and perinatal characteristics described earlier. As in the univariate models, we accounted for nesting of children within recruitment site. The results were pooled across imputation datasets prior to interpretation. In a final step, we entered interaction terms into the model to determine GA-specific correlates of NNNS profile membership. All interaction terms were entered simultaneously, and non-significant (p > 0.15) terms were removed from the model prior to interpretation. Significant interactions were probed at clinically relevant levels of GA.

Results

Descriptive statistics

Descriptive statistics for the sample are presented in Table 2. Overall, 80% of the infants were born preterm (<37 weeks GA; N = 886). Of the mothers, 6% identified as Asian, 25% as Black, 16% as Hispanic, 51% as White, and 8% as Other Race or Ethnicity. Most mothers (71%) were married, and approximately one quarter (28%) held a 4-year college degree. Obesity (33%) and pre-eclampsia (20%) were the most prevalent maternal medical problems. Additionally, 8.6% of mothers experienced depression; 7.7% experienced anxiety; and 15% used opioids during pregnancy.

NNNS profiles

We estimated LPA models with 1–7 profiles and compared model fit statistics (Table 3). A model with 6 profiles was determined to have the best fit based on a low BIC, high entropy, high average class probabilities, and a sufficient fraction in the smallest class (6% of the sample). The BLRT indicated that a model with 6 profiles fit significantly better than a model with 5 profiles. Despite the 7-profile solution having a lower BIC, a 6-profile solution was preferred due to convergence problems in the 7-profile solution. Entropy and average class probabilities were lower in the 7-profile solution compared with the 6-profile solution. In addition, the LMRT was not significant, and the BLRT was inconclusive due to convergence issues.

Table 3 Model fit statistics for LPA models.
Full size table

Using the results from the 6-profile solution, we next examined patterns of NNNS summary scores in each profile. Figure 1 shows the means of the standardized scores by profile while Table 4 shows the means and standard deviations of the NNNS summary scores by profile. Infants in profile 1 (blue in Fig. 1; 42%) had the lowest handling, excitability, and stress/abstinence scores. All remaining summary scores were approximately average. Infants in profile 2 (green in Fig. 1; 12%) had the highest attention and regulation scores. Infants in profile 3 (purple in Fig. 1; 16%) had average attention and regulation with somewhat elevated arousal and excitability. Infants in profile 4 (yellow in Fig. 1; 8%) required more handling, had below average regulation, and had elevated arousal and excitability. Infants in profiles 5 and 6 exhibited clinically significant dysregulated neurobehavioral phenotypes. Infants in profile 5 (orange in Fig. 1; 16%) were hypo-aroused. These infants had the highest lethargy, hypotonicity, and nonoptimal reflexes compared with all other profiles. Infants in profile 6 (black in Fig. 1; 6%) were hyper-aroused. These infants had the highest arousal and excitability, lowest regulation and quality of movement, and the most signs of stress/abstinence compared with all other profiles.

Fig. 1: Means of NNNS summary scores by profile.
figure 1

NNNS NICU Network Neurobehavioral Scale.

Full size image
Table 4 NNNS summary score means (SD) by profile.
Full size table

Predictors of NNNS profiles

Profiles 5 (hypo-aroused) and 6 (hyper-aroused) represent clinically relevant phenotypes given that NNNS scores fell approximately one SD above the mean for lethargy, hypotonicity, and nonoptimal reflexes (profile 5); at least one SD above the mean for handling, arousal, excitability, and stress abstinence (profile 6); and almost two SDs below the mean for regulation and quality of movement (profile 6). Therefore, subsequent analyses examined predictors of membership in these two dysregulated profiles compared with profiles 1-4, which contained infants with typical patterns of neurobehavior.

Descriptively, infants born premature were more likely to be in the hypo-aroused profile (17% vs 11%) but less likely to be in the hyper-aroused profile (4% vs 15%) compared with infants born full-term. This difference was only significant for the hyper-aroused profile (odds ratio [OR] = 0.33, 95% confidence interval [CI] = 0.15–0.74). Continuous GA was not associated with membership in the hypo- or hyper-aroused profile.

The results for the multivariate analyses adjusting for GA and all other neonatal and maternal characteristics are shown in the right-hand panel of Table 5. GA was not associated with profile membership after adjusting for all other predictors.

Table 5 Means and frequencies of predictor variables by NNNS profile.
Full size table

Adjusted analyses showed that boys were more likely to be in the hypo-aroused profile compared with girls (OR = 1.37; 95% CI = 1.14–1.64). Maternal age at birth was negatively associated with membership in the hypo-aroused profile (OR = 0.97, 95% CI = 0.94–0.99). Maternal depression was positively associated with membership in the hypo-aroused profile (OR = 1.51, 95% CI = 1.06–2.51).

Infants in the hyper-aroused profile were more likely to be of Hispanic/Latino ethnicity (OR = 1.67, 95% CI = 1.04–2.68). Similar to the hypo-aroused profile, maternal depression was associated with a greater likelihood of being in the hyper-aroused profile (OR = 2.50, 95% CI = 1.02–6.15). Finally, infants born to mothers who used tobacco prenatally were more likely to be in the hyper-aroused profile (OR = 1.73, 95% CI = 1.01–2.95).

Because the sample was primarily comprised of children born at <30 weeks GA, we ran sensitivity analyses where we replaced continuous GA with a dichotomous variable indicating whether infants were born at <30 weeks GA. Model findings were very similar. Infant sex, maternal age, and maternal depression continued to be associated with membership in the hypo-aroused profile. Hispanic/Latino ethnicity was associated with the hyper-aroused profile, and maternal depression and tobacco use were marginally associated (ps = 0.06). Full results from these sensitivity models are presented in Supp. Table 2.

A secondary investigation of GA-specific predictors indicated a significant interaction between continuous GA and opioid use (GA × Opioid: b = 0.15, p = 0.001). Opioid use was associated with an increased likelihood of being in the hypo-aroused profile but only for children born at a later GA. For example, opioid exposure was not significantly associated with an increased likelihood for infants born at 30 weeks GA (OR = 1.24; 95% CI = 0.80–1.94) but was associated with an increased likelihood for infants born at 37 weeks GA (OR = 3.52; 95% CI = 1.78–6.95).

Analysis of the hyper-aroused profile indicated a significant interaction between continuous GA and White race (GA × White race: b = 0.15, p = 0.001). No significant association was found between race and likelihood of being in the hyper-aroused profile for infants born at 30 weeks GA (OR = 0.74, 95% CI = 0.41–1.33), but White race was associated with a lower likelihood of being in the hyper-aroused profile for infants born at 27 weeks GA (OR = 0.47, 95% CI = 0.25–0.93).

Discussion

The current study aimed to identify neonatal neurobehavioral profiles and predictors of profile membership in a heterogeneous cohort of infants with different GAs and perinatal exposures. We found evidence for six distinct neurobehavioral profiles, with two dysregulated profiles similar to those reported in prior work. Infants in the hypo-aroused profile were lethargic and hypotonic, whereas children in the hyper-aroused profile were highly aroused and excitable, with poor regulation and quality of movement and many signs of stress and abstinence. We identified infant and maternal characteristics that were associated with membership in each of these two atypical profiles, with some effects that were contingent on infant GA.

Prior studies investigating neonatal neurobehavior using the NNNS have also identified the presence of hypo- and hyper-aroused profiles. For example, the hyper-aroused profile has been identified in samples of drug-exposed infants2,28,29 and term, healthy infants,18,19,20,27 with similar patterns of atypical scores for arousal, excitability, hypertonia, quality of movement, regulation, and stress signs. The hypo-aroused profile has also been seen in prior studies of term, healthy infants19,20,27 and premature infants,6,7 which is typically characterized by high lethargy and hypotonicity along with non-optimal reflexes. While the magnitude of extreme scores may vary based on the study population, it is striking that qualitatively similar profiles of infants have emerged across these different investigations. The current analyses of infants with different GAs and exposures confirm that these profiles can be observed across a range of infants who vary in terms of their risk for neurobehavioral impairment.

Distinguishing between different types of dysregulated neurobehavior is important for at least two reasons. First, understanding whether infants are hypo- or hyper-aroused could lead to the development of personalized interventions with strategies that are tailored to the specific needs of these infants (for example, teaching caregivers different strategies to support up- or down-regulation of arousal and excitability). Existing NICU interventions do not specifically target neurobehavioral dysregulation, nor does the standard of care in most NICUs involve a standardized neurobehavioral exam such as the NNNS. Thus this information could be used to improve upon existing interventions for high-risk neonates prior to and immediately following NICU discharge, especially since the efficacy of many existing interventions have not been robustly supported by randomized control trials.31,32 Second, infants with different types of dysregulated neurobehavior may be at risk for different downstream outcomes (e.g., cognitive versus behavioral problems, internalizing versus externalizing disorders), and appropriate supports following NICU discharge (e.g., early screening, referrals) could look different depending on these risks.

Predictors of atypical profiles included demographic, medical, and substance use variables. Infants in the hypo-aroused profile were more likely to be male, have mothers of a younger age, and have mothers who were depressed prenatally. Others have similarly reported a preponderance of males in hypotonic profiles,20,29 with hypotonia possibly representing neurodevelopmental immaturity. Suggestive links between maternal depression and hypotonic profiles and/or characteristics of the hypotonic profile (i.e., high lethargy) have also been reported.5,19,33,34 Links between maternal age and NNNS scores are novel. Older maternal age may be protective because of its association with greater educational and career attainment, financial and social resources, access to high quality care during pregnancy, social support during pregnancy, and socioemotional skills and maturity.35 However, these findings do not necessarily generalize to samples that would be considered of advanced maternal age given the relatively low proportion of women above age 35 (24%) or age 40 (7%) in the sample.

Infants in the hyper-aroused profile were more likely to be Hispanic/Latino and have mothers who were depressed or used tobacco prenatally. Sociodemographic factors, including minority race/ethnicity, have been associated with atypical NNNS scores.6,36 This association is likely due to the widespread, pernicious impacts of institutional racism in the U.S. For example, oppressed groups are afforded less access to healthy foods, experience greater financial and other stressors, and are exposed to more crime and pollution, all of which may affect fetal central nervous system development and newborn neurobehavior.37,38 Further, issues unique to members of the LatinX community, such as immigration-related stressors,39 may explain why we found a main effect of Hispanic/Latino ethnicity and not Black race. As discussed before, maternal depression has been associated with atypical neurobehavior, although these links have been more apparent for hypo- rather than hyper-arousal. Finally, tobacco use, either on its own or in combination with other prenatal drug exposures, has been shown to be associated with a greater likelihood of hyper-arousal in infants.1,29,40,41

Our sample was primarily comprised of premature infants, with small cell sizes for some GA categories and several prenatal and perinatal predictors within each latent profile. Therefore, we were limited in our ability to test GA-specific effects. However, our findings suggest that neurobehavioral effects associated with opioid exposure may be more pronounced among term infants than preterm infants. Indeed, prior studies have shown that the clinical presentation of neonatal opioid withdrawal syndrome (NOWS) is markedly different in preterm versus term infants42,43 to the extent that different NOWS diagnostic tools may be needed for preterm infants. For example, hyperexcitability is considered to be a hallmark presentation of NOWS, yet it may not be as common a presentation in preterm newborns with NOWS, whereas other symptoms such as poor feeding may be more common.42 Conversely, we found that maternal race was more predictive of neurobehavior in infants born at extremely low GA. These findings suggest that belonging to a non-oppressed racial/ethnic group may be protective for very premature infants, a trend that has been noted in the recent literature.44 It is likely that lower maternal stress,45 better access to high quality prenatal and neonatal healthcare,46 and relatively better socioeconomic position47 may provide some advantage for White premature infants.

Interestingly, no significant associations were observed between GA and NNNS profile membership after adjusting for potentially confounding infant and maternal characteristics. These findings suggest that factors associated with prematurity may predict profile membership rather than prematurity per se. Prior studies have found poorer neurodevelopment in preterm versus full-term infants based on NNNS summary scales, but these studies did not estimate NNNS profiles nor did they account for the wide number of factors considered in the present study.36,48 Follow-up multicohort studies with preterm and term-born children are needed to untangle the associations among prenatal risk factors, neonatal illness, and neonatal neurobehavior.

This study was unique in that it incorporated a large, heterogeneous sample of neonates—the largest ever investigated with the NNNS scale—with racial/ethnic, socioeconomic, and geographic diversity. However, several limitations remain. First, although a wide range of GAs were represented in the sample, there was a preponderance of very preterm infants and a relative dearth of term infants. Additionally, although we included a large number of risk factors, not all relevant medical and demographic factors were available. Because maternal health complications were exclusionary for some of the cohorts included in this analysis, we may have been biased toward underestimating the effect of some maternal medical problems on infant neurodevelopment. Because of the nature of the harmonizing efforts in the ECHO Program, some variables were collapsed across the immediate pre-pregnancy, pregnancy, and early postnatal periods. For example, prenatal opioid use was coded as present if there was any evidence of use from 3 months before to 6 months after the target pregnancy. Therefore, these variables are best understood as representing perinatal characteristics or exposures. Finally, although we found some associations between maternal mental health and child neurobehavioral profiles, insufficient data were available regarding psychiatric medication use; therefore, we could not examine the specific effects of different psychiatric medications (e.g., selective serotonin reuptake inhibitors).

In conclusion, we found evidence for two dysregulated profiles in a large sample of preterm and term infants drawn from a national consortium of cohorts. These hypo- and hyper-aroused profiles have clinical relevance and different perinatal antecedents. Given the links between these two dysregulated profiles and later cognitive and socioemotional outcomes,1,19,21 identifying the precursors associated with these neurobehavioral profiles in infants of varying GA could identify risk and inform practices to promote positive developmental outcomes starting at the very beginning of life. Clinicians could consider screening infants exposed to adverse prenatal conditions using a neurobehavioral exam such as the NNNS to aid in the infant’s care and discharge plan.