Inflammatory Predictors of Neurobehavior in Very Preterm Infants

Abstract

Background:

Preterm infants are at risk for impaired neurodevelopment. Inflammation may be an important modifiable mediator of preterm birth and neurodevelopmental impairment, but few studies have examined longitudinal measures of inflammation.

Objective:

To determine the relationship between longitudinal measures of inflammation and neurobehavior in very preterm infants.

Study design:

Non-experimental, repeated measures cohort study

Methods:

Very preterm infants were enrolled between October 2017 and December 2018. Blood was collected weekly until 35 weeks post-menstrual age for the quantification of plasma cytokines. Neurobehavior was assessed at 35 weeks post-menstrual age using the cluster scores for motor development and vigor and alertness/orientation from the Neurobehavioral Assessment of the Preterm Infant. Multiple linear regression models with robust standard errors were used to analyze the data. Average levels of individual cytokines, cytokine trends, and composite scores were used as measures of inflammation.

Results:

Seventy-three infants were enrolled in the study. Interleukin-1 receptor antagonist was associated with motor development and vigor scores. Interleukin-6 was associated with alertness/orientation scores. Tumor necrosis factor-alpha and composite scores of inflammation were associated with motor development and vigor and alertness/orientation scores. There were interactions with post-menstrual age at birth and infant sex.

Conclusion:

Inflammation may be an important predictor of short-term neurobehavior in preterm infants. Interleukin-1 receptor antagonist, interleukin-6, and tumor necrosis factor-alpha are key cytokines for studies of preterm infants, but composite scores may be a better measure of inflammation than individual cytokines. Inflammation can be damaging to the immature brain and may be a specific target for future interventions to improve outcomes.

Introduction

Preterm birth is the leading cause of mortality for children less than five years old across the world (1). In the United States, approximately 10% of infants are born preterm (2), and, in recent years, the rate of preterm birth has been increasing (2, 3). Advanced medical technology has improved the rate of survival for the majority of infants born after 23 weeks post-menstrual age (PMA) (4). However, the survivors of preterm birth are at high risk for neurodevelopmental impairment (5, 6). Impairments in cognition (7, 8), motor development (9), sensory processing (10, 11), and socio-emotional and behavioral functioning (12, 13) are common among children born preterm. Neurodevelopmental impairments continue to affect the survivors of preterm birth throughout the life course, as adolescents and young adults who were born preterm report higher rates of mental health concerns (14–16) and poorer academic performance (17) compared to those who were born at term. Novel interventions are needed to improve outcomes for preterm infants to decrease some of the human, societal, and economic burdens of preterm birth- associated neurodevelopmental impairment. To provide the foundation for such interventions, the potentially modifiable causes of neurodevelopmental impairment must be examined.

While there are many potential contributors to poor outcomes in preterm infants, chronic inflammation has been identified as a potentially important cause of brain injury and neurodevelopmental impairment (18). The association between inflammation and neurodevelopment in high-risk preterm infants, including those born extremely preterm (19, 20), small for gestational age (21), or with exposure to chorioamnionitis (22), has been previously documented. However, the contribution of inflammation to neurodevelopmental impairment in very preterm infants without overt exposure to prenatal infections has been less well defined. Thus, the purpose of this study was to determine the relationship between longitudinal measures of inflammation and early neurobehavior in very preterm infants without exposure to prenatal infection.

Inflammatory cytokines, including interleukin (IL)-1β, IL-6, IL-8, and tumor necrosis factor-alpha (TNF-α), can predict neurodevelopment of preterm infants when measured in the first weeks of life (23). However, results across studies have been inconsistent. While some researchers have reported that higher cord blood levels of IL-6, IL-8, and TNF-α predicted brain injury in preterm infants (22), others have reported no association (24). The majority of researchers studying inflammation and neurodevelopment in preterm infants have analyzed individual cytokines at distinct time points (23), which may not fully describe the inflammatory milieu or distinguish infants with transient inflammation from those with more chronic inflammation. Researchers examining repeated measures of cytokines have found that persistent inflammation, defined by repeated elevations in individual cytokines, is predictive of brain injury (25, 26), cognitive impairment (27, 28), and cerebral palsy (19, 25); the majority of these research results emanated from the Extremely Low Gestational Age Newborn (ELGAN) study of preterm infants born before 28 weeks PMA (29). While extremely preterm infants (i.e. less than 28 weeks PMA) suffer the most significant neurodevelopmental deficits, infants bom less than 32 weeks PMA but greater than 28 weeks account for a significant proportion of the population- level rates of neurodevelopmental impairment (30). Moreover, these infants bom very preterm (i.e. 28–31 weeks PMA) comprise a larger proportion of preterm infants compared to those born extremely preterm (31). Thus, the purpose of our study was to determine the relationship between repeated measures of inflammation and early neurobehavior in preterm infants born between 28–31 weeks PMA (i.e. very preterm) and without exposure to prenatal infection.

Methods

This analysis was part of a larger non-experimental, repeated measures study on the relationships among stress exposure, inflammation, and neurodevelopment in very preterm infants (32). We used a convenience sample of very preterm infants, enrolled between October 2017 and December 2018 from four neonatal intensive care units (NICUs) in a 276-bed NICU system in a large U.S. metropolitan area. Infants were enrolled within the first two weeks of life. Blood was collected weekly for the quantification of cytokines until 35 weeks PMA. Neurobehavior was measured at 35 weeks PMA.

Infants were included if they (1) were born between 28-31 weeks PMA, (2) were born or transferred into one of the four study NICUs before the second day of life (DOL), and (3) were born to mothers who were English-speaking and able to provide informed consent. Infants were excluded if they (1) were born small for gestational age, (2) were born to mothers with oligohydramnios, (3) displayed signs of neonatal abstinence syndrome or were born to mothers with documented opioid drug abuse, (4) were born to mothers with confirmed chorioamnionitis or who were febrile with purulent amniotic fluid at the time of delivery, (5) had positive initial blood cultures indicating congenital infection, (6) had congenital or genetic anomalies known to affect neurodevelopment, or (7) developed severe neurologic injury (e.g. intraventricular hemorrhage with ventricular dilation, periventricular leukomalacia, neonatal encephalopathy) within the enrollment period. Information for screening was viewed in the electronic health record.

The study was approved by the hospital and university Institutional Review Boards prior to electronic health record screening of the first potential participant. Mothers of enrolled infants provided written informed consent prior to any study activities. To minimize stress to participants, research blood was collected only with a clinically-indicated lab collection. Although the NAPI is administered in a manner that minimizes infant stress, we only performed the assessment when infants required no more than nasal cannula for respiratory support and discontinued the assessment if infants experienced bradycardia or oxygen desaturation.

Data Collection

Electronic health record data.

Demographic and clinical data, including prenatal exposures, clinical comorbidities, and clinical treatments, were extracted from the electronic health record. We also collected data to assign illness severity scores to infants using the Neonatal Medical Index (NMI) (33). Infants were assigned a score from 1 (birth weight >1000g and no medical complications) to 5 (severe medical complications).

Cytokine measures.

Whole blood (300 μL) was collected weekly in a dipotassium ethylenediaminetetraacetic acid (K₂EDTA) microtainer tube (Becton Dickinson; Franklin Lakes, NJ) from the first week of life through 35 weeks PMA. Blood was collected by heelstick, venipuncture, or arterial puncture or from an indwelling line at the same time as a clinically-indicated lab draw. Samples were stored at 4° C and processed as soon as possible after collection. Blood samples were centrifuged at room temperature for 10 minutes at 2000 x_g. When collected in K₂EDTA tubes, plasma cytokines are stable for approximately 48 hours at 4° C (34). Plasma was pipetted into aliquots and stored at −80° C until analysis, which occurred within 18 months of collection.

We quantified IL-6, IL-8, IL-1β, IL-17A, TNF-α, monocyte chemoattractant protein (MCP)-1, IL-4, IL-10, and IL-1 receptor antagonist (RA) simultaneously using a custom Bio-Plex multiplex assay (Bio-Rad Laboratories; Hercules, CA), according to the manufacturer’s instructions. Samples were run in duplicate. The cytokines were theoretically selected as previously described (32). Cytokines were quantified in picograms per milliliter (μg/mL).

Neurobehavioral assessment.

We measured neurobehavior at 35 weeks PMA using the motor development and vigor (MDV) and alertness/orientation (AO) clusters from the Neurobehavioral Assessment of the Preterm Infant (NAPI) (35). NAPI MDV scores are based on performance on 7 unique items. NAPI AO scores are based on performance on 10 unique items. Assessments were started approximately one hour prior to a scheduled feeding. For statistical analysis, converted cluster scores were transformed to z-scores based on published means and standard deviations for the norm reference group (35).

Data Analysis

Due to the sample size and potential collinearity among predictor variables, we performed bivariate analyses (Spearman’s rho, Kruskal-Wallis, and Wilcoxon rank sum, as appropriate) to identify potential confounders. Specifically, we defined a possible confounder as a variable that was significantly associated (at the p<.05 level) with both the predictor (i.e. measures of inflammation) and the outcome (i.e. NAPI MDV or AO). We used multiple linear regression models with Huber-White sandwich variance estimators (i.e. robust standard errors) to estimate the association between inflammation and neurobehavior without adjustment, with adjustment for potential confounders, and with interactions that allowed the effect of inflammation to differ by PMA at birth and infant sex. PMA was centered on the lowest possible PMA of infants in the study (i.e. 196 days or 28 weeks).

A single measure of inflammation for each infant was operationalized as (1) average level of each cytokine over time, (2) individual-level trend for each cytokine over time, and (3) a composite score based on average levels of all cytokines. For average levels, the average level of each cytokine was calculated for each infant who had at least one plasma collection. For cytokine trends, the linear slope for each cytokine trajectory was calculated for each infant with at least two cytokine measures; the individual-level slope was then used as the inflammation variable in the regression models. Finally, we calculated a composite inflammation score as a z-score, using the empirical means and standard deviations for each cytokine (32, 36, 37). Average z-scores over time for each cytokine were averaged for a composite score.

We standardized the regression coefficients in the models with average cytokine levels and cytokine trends to facilitate interpretation and comparison across cytokines. Thus, regression coefficients in these models represent the change in NAPIMDV or AO z-score for each 1 standard deviation increase in average cytokine level or cytokine linear slope. For composite scores, regression coefficients represent the change in NAPI MDV or AO z-score for a 1 unit change in composite score. All analyses were performed using STATA (version 14; StataCorp; College Station, TX). Graphs were created using STATA or PROCESS (version 3) (38) and SPSS (version 26; IBM Corp.). Results were considered significant at α = .05.

Results

Of the 232 eligible infants born between 28-31 weeks PMA and hospitalized in a participating NICU, mothers of 73 infants consented to participation in the study (Figure 1). Of the ineligible infants born in the specified PMA range, most were small for gestational age (n = 29), had in utero exposure to opioids (n = 19), were bom with congenital anomalies (n = 17), or suffered neurologic injuries or were diagnosed with neural deficits prior to enrollment (n = 15). We were unable to contact mothers of 98 eligible infants. Mothers of the 61 eligible infants who declined participation most commonly did not respond to the invitation to participate (n = 24) or did not provide a reason for declining (n = 17). Fifty-four percent of those who were invited to participate in the study consented to enrollment. We excluded 2 of 73 study participants from this analysis for neurologic injuries that were not apparent at the time of their enrollment but were diagnosed within the first two weeks of life. One infant was unable to complete the study due to clinical instability that prevented neurobehavioral assessment. We were unable to collect blood from two infants. Thus, the results reported here include data from 68 infants for whom we had outcome assessments and at least one cytokine measure.

Figure 1.

Participant recruitment and enrollment.

Note. NAS, neonatal abstinence syndrome; IVH, intraventricular hemorrhage; PVL, periventricular leukomalacia; HIE, hypoxic-ischemic encephalopathy; CPS, children’s protective services; DOL, day of life; NAPI, Neurobehavioral Assessment of the Preterm Infant.

We present the demographic and clinical characteristics of included infants (n = 68) in Table 1 (39, 40). The majority of infants were male (66.2%), bom at an average of 30 weeks PMA (SD = 1.1), and with an average birthweight of 1427.1 grams (SD = 320.4). Using NMI as a measure of illness severity (33), we assigned approximately three-fourths of infants a score of 4 (20.6%) or 5 (57.4%), indicating these infants required at least 2 weeks of respiratory support, had a birthweight less than 1000 grams, or suffered neurologic injury after the initial enrollment period. Bronchopulmonary dysplasia was the most frequently diagnosed comorbidity (61.7%). Mothers of enrolled infants primarily self-identified as White (n = 76.4%) and were an average of 29.1 years old (SD = 6.9). Most mothers received prenatal steroids (86.8%) and antibiotics (57.6%). Less than half (44.8%) received magnesium for neonatal neuroprotection or the treatment of preeclampsia.

Table 1.

Clinical and Demographic Characteristics of Study Participants

			n (%)
Infant Variables
Male sex			45 (66.2)
Twin gestation			16 (23.5)
Neonatal Medical Indexa
Score=1			1 (1.5)
Score=2			2 (2.9)
Score=3			12 (17.7)
Score=4			14 (20.6)
Score=5			39 (57.4)
Clinical comorbidities
IVH without dilation			16 (23.5)
Periventricular leukomalacia			5 (7.4)
Bronchopulmonary dysplasiab			42 (61.8)
Mild			16 (23.5)
Moderate			12 (17.7)
Severe			14 (20.6)
Blood stream infection			1 (1.5)
Urinary tract infection			2 (2.9)
Necrotizing enterocolitisc			2 (2.9)
Postnatal exposures
Caffeine			53 (77.9)
Inhaled glucocorticoids			9 (13.2)
Intravenous glucocorticoids			2 (2.9)
Antibiotics			45 (66.2)
Maternal Variables
Race
White			53 (77.9)
Black/African-American			12 (17.7)
Asian			3 (4.4)
Ethnicity
Hispanic/Latina			1 (1.5)
Non-Hispanic			58 (85.3)
Not reported			9 (13.2)
Prenatal complications
Suspected chorioamnionitis			3 (4.4)
Hypertension			24 (35.3)
Diabetes			8 (11.9)
Prenatal exposures
Steroids			59 (86.8)
Antibiotics			38 (57.6)
Magnesium			30 (44.8)
Smoking			9 (13.2)
Illicit drug use			8 (11.8)
	Mean (SD)	Median	Min-Max
Infant Variables
Birth post-menstrual age (weeks)	30 (1.1)	30.1	28.1-31.9
Birthweight (grams)	1427.1 (320.4)	1346.5	770-2331
Length of stay (days)	57.8 (21.4)	57.5	20-129
Total days CPAP	24.9 (15.1)	27	0-70
Total days supplemental oxygen	11.1 (18.2)	3	0-97
Maternal Variables
Age (years)	29.1 (6.9)	29	15-43

Note. n=68;

^aNeonatal Medical Index (31) is scored 1 (low acuity) to 5 (high acuity);

^bclassification (37);

^cStage II necrotizing enterocolitis (38); IVH, intraventricular hemorrhage

Measures

Cytokines.

The number of expected plasma collections for the quantification of cytokines differed for each infant and was based on the infant’s PMA at birth. While infants born at 28 weeks PMA could have up to 7 collections, we expected infants born at 31 weeks to have no more than 4 collections. In addition to samples that were not expected due to infant age, there were some missing samples. Categories of missing data are displayed in Table 2. In total, 263 plasma samples were collected across the 7 possible time points.

Table 2.

Measured and missing samples for plasma cytokine quantification

	Timepoint n (%)
	1	2	3	4	5	6	7
Measured	50	46	41	52	41	22	11
	(73.5)	(67.7)	(60.3)	(76.5)	(60.3)	(32.4)	(16.2)
Not expected	0	0	0	0	16	39	54
	(0)	(0)	(0)	(0)	(23.5)	(57.4)	(79.4)
Missing	18	22	27	16	11	7	3
	(26.5)	(32.4)	(39.7)	(23.5)	(16.2)	(10.3)	(4.4)
	100%	100%	100%	100%	100%	100%	100%
No clinical lab draw	0	15	19	9 (56.3)	9 (81.8)	6 (85.7)	2 (66.7)
	(0)	(68.2)	(70.4)
Collection missed	0	4 (18.2)	7 (25.9)	6 (37.5)	1	1 (14.3)	1 (33.3)
	(0)				(9.1)
Infant not yet enrolled	13	2	0	0	0	0	0
	(72.2)	(9.1)	(0)	(0)	(0)	(0)	(0)
Infant discharged	0	0	0	0	1	0	0
	(0)	(0)	(0)	(0)	(9.1)	(0)	(0)
Too soon after previous	5 (27.8)	1	1	1	0	0	0
		(4.5)	(3.7)	(6.3)	(0)	(0)	(0)
	100%	100%	100%	100%	100%	100%	100%

Plasma cytokines were quantified on 8 Bio-Rad multiplex immunoassay plates (Bio-Rad; Hercules, CA). The average intra-assay coefficients of variation (CV) for sample duplicates for each of the 9 cytokines were 11.3% for IL-6, 10.8% for TNF-α, 10.2% for MCP-1, 13% for IL-8, 7.2% for IL-1β, 18.2% for IL-1RA, 9% for IL-10, 8.9% for IL-4, and 9.6% for IL-17A. The manufacturer reports an intra-assay CV of less than 15% and an inter-assay CV less than 25%.

For the majority of plasma samples, analyzed values fell within the assay’s detectable range or could be extrapolated from the equation for the standard curve for IL-6 (94.7%), TNF-α (100%), MCP-1 (100%), IL-8 (100%), and IL-1RA (99.6%). IL-6 was not detectable in 14 samples and IL-1RA was not detectable in 1 sample. These undetectable values were changed to 0 for subsequent analyses. In more than 80% of samples, IL-1β, IL-10, IL-4, and IL-17A levels were either extrapolated or undetectable. Thus, these cytokines were not included in subsequent analyses.

Neurobehavioral assessments.

Neurobehavioral assessments using the NAPI were performed for each infant between 34 and 38 weeks PMA (M = 35.8, SD = 1). Because the PMA differed among infants at the time of assessment due to early discharge or continued requirement for respiratory support beyond a nasal cannula, we calculated z-scores from the NAPI cluster converted scores based on the published norm-referenced means and standard deviations for infants between 32 and 37 weeks PMA (35). We excluded AO scores for two infants from the analyses because these infants had their eyes dilated for ophthalmologic screening at the time of the assessment, rendering the AO cluster invalid.

Analyses for Confounding

We conducted analyses for each variable in Table 1 to identify potential confounders of the relationship between inflammation and NAPI MDV or AO. We first performed bivariate analyses to identify significant associations between the variables in Table 1 and NAPI MDV and AO. In this first step, we determined that prenatal magnesium exposure, maternal age, and infant length of stay (LOS) were associated with NAPI MDV. Postnatal caffeine exposure, inhaled glucocorticoid exposure, days of continuous positive airway pressure, and LOS were associated with NAPI AO. We then repeated the bivariate analyses to determine whether these potential confounders were also associated with our measures of inflammation. Only infant length of stay (LOS) was associated with measures of inflammation and MDV. Postnatal exposure to caffeine, birthweight, and LOS were associated with measures of inflammation and AO. Because caffeine exposure and birthweight were significantly associated with LOS, we controlled for only LOS in the regression analyses.

Average Cytokine Levels

From models using average cytokine levels over time, we found that higher average levels of TNF-α and IL-1RA predicted poorer NAPI MDV scores, controlling for LOS (Table 3). Average IL-6, MCP-1, and IL-8 levels were not associated with MDV. There were no interaction effects for PMA or sex. In the models for AO, we found associations with IL-6 and TNF-α, controlling for LOS (Table 3). In these models, higher levels of IL-6 and TNF-α were associated with higher AO scores. There were no interactions with PMA at birth or infant sex. Between 11– 18% of the variance in MDV and AO was explained by the interaction models.

Table 3.

Multiple linear regression models with significant results using average cytokine levels

	Unadjusted β [95% CI]	p-value	R2	Adjusted β1 [95% CI]	p-value	R2	β1 [95% CI]	p-value	R2
NAPI MDV
TNF-α	−0.16 [−0.42, 0.12]	0.26	0.018	−0.11 [−0.35, 0.15]	0.41	0.087	−0.33 [−0.60, −0.06]	0.018	0.11
TNF-α*PMA							0.03 [−0.02, 0.08]	0.3
TNF-α*sex							−0.04 [−0.74, 0.67]	0.92
IL-1RA	−0.42 [−0.70, −0.16]	0.002	0.13	−0.37 [−0.64, −0.08]	0.013	0.16	−0.61 [−1.05, −0.16]	0.008	0.18
IL-1RA*PMA							0.02 [−0.03, 0.07]	0.38
IL-1RA*sex							0.01 [−0.49, 0.52]	0.96
NAPI AO
IL-6	0.09 [0.02, 0.15]	0.01	0.009	0.11 [0.03, 0.20]	0.011	0.068	1.29 [0.005, 2.52]	0.049	0.13
IL-6*PMA							−0.11 [−0.25, 0.02]	0.09
IL-6*sex							0.49 [−1.31, 2.35]	0.59
TNF-α	−0.02 [−0.19, 0.15]	0.78	0.0006	0.01 [−0.17, 0.20]	0.88	0.054	0.22 [0.01, 0.42]	0.042	0.11
TNF-α*PMA							−0.03 [−0.07, 0.01]	0.13
TNF-α*sex							0.15 [−0.39, 0.70]	0.58

Note.

¹Adjusted for length of stay. NAPI, Neurobehavioral Assessment of the Preterm Infant; MDV, motor development and vigor; AO, alertness/orientation; TNF-α, tumor necrosis factor-alpha; IL-1RA, interleukin-1 receptor antagonist; PMA, post-menstrual age; β = standardized regression coefficient; CI, confidence interval; R² = variance explained by model.

Cytokine Trends over Time

Using cytokine slopes for each participant in the multiple linear regression model, we found that the trend for IL-1RA over time predicted NAPIMDV and that this association was moderated by PMA at birth (Table 4). Although sex did not moderate the relationship between the IL-1RA trend and NAPI MDV, there was an interaction effect between sex and the TNF-α trend. Compared to females, the negative association between TNF-α and NAPI MDV scores was attenuated in males. There were no other significant findings for cytokine trends and NAPI MDV. For NAPI AO, we found an association between the trend for IL-6 and AO scores, controlling for LOS. There were no interactions with PMA at birth. Only the effect of the IL-8 trend on NAPI AO was moderated by sex (Table 4). Between 12-23% of the variance in MDV and AO was explained by the interaction models.

Table 4.

Multiple linear regression models with significant results using cytokine trends over time

	Unadjusted β [95% CI]	p-value	R2	Adjusted β1 [95% CI]	p-value	R2	β1 [95% CI]	p-value	R2
NAPI MDV
TNF-α	−0.06 [−0.35, 0.23]	0.69	0.0026	−0.06 [−0.30, 0.19]	0.62	0.088	−0.09 [−0.86, 0.68]	0.82	0.16
TNF-α*PMA							−0.03 [−0.07, 0.01]	0.14
TNF-α*sex							0.83 [0.07, 1.61]	0.032
IL-1RA	0.22 [−0.30, 0.75]	0.4	0.037	0.14 [−0.40, 0.68]	0.6	0.10	0.89 [0.47, 1.30]	<0.0005	0.23
IL-1RA*PMA							−0.04 [−0.07, −0.02]	0.002
IL-1RA*sex							−0.18 [−0.60, 0.22]	0.35
NAPI AO
IL-6	−0.19 [−0.33, −0.05]	0.008	0.044	−0.23 [−0.37, −0.09]	0.002	0.091	0.06 [−0.76, 0.88]	0.88	0.12
IL-6*PMA							−0.01 [−0.03, 0.02]	0.67
IL-6*sex							−0.21 [−1.01, 0.59]	0.6
IL-8	0.01 [−0.21, 0.23]	0.91	0.0002	0.03 [−0.19, 0.27]	0.78	0.032	0.74 [−0.23, 1.71]	0.13	0.14
IL-8*PMA							0.01 [−0.04, 0.07]	0.68
IL-8*sex							−0.97 [−1.67, −0.29]	0.006

Note.

¹Adjusted for length of stay. NAPI, Neurobehavioral Assessment of the Preterm Infant; MDV, motor development and vigor; AO, alertness/orientation; TNF-α, tumor necrosis factor-alpha; IL-1RA, interleukin-1 receptor antagonist; PMA, post-menstrual age; β = standardized regression coefficient; CI, confidence interval. R² = variance explained by model

Composite Scores

Using composite scores that included IL-6, TNF-α, MCP-1, IL-8, and IL-1RA, we found that higher composite scores of inflammation were associated with decreases in NAPI MDV scores (Table 5, Figure 2), controlling for LOS. This association was not moderated by PMA at birth or infant sex. The interaction model explained 11% of the variance in MDV scores.

Table 5.

Multiple linear regression results for the effect of composite scores of inflammation on neurobehavior

	Unadjusted β [95% CI]	p-value	R2	Adjusted β1 [95% CI]	p-value	R2	β1 [95% CI]	p-value	R2
NAPI MDV
Composite score	−0.53 [−1.09, 0.042]	0.069	0.032	−0.32 [−0.91, 0.26]	0.28	0.09	−0.76 [−1.32, −0.20]	0.009	0.11
Composite*PMA							0.016 [−0.10, 0.13]	0.79
Composite*sex							0.56 [−0.96, 2.07]	0.47
NAPI AO
Composite score	0.096 [−0.32, 0.51]	0.65	0.0017	0.27 [−0.21, 0.75]	0.27	0.066	0.70 [0.17, 1.23]	0.01	0.21
Composite*PMA							−0.14 [−0.22, −0.053]	0.002
Composite*sex							1.77 [0.36, 3.17]	0.015

Note.

¹Adjusted for length of stay. NAPI, Neurobehavioral Assessment of the Preterm Infant; MDV, motor development and vigor; AO, alertness/orientation; PMA, post-menstrual age; β = regression coefficient; CI, confidence interval. R² = variance explained by model

Figure 2.

Association between composite inflammation score and NAPI MDV.

Note. Composite inflammation scores include averages of interleukin (IL)-6, IL-8, IL-1 receptor antagonist, tumor necrosis factor-alpha, and monocyte chemoattractant protein-1 over time. Higher composite inflammation scores were associated with lower NAPI MDV scores (p = .009). This association was not significantly moderated by infant sex or post-menstrual age at birth. Figure created with STATA version 14. NAPI, Neurobehavioral Assessment of the Preterm Infant; MDV, motor development and vigor.

As can be observed in Figure 2, there are several potentially influential observations, including one observation with a composite inflammation score greater than 2 and three observations with composite inflammation scores greater than 0.8. When we removed these potentially influential observations, the value of the point estimate increased but was no longer statistically significant (β = 1.21, [−3.69, 1.26], p=.33).

We found a positive association between composite scores and NAPI AO (Table 5, Figure 3). Furthermore, this association was moderated by PMA at birth and infant sex. The positive association between composite score and NAPI AO was lessened as PMA at birth increased and was higher for males than females. The interaction model explained 21% of the variance in AO scores.

Figure 3.

Association between composite inflammation score and NAPI AO.

Note. Composite inflammation scores include averages of interleukin (IL)-6, IL-8, IL-1 receptor antagonist, tumor necrosis factor-alpha, and monocyte chemoattractant protein-1 over time. Higher composite inflammation scores were associated with higher NAPI AO scores (p = .01), but this association was moderated by infant sex (p = .015) and post-menstrual age at birth (p = .002). Figure created with PROCESS version 3 (38) and SPSS version 26. NAPI, Neurobehavioral Assessment of the Preterm Infant; AO, alertness/orientation; PMA, post-menstrual age.

Discussion

Inflammation, which can be measured by quantifying circulating cytokine levels, may be an important predictor of long-term neurodevelopment in preterm infants (23). The most commonly reported cytokine predictors of neurodevelopment in preterm infants include interleukin IL-6, IL-8, TNF-α, and IL-1β (23). In this study, we found that average levels of TNF-α and average levels, as well as individual slopes, of IL-1RA predicted NAPI MDV scores in infants born between 28-31 weeks PMA. Average TNF-α levels, as well as average levels and individual slopes of IL-6, predicted NAPI AO. Although we did not find an association between IL-8 and neurobehavior in this study, other researchers have reported that elevated levels of IL-8 predict impaired cognition (41, 42), motor development (43), and behavioral outcomes (41, 44). We were unable to measure IL-1β in the majority of our samples; these levels were mostly extrapolated or not detectable.

Inconsistencies across studies may be due to sample selection, timing of cytokine measurement, or inherent differences in preterm infants. Some researchers have examined complex interactions of cytokines and neurodevelopment in preterm infants (45, 46) and have found that these interactions better predict neurodevelopment than specific cytokines (46). Differences in methods of collection and processing may account for the discrepancy between our results and those of other researchers. For example, serum samples tend to yield higher levels of IL-Ιβ than plasma samples due in part to the release of IL-1β during the clotting process (47). Similarly, IL-6, IL-8, IL-4, IL-10, and MCP-1 are typically higher in dried blood spots collected on filter paper than in plasma isolated from whole blood (48). Many researchers analyzing neonatal blood samples have used either serum (41, 42) or dried blood spots (29). Thus, the results across studies cannot be directly compared.

Our study revealed that IL-1RA was associated with neurodevelopment in preterm infants. This association has not been reported previously to our knowledge. In an integrative review of the cytokine predictors of neurodevelopment, Nist and Pickier (2019) included 37 publications from 20 independent studies and found only one study (46) that measured IL-1RA. Bass et al. (2008) found that neither IL-1β nor IL-1RA independently predicted white matter injury in a sample of preterm infants, but the interaction of these two cytokines improved the prediction model. In our study, IL-1RA, measured longitudinally as an average level and a slope, was a predictor of NAPIMDV. Researchers have reported high correlations among pro- and anti-inflammatory cytokines in newborns admitted to the NICU (49). In our study, IL-1RA may be a more sensitive predictor of short-term neurodevelopment than the other cytokines; IL-1RA was relatively simple to detect in our neonatal plasma samples using multiplex assay.

IL-1RA is important in regulating the damaging effects of IL-1β and, under normal conditions, is secreted in response to IL-1β (50). Infants deficient in IL-1RA suffer extreme systemic inflammation mediated by unopposed IL-1β production and must be rapidly treated with exogenous IL-1RA to prevent death (51). While IL-1β has been implicated in the neuronal injury that characterizes many disease processes, IL-1RA inhibits the inflammatory effects of IL-1β by competitively binding the IL-1 transmembrane receptor (51, 52). Using animal models, researchers have shown that peripherally administered IL-1β impairs myelination and affects subsequent behavior in mice (53), but peripherally administered IL-1RA prevents neurologic injury after exposure to infection (54). We suspect that IL-1RA was a proxy measure of IL-1β, which was more difficult to detect with our methods. Future research could continue the work of Bass et al. (2008) and examine how differences in the IL-ip/IL-IRA balance relate to neurodevelopment in preterm infants.

TNF-α was also associated with NAPIMDV and AO in our sample. Other researchers have also found an association between TNF-α and motor outcomes in preterm infants (43, 55). In these studies, higher levels of TNF-α predicted motor development on the Bayley Scales of Infant Development (BSID) at 2 years corrected age. While the association between TNF-α and NAPI MDV and AO scores may have been a spurious finding in our study, there is strong empirical evidence for the negative association between TNF-α and neurodevelopment in preterm infants (23).

We calculated average composite scores for IL-6, TNF-α, MCP-1, IL-8, and IL-1RA to better describe the inflammatory milieu without preference to any particular cytokine, a method that has been used by researchers studying various disease processes (36, 37). We found that composite scores predicted both NAPI MDV and AO in our sample. Higher composite scores, indicating higher average levels of IL-6, TNF-α, MCP-1, IL-8, and IL-1RA, predicted poorer NAPI MDV and better NAPI AO among females at the youngest PMA. The association of higher composite scores with poorer NAPI MDV was not surprising and is consistent with findings from other researchers reporting an association between IL-6, TNF-α, MCP-1, and IL-8 and abnormal motor development in preterm infants (20, 25, 55).

We were surprised to find that higher composite scores were associated with better NAPI AO scores, as only a few researchers have found higher levels of inflammatory cytokines to be predictive of better outcomes (20, 41). There may be several explanations for this finding. First, the association between composite scores and NAPI AO in our statistical models is a conditional effect that is dependent on infant sex and PMA at birth. There was a negative interaction between composite score and PMA at birth, suggesting that infants bom later than 28 weeks PMA experienced poorer NAPI AO as the composite score increased (Figure 3). Second, while NAPI AO scores predict BSID mental development scores at 18 and 30 months corrected age (56), these measures of neurodevelopment do not account for subtle cognitive impairments in executive function that are not apparent until late childhood. Recently, researchers have found that elevated levels of cytokines in preterm infants predict impairments in multiple domains of executive functioning at 10 years (57). These effects may not be detectable in early infancy and may not correlate with very early measures of neurobehavior.

In addition to the conditional associations between TNF-α, IL-1RA, IL-6, and composite scores and NAPI MDV and AO scores, there were also multiple interaction effects for various cytokines and either PMA at birth or infant sex on NAPI outcomes. We found that PMA at birth moderated the association between IL-1RA slopes and NAPI MDV and composite scores and NAPI AO. Compared to infants bom at 28 weeks PMA, higher levels of IL-1RA or composite z-scores predicted poorer NAPI outcomes in infants bom at a more advanced PMA. This may be due to age-related vulnerabilities of the developing brain (58). Using animal models, researchers have shown that inflammatory stimuli induce different cytokine profiles in the brain and different developmental outcomes based on the timing of exposure (59, 60).

In addition to PMA at birth, we also found that infant sex moderated the association between TNF-α slopes and NAPI MDV, the association between IL-8 slopes and NAPI AO, and the association between composite scores and NAPI AO. The timing and process of brain development, including neural migration, synaptic pruning, developmental apoptosis, and glial maturation, differ between males and females (61), therefore, inflammatory insults occurring at the same gestational age may affect males and females differently. Using a mouse model of prenatal inflammation, researchers have found that intrauterine inflammation induced immune responses in the CNS of offspring that differed by sex (62). Though studies of inflammation and neurodevelopment occasionally control for sex, very few examine interactions with sex (23). Infant sex may be an important moderator in models to predict infant neurodevelopment and an important future consideration for interventions to improve outcomes.

Our interaction models explained a modest percent of variance in NAPI MDV and AO scores. Given the multitude of factors that affect neurobehavior in preterm infants, this finding was expected as inflammation is only one biological variable that predicts outcomes in very preterm infants. Other potential predictors include neonatal apnea, clinical course, and stress exposure, among others (63).

Limitations

There are several limitations to our study. First, we were unable to quantify all the selected cytokines. IL-1β, IL-4, IL-10, and IL-17A levels were either extrapolated or not detectable in the majority of our samples. Thus, we could not include these potentially important cytokines in our analysis. Researchers could use filter paper to collect blood samples, which have been shown to be able to detect cytokines (29).

We used two clusters of the NAPI as measures of short-term neurobehavior. Although the MDV and AO clusters of the NAPI have adequate test-retest and excellent inter-rater reliability, the assessment is used primarily to evaluate the developmental maturation of preterm infants over time (35). Thus, infant performance on the MDV and AO tends to improve over time, and, ultimately, there is a ceiling effect for both clusters. Norm references are only available for infants between 32–37 weeks PMA. We planned the assessments for 35 weeks PMA to avoid this effect. However, some assessments in our study were delayed due to infants’ clinical conditions. In addition to measuring the relative maturation of infants over time, the NAPI can also discriminate among groups of infants based on degree of prematurity (56) and clinical acuity (35). However, compared to other neonatal developmental assessments, data on the predictive validity of the NAPI are limited (64). Constantinou et al. (2005) found that AO performance assessed at 36 weeks PMA was positively correlated with scores on the psychomotor and mental development indices of the BSID, 2^nd edition, at 18 and 30 months corrected age, but confirmatory data are lacking.

We examined NAPI MDV and NAPI AO as outcomes in our study of inflammation and neurobehavior and included multiple measures of inflammation (i.e. individual average cytokines, individual cytokine trends, composite scores), but we did not control for multiple comparisons, increasing the chance of a Type I error. Thus, our results should be interpreted with caution.

Finally, our results may not be generalizable to all preterm infants. The purpose of this study was to determine the association between inflammation and early neurobehavior in very preterm infants. Infants were excluded from enrollment for congenital infection, size small for gestational age, and exposure to confirmed maternal chorioamnionitis. Thus, the results of this study cannot be generalized to more high-risk or extremely preterm infants or infants with prenatal exposures to infection. In addition, there was very little racial and ethnic diversity in our sample and low variance for illness severity. Moreover, we were able to enroll only 73 of 232 eligible infants. Low enrollment was due to parental refusal and difficulties contacting parents of eligible infants. This low enrollment as well as limited sample diversity may have biased the results.

Conclusion

This study provides several important contributions to the study of inflammation and neurodevelopment in preterm infants. First, we have shown that IL-1RA, a cytokine rarely included in such studies, is a potentially important and sensitive measure in preterm infants. In contrast to the other cytokines included in our study, IL-1RA was readily detectable in our samples and significantly predicted NAPIMDV. In addition, we have shown that composite scores of inflammation may be better predictors of neurobehavior than individual cytokines. Composite scores are rarely used in studies of inflammation and neurodevelopment, but they account for average levels of multiple cytokines and may better quantify the inflammatory state of the infant compared to individual cytokines. We have demonstrated the importance of the moderation of the association between inflammation and neurobehavior by infant sex and PMA at birth in studies of preterm infants. Simply controlling for these variables may mask important associations that could influence the development of future interventions. Future studies could continue our work to examine repeated measures of inflammation on more long-term outcomes.