Individual variability in neural correlates of visual engagement in premature toddlers

Abstract

Prematurity disrupts in-utero processes that support neurocognitive development, with children born preterm at increased risk for long-term difficulties. Little is known about how prematurity affects early development of neural systems supporting visual engagement, or whether early variability in engagement relates to later attention outcomes. The present study examined behavioral and neural correlates of visual engagement in toddlers born prematurely using a developmentally appropriate computerized attention paradigm, involving attention capturing stimuli varying in duration, location, and motion while EEG was recorded. Visual engagement based on looking time and alpha power, a neural marker of cortical activation modulated by visual attention, were quantified in response to task stimuli and a resting baseline. Toddlers whose parents reported higher education levels showed greater overall looking time to visual stimuli. There was not a significant effect of gestational status on this behavioral response. In contrast, gestational status was associated with greater alpha power during rest relative to visual engagement. Toddlers born closer to term showed a larger rest–engagement difference in alpha power, whereas toddlers born at earlier gestational ages showed a smaller difference driven by lower resting alpha. This pattern in earlier-born toddlers may reflect delayed or disrupted maturation of this neural signature and may represent a neural substrate of prospective attention difficulties.

Highlights

• •Alpha power was higher during rest than during visual engagement across toddlers.
• •Toddlers born closer to term showed a larger rest–engagement alpha difference.
• •Earlier-born toddlers showed a smaller difference, driven by lower resting alpha.
• •Higher parental education was associated with longer looking to visual stimuli.

1. Introduction

Premature birth, defined as delivery before 37 weeks of gestation, affects approximately 11% of live births worldwide (Walani, 2020). Prematurity interrupts the in-utero developmental processes during which essential neurophysiological systems, such as the brain and vital organs, undergo rapid formation (Stiles, 2008). As a result, preterm children are at heightened risk for a range of long-term health and neurodevelopmental challenges. Cognitive development is particularly sensitive to early disruption following preterm birth, including more basic neurocognitive skills like visual attention, a foundational ability that supports environmental selection, learning, executive functioning, and self-regulation (Hendry et al., 2019). Importantly, less is known about how prematurity may influence the development of neural systems that underpin early visual attention in premature children, especially in the toddler years (12–36 months), or how variability in attentional development in these formative early life stages may contribute to broader neurocognitive outcomes.

1.1. Visual attention development

Visual attention is an early-developing process that undergoes substantial maturation throughout infancy and childhood (Oakes, 2023). A seminal framework conceptualizes attention as comprising three core components: orienting (aligning attention to stimuli), alerting (maintaining readiness to incoming stimuli), and executive control (monitoring conflict and problem-solving) (Colombo, 2001, Posner and Petersen, 1990, Posner and Rothbart, 2007). Together, these processes interact to encompass attention, undergo substantial development in the first years of life, and allow individuals to efficiently interact with the world (Landry et al., 2019, Mullane et al., 2016).

However, in early development, attentional processes do not operate in isolation. Attentional processes mature in concert with other domains such as fine motor skills, language, and executive functioning (Dandar et al., 2023, Lachambre et al., 2021). For example, orienting depends not only on attentional control but also on visual acuity and motor coordination; thus, early disruptions in motor development could indirectly impair the development of attentional skills, with downstream effects on higher-order cognitive functioning (Hendry et al., 2016, Veldman et al., 2019). This cascading developmental framework highlights the importance of studying attention within the broader context of developmental systems, where early vulnerabilities can shape trajectories across multiple domains over time (Iverson, 2021, Oakes, 2023).

Given these developmental interdependencies, examining voluntary looking behavior during engaging tasks provides a practical approach to examining early attentional capacity without requiring fine motor control or language comprehension. The ability to visually engage (to sustain gaze on a stimulus), emerges within the first months of life and shows marked variability. While less fine-grained than measures isolating specific attention components, visual engagement captures integrated behavioral output of multiple developing systems, including attentional control, arousal, regulation, and motivational processes (Geeraerts et al., 2019, Moyano et al., 2022). Individual differences in looking duration and stability during infancy have been linked to later cognitive outcomes. For example, around 6 months of age, infants begin to spend less time visually engaging with objects in their central field of view, a shift thought to reflect more efficient encoding and faster global feature processing. Shorter look durations have been associated with better later performance in memory and executive function tasks and are influenced by factors such as novelty preference and arousal reactivity (Colombo, 2001, Xie and Mallin, 2018, Xie and Richards, 2016, Xie and Richards, 2017).

With this perspective in mind, studying attentional processes during toddlerhood presents unique challenges. Toddlers may not yet have the language or motor skills necessary to follow complex task instructions, making traditional assessments difficult (e.g., Attention Network Test; Fan et al., 2002). To address this, researchers often adopt age-appropriate measures to assess attentional processes (Nakagawa et al., 2025, Xie and Richards, 2017). Despite these adaptations, many studies still restrict analyses to children who successfully complete experimental tasks, potentially overlooking important sources of variability. Particularly during infancy and toddlerhood, when engagement is naturally variable, treating failures to attend as mere noise may obstruct insights into the emergence of attentional control (De Barbaro et al., 2017). The observation of naturalistic variability in engagement duration itself may provide valuable information about resilience and vulnerability, especially in high-risk populations such as children born preterm.

Attention emerges through an interactive specialization process, wherein neural networks are shaped by dynamic interactions that include biological maturation and environmental experiences, and the child’s own activity and engagement (Johnson, 2011). Individual differences in attention, even in toddlerhood, likely reflect the cumulative influence of these factors rather than the isolated effects of any single domain. Within this framework, visual engagement represents an early precursor for more mature attention networks. Yet despite serving as a foundational neurocognitive building block for higher level cognition, relatively little is known about how visual engagement unfolds in prematurely born children, or other neurodevelopmentally at-risk groups, especially during toddlerhood. Examining neural mechanisms that support visual engagement in the first years of life may highlight pathways to potential risk in children born preterm.

1.2. Premature birth effects on visual attention development

Prematurity appears to alter typical attention trajectories, with some children demonstrating persistent deficits while others show evidence of improvement over time (Finke et al., 2015, Sripada et al., 2018). Longitudinal work shows that prematurely born infants as young as 5 months exhibit less efficient attentional behaviors (i.e. looking duration, shift rates, on task behavior, novelty scores) compared to their full-term counterparts, and these inefficiencies persisted across the first year of life (Rose et al., 2001). These behavioral markers of attentional inefficiency have been observed across development in other studies. In children born prematurely, the most common deficits include slower shifts to novel stimuli and less efficient spatial orienting in infancy (Van De Weijer-Bergsma et al., 2008), During toddlerhood, these children often show shorter sustained attention spans and greater distractibility (Burstein et al., 2021). In middle childhood, they tend to perform more poorly on tasks requiring sustained and shifting attention in middle childhood (Anderson et al., 2011, Wheelock et al., 2021). Even among adolescents who perform comparably to full-term peers on behavioral measures, distinct patterns of neural activation suggest that compensatory mechanisms may emerge over time (Fernandez et al., 2024).

Despite these well-documented behavioral findings, two major gaps remain in our understanding of how prematurity impacts attention development. First, relatively few studies have investigated the neural substrates supporting task-based visual engagement in preterm populations during the first years of life, limiting our understanding of early brain–behavior dynamics in this high-risk group. Second, there is a notable “data desert” in toddlerhood (12–36 months) for both clinical and normative populations, creating substantial gaps in neurodevelopmental models of attention for this age group (Kerr-German et al., 2023). One of the few studies examining attention in premature toddlers found that at 18 months, preterm children showed poorer orienting and alerting attention on eye-tracking tasks compared to full-term peers, whereas parental ratings of attention did not differ between groups, suggesting that early attentional challenges may not yet be readily observable in everyday behavior (de Jong et al., 2015). Given these early emerging behavioral vulnerabilities and the significant gaps in understanding toddlerhood research, characterizing neural correlations of visual engagement, will be an important next step in understanding how attentional systems develop in preterm children.

1.3. Neural indicators of visual attention

1.3.1. Normative samples

One way to systematically examine engagement during early development is through measurement of neural activity recorded during tasks that elicit visual attention. Electroencephalography (EEG) provides a powerful method for capturing developmental changes in neural functioning, particularly within the alpha frequency band, a range of oscillatory activity that has often mapped on to attention, arousal, and visual processing. Alpha oscillations emerge early in infancy; a dominant posterior rhythm appears around three months of age at approximately 3–5 Hz and increases to ∼6–7 Hz by 12 months, continuing to rise in peak frequency and power throughout childhood reaching around an 8 Hz peak by middle childhood (Stroganova et al., 1999). In terms of regional differences, evidence supports an increase in 5–9 Hz alpha power across occipital, central, and frontal regions in the first year of life (Elhamiasl et al., 2024).

Early infancy is characterized by sensory motor development which is also reflected in well-established patterns of alpha power development. For example, by around 6–7 months, infants show higher posterior alpha in darkness and reduced alpha in light or during visual stimulation (Stroganova et al., 1999), suggesting early differentiation between rest-like and visually engaged neural states. Throughout the first year of life, children can increasingly regulate their attention, and this is also reflected by changes in alpha power. By approximately 10–12 months, children show a reliable suppression in alpha power during a sustained visual attention task relative to rest and attention termination, which appears broadly over posterior, parietal, central, and frontal regions (Xie et al., 2018). Social motivation and anticipation is also shown to modulate alpha related attention. For example, in a classic naturalistic peek-a-boo paradigm, 8–11-month-old infants showed higher central alpha when attending to visible objects (externally controlled attention), whereas posterior alpha increased when anticipating the reappearance of a hidden face (internally controlled attention), highlighting differentiated neural pathways supporting emerging attentional abilities (Orekhova et al., 2001).

Alpha suppression, or the relative decrease in alpha power during attention engagement versus baseline or termination, is thought to reflect increased neural resources allocated to support attention engagement; higher alpha power, in contrast, is associated with reduced cortical excitability, (Foxe & Snyder, 2011), thus providing context for higher alpha during periods of darkness in visual-sensory paradigms or in baseline states, relative to active engagement. At around 9 months, infants exhibit greater central alpha suppression during interleaved action-observation sequences, which require trial-by-trial attentional switching, compared to blocked sequences where observation and execution trials are grouped together in predictable sets (Thinakaran et al., 2025), illustrating that alpha activity is already sensitive to fluctuations in attentional load before the more robust sustained-attention suppression patterns that emerge toward the end of the first year. Together, these findings suggest that alpha activity in infancy indexes not only whether attention is engaged, but also the degree of attentional control required to maintain and shift focus. However, little is known about how alpha power may serve as an early neural indicator of risk, particularly population vulnerable to attention difficulties such as preterm children.

1.3.2. Premature samples

Although alpha rhythms are sensitive to attentional demands in typically developing infants and toddlers, research examining alpha in preterm populations has primarily focused on resting-state or sleep recordings in infancy. For instance, in a sample of 26 preterm infants, spectral EEG frequency patterns measured during sleep predicted attention deficits at school age: infants who showed reduced delta and alpha power were more likely to perform poorly on attentional tasks at age 6 (Cainelli et al., 2021). Similarly, Magnetoencephalography (MEG), studies have revealed prematurity-related differences in resting-state alpha oscillations, very preterm school age children exhibited region-specific slowing of spontaneous alpha oscillations, particularly within posterior parietal and inferior temporal regions, and that this slowing was associated with visual perceptual deficits, but not with motor or IQ outcomes) (Doesburg et al., 2013). Other longitudinal work examining executive attention and alpha and theta PSD at ages 4 and 8 found that EEG indicators at age 4 did not significantly predict executive attention ability at age 8; however, frontal EEG power explained significant variance in executive attention concurrently at age 8. Authors suggest that the executive attention system may be too underdeveloped at age 4 to observe continuity across this developmental window, but (Joyce et al., 2018).

Together, these findings suggest that while resting state alpha measures in infancy can predict later attention outcomes, the sensitivity of alpha measures may depend on developmental timing and measurement context. To understand how prematurity shapes domain-specific neural systems supporting visual engagement, measuring EEG during developmentally appropriate, visually engaging tasks that capture fluctuations in attention, may reveal compensatory mechanisms or early markers of risk that are not captured by resting-state or sleep-based measures alone.

1.4. The current study

The purpose of the current study was to adopt a developmentally informed and ecologically grounded approach to examining visual attention and neural engagement in toddlers born preterm. Rather than restricting analyses to periods of successful attentional behavior, variability in visual engagement is modeled across a naturalistic visual attention task as a meaningful developmental indicator. Both behavioral measures of visual engagement (overall looking proportion) and neural measures of engagement (alpha power spectral density [PSD]) were examined. By linking behavioral and neural indices with biological (gestational age) factors, this study aims to advance understanding of early attentional development pathways in a high-risk population.

Specifically, this study addresses two aims. Aim one is to assess how visual engagement differs behaviorally across gestational status (from extreme to late prematurity) and child age (from early to late toddlerhood). Second, this project will examine developmental differences in alpha power during visual engagement (compared to rest) as a function of gestational age and child age. For aim one, it was hypothesized that children born at earlier gestational ages would display lower visual engagement. This effect may be moderated by child age, such that younger children born at the earliest gestational ages would show the lowest levels of visual engagement, given that visual attentional control typically increases with age. For the second aim, it was hypothesized that toddlers would show decreased alpha power during visual engagement compared to rest, and that this pattern would be most pronounced in older children born closest to term.

2. Methods

2.1. Participants

Participants were drawn from a sample of children who completed baseline behavioral and EEG assessments as part of a larger randomized clinical trial examining whether enhancing parental responsiveness promotes more neurotypical brain and behavioral development in toddlers born preterm. Of the 97 toddlers initially recruited for EEG data collection, 81 were retained in the final analytic sample. Sixteen participants were excluded for the following reasons: equipment malfunction resulting in unusable EEG triggers (n = 1) or broken ground electrode (n = 1); loss or desynchronization of EEG and behavioral video data (n = 2); inability to obtain valid looking-time data due to camera malfunction (n = 1); EEG Recorder software crash during acquisition (n = 1); insufficient usable EEG or looking-time data after preprocessing (n = 3); and data export or hand-editing errors that prevented reliable preprocessing (n = 2). Additionally, five participants were collected after data extraction for the present project was completed and were therefore not included in analyses.

Children (chronological ages 12–36 months) in this cohort varied in prematurity severity, defined as late (32–36 weeks), very (<32 weeks), and extremely (<28 weeks) premature. Participants were recruited from the UT-Health Neonatal High-Risk and General Pediatric Clinic. Exclusion criteria for our samples included the following: 1) presence of known/suspected congenital anomalies including chromosomal or complex congenital heart disease; 2) congenital infection including TORCH (Toxoplasmosis, Rubella, Cytomegalovirus, Herpes Simplex and others), untreated maternal HIV, or maternal syphilis; 3) bilateral grade 3/4 intraventricular hemorrhage, intraparenchymal hemorrhage, hydrocephalus; 4) Cerebral palsy with Gross Motor Function Classification of III or higher; 5) blindness; 6) deafness; 7) Current maternal drug use or maternal drug use during pregnancy; 8) Maternal age under 18-years when child was born; 9) families who reside outside the catchment area (>1 h drive from the Texas Medical Center).

Parent education was used as a proxy for socioeconomic status. Parent educational attainment ranged from 1 (Primary School) to 10 (9 = “Doctorate or advanced degree” and 10 = “Other”), with a mean and median of 6.0 = “Some college but no degree”. Income ranges from < $50,000 per year for the household to > $200,000 per year. Income ranged from $75,000 to $100,000. Participants completed the study at the University of Houston Health 1 Building. The Institutional Review Board at the University of Houston and the University of Texas Health and Science Center approved all recruitment and experimental procedures for this study. Informed consent was obtained from all caregivers. Assent was provided on behalf of the participants by their caregivers, as participants were too young to provide assent themselves.

2.2. Measures

2.2.1. Visual engagement paradigm

The task was an measure of visual attention (Richards, 1997), presented using Presentation software (Neurobehavioral Systems ©). Children sat on their caregiver’s lap, 50 in. from a 21 × 12-inch monitor. They viewed stationary movie clips (30 frames/second) of Sesame Street characters displayed in 4 × 2-inch windows positioned on the left or right side of the screen, see Fig. 1 to see an example stimulus. After 8–12 s, clips disappeared, reappeared, or shifted position. Instrumental music accompanied each trial and stopped between trials. The task consisted of 5 blocks of 14 trials each. For each block the same video played on a loop for each trial, the stimuli would only change at the start of a new block. Parents were instructed to remain silent to minimize distraction. When children were not looking, research assistants silently pointed to the monitor to reorient attention. A single front-facing camera recorded the child’s face, and a mirror positioned behind the child reflected the on-screen stimuli to facilitate precise looking-time coding.

Fig. 1.

Visual Engagement Stimulus. Above is an example stimulus from the visual engagement paradigm.

2.2.2. EEG recording and preprocessing

Resting State Paradigm. Resting-state EEG was recorded for three minutes while children sat on their parent’s lap in the same location as the visual engagement paradigm. To maintain quiet, neutral attention during the recording, a research assistant sat directly in front of the child and continuously spun a bingo-ball wheel containing three brightly colored toys. This minimally engaging visual baseline is a widely used approach for eliciting resting-state EEG in infants and young children and has been used in several prior studies (MacNeill et al., 2018; Marshall et al., 2002; Norton et al., 2021). For each participant, average alpha power (6–9 Hz) across the full three-minute period was computed state.

EEG Acquisition. EEG was recorded using Brain Vision Recorder software (Brain Products, Munich, Germany) and a 64-channel actiCHamp system. Electrode impedances were checked prior to recording and maintained below 20 kΩ. Online recordings used FCz as the reference, re-referencing to an average occurred during pre-processing (described below). The sampling rate was 1000 Hz for most subjects, although the first seven subjects used a sampling rate of 500 Hz. Data were downsampled to 250 Hz making this issue trivial After each recording session, research assistants visually inspected recordings to identify excessive noise, drift, or flat channels prior to preprocessing.

Preprocessing Pipeline. EEG preprocessing was conducted in MATLAB using the Batch Electroencephalography Automated Processing Platform (BEAPP; Levin et al., 2018). Raw EEG files were first converted into BEAPP-compatible format. Next, the PREP module was applied to remove 60-Hz power line noise using CleanLine algorithms. After PREP, data underwent band-pass filtering (1–100 Hz) and were then resampled to 250 Hz to standardize the sampling rate across participants. Artifact correction was performed using the ICA module of the Harvard Automated Processing Pipeline for Electroencephalography (HAPPE; Gabard-Durnam et al., 2018), which implements wavelet-enhanced ICA followed by automatic component rejection using MARA to identify non-neural artifacts. Bad channels were interpolated after artifact correction. Following artifact removal, the data were re-referenced to the average reference across all scalp channels. A linear detrending step was then applied to remove slow drifts in the signal.

For the visual engagement paradigm, event-related segmentation was conducted by identifying looking-related event markers and extracting 1-s epochs time-locked to the child’s looking behavior. Epochs exceeding ±200 μV or containing residual artifacts were excluded, and poor-quality channels were interpolated using spherical spline interpolation. For the resting-state data, the same preprocessing pipeline was applied, with the continuous 3-minute recording segmented into consecutive 1-s epochs. Artifactual segments were removed using the same ±200 μV criterion. Power spectral density (PSD) was computed using the BEAPP PSD module, which applies Welch’s method with a 2-s Hanning window and 50% overlap (FFT length = 256), implemented through the FieldTrip toolbox. Alpha-band power (6–9 Hz) was extracted for each channel, and PSD values were log₁₀-transformed (µV²/Hz) prior to statistical analyses. Alpha power was analyzed across four EEG regions (frontal, central, parietal, and occipital); see Fig. 2 for a depiction of how these regions were defined.

Fig. 2.

EEG Regions for Analyses. Each color denotes a region analyzed in the scalp analyses. The three sections for each region represent the hemispheres (left, midline, right).

Quality Control. For the visual engagement task, participants contributed an average of 218.8 usable looking-condition segments (SD = 46.9) after artifact rejection. Before interpolation, approximately 24.5 channels (SD = 8.7) were marked as bad, after which BEAPP restored full 64-channel montages for analysis. For the resting-state recording, prior to interpolation, children exhibited an average of 25.3 bad channels (SD = 8.6), which is typical of toddler EEG. Following HAPPE's automatic channel-level artifact correction and interpolation procedures, all recordings were restored to a full 64-channel montage for power spectral estimation. After segment-level artifact rejection, participants contributed an average of 193 clean 1-s segments (SD = 14.8). The retained data quantities and preprocessing outcomes meet or exceed published methodological recommendations for developmental EEG, including thresholds established by the HAPPE pipeline (Gabard-Durnam et al., 2018). Three participants did not meet the minimum requirement of 40 s of clean resting EEG and were excluded from analyses, as noted in the Participant section of this manuscript.

2.2.3. Visual engagement (looking behavior)

Video recordings of the task were coded frame-by-frame to identify the onset and offset of each instance in which a child looked at the stimulus. Coding was conducted post–data collection using Shotcut video-editing software (Meltytech, LLC, 2025). For each trial, the total duration of looking was divided by the trial length to derive a proportion of time spent looking. A global measure of visual engagement was then calculated for each child by averaging these trial-level proportions across the entire task. This looking-time proportion served as an index of between-subject variability in overall engagement with the visual stimuli and was used as the primary behavioral measure of visual attention in subsequent analyses.

3. Results

3.1. Descriptive statistics

All variable distributions were inspected in R (v4.4.1) to assess normality and outliers. Outliers were defined using Tukey’s criterion, values falling below the first quartile minus 1.5 times the interquartile range or above the third quartile plus 1.5 times the interquartile range, a method commonly used in developmental EEG research (Kulke et al., 2017, Tukey, 1997). Identified outliers were winsorized to the nearest non-outlier value (Kwak and Kim, 2017). Skewness and kurtosis were computed using the psych package (v2.3.6), applying widely used interpretive norms of |skewness| ≤ 1 and |excess kurtosis| ≤ 4. Based on these criteria, frontal, central, parietal, and occipital alpha PSD variables, and number of segments per participant required winsorizing. A small proportion of participants (9–12 per region; ∼10–15% of the sample) contributed PSD values identified as outliers and underwent winsorizing. For the number of segments, 12 participants (14.8%) were identified as outliers based on the outlier criteria. All other variables met normality criteria and did not require adjustment (Table 1).

Table 1.

Participant characteristics and EEG alpha power.

Variable	n	% or M	SD	Range
Demographics
Total sample	81	100.0%	--	--
Sex (male)	47	58%	--	--
Ethnicity (Hispanic/Latino)	45	55.6%	--	--
Race (White)	41	50.6%	--	--
Race (Black/African American)	26	32.1%	--	--
Race (Native Hawaiian or Other Pacific Islander)	1	1.2%
Race (Asian)	4	4.9%
Race (American Indian or Alaska Native)	2	2.5%
Parent education (≤ high school)	26	32.1%	--	--
Parent education (some college)	20	24.7%	--	--
Parent education (bachelor's degree)	20	24.7%	--	--
Parent education (graduate degree)	8	9.9%	--	--
Other	6	7.4%	--	--
Participant characteristics
Gestational age (weeks)	81	30.36	4.78	23.00–36.00
Chronological age (months)	81	23.31	4.94	14.00–35.82
Looking Proportion	81	0.61	0.18	0.15–0.90
EEG alpha power by region
Frontal Alpha Power	81	-0.46	0.45	-1.45–0.58
Central Alpha Power	81	-0.54	0.53	-1.76–0.74
Parietal Alpha Power	81	-0.61	0.53	-1.83–0.67
Occipital Alpha Power	81	-0.50	0.52	-1.63–0.74
EEG alpha power by engagement condition
Looking Alpha Power	81	-0.59	0.42	-1.67 – 0.02
Rest Alpha Power	81	-0.46	0.52	-1.67– 0.68
Number of Segments Per Condition (1 s)
Looking	81	218.8	46.9	88–405
Rest	81	192.8	14.8	180–271

Note. Descriptive statistics for participant characteristics, looking proportion, and EEG alpha power across regions and conditions.

To address aim one, multiple linear regression was used to assess how visual engagement was influenced by gestational age and child age. Linear mixed-effects models were conducted using lme4 (v1.1–34) to examine aim two, which was to assess developmental differences in alpha power during visual engagement as a function of gestational age and child age. For both analyses, parent education was included as a covariate for socioeconomic status. Parent education contained less missing data, had more variability, and was closer to normality compared to household income. Aim two mixed-effect models were corrected for multiple comparisons using False Discovery Rate (FDR)(Benjamini and Hochberg, 1995).

For linear regression models, assumptions were assessed via Q–Q plots of residuals, residuals-versus-fitted plots, and autocorrelation functions (ACF) of residuals. For mixed-effects models, models were estimated via restricted maximum likelihood (REML) and tested using Satterthwaite-approximated degrees of freedom via the lmerTest package (v3.1.5). Assumptions were evaluated through Q–Q plots of both residuals and random effects, residuals-versus-fitted plots, and ACF plots of residuals. No serious violations of assumptions were detected for either model type.

3.2. Aim 1: To examine whether visual engagement varies by age and gestational age

Multiple linear regression model examined whether gestational age, chronological age, parental education, and sex predicted children’s total looking proportion (the indicator of visual engagement) across the task. The model explained 5.5% of the variance in looking proportion, adjusted R² = .055, F(4, 74) = 2.75, p = .084. Gestational age was not a significant predictor, β = 0.107, SE = 0.11, t(74) = 0.96, p = .340, nor was chronological age, β = 0.043, SE = 0.11, t(71) = 0.384, p = .701 of looking proportion. Parental education significantly predicted looking proportion, β = 0.128, SE = 0.050, t(72) = 2.56, p = .012 such that higher education levels were associated with greater looking proportion to task stimuli (See Fig. 3). The interaction between gestational and chronological age was not significant, β = –0.125, SE = 0.107, t(72) = –1.17, p = .246. See Table 2. regression results.

Fig. 3.

Parent Education on Looking Proportion. Association between parental education and toddlers’ looking proportion during the visual attention paradigm. The black line represents the fitted regression line with 95% confidence intervals (gray shading), and individual data points are shown in gray (looking proportion values are centered).

Table 2.

Multiple linear regression predicting looking duration from gestational age, child age.

Parameter	β	SE	t	p
(Intercept)	-0.776	0.311	-2.50	0.015	*
Chronological Age	0.043	0.111	0.38	0.702
Gestational Weeks	0.107	0.111	0.96	0.340
Parent Education	0.128	0.050	2.56	0.013	*
Chronological Age × Gestational Weeks	-0.125	0.107	-1.17	0.246

Note. R² = 0.104, Adjusted R² = 0.055, F(4, 74) = 2.14, p 0.084. * p < .05, ** p < .01, *** p < .001

3.3. Aim 2: To assess developmental differences in alpha power during visual engagement as a function of gestational age and child age

Mixed effects repeated measure ANOVAs were conducted. Separate models were conducted for each of the four EEG regions: frontal, central, parietal occipital. Each model included the centered and standardized alpha power (6–9 Hz) in for each region. For each model, child age and gestational age were included as between subject predictors (also standardized and centered). Engagement condition was included as a within-subject factor with two levels (looking, resting), and lastly hemisphere was included (left, midline, right) to test for laterality effects. Gestational age, child age, and condition were included as interaction terms, number of segments of EEG data was a covariate, and participant was included as a random effect in the model. Parent education and sex were not included as covariates as they did not improve model fit and were not significantly associated with the variables of interest.

3.4. Frontal

For the frontal model, there was a significant main effect of task-related engagement (looking vs resting), F(1409)= 14.04, p(FDR) = <.001, ηp²= 0.033, see Fig. 4. The estimated marginal means indicated lower frontal alpha power during task-related visual engagement (M = −1.05, SE = 0.005, 95% CI [−1.06, −1.04]) compared with the resting condition (M = −1.04, SE = 0.005, 95% CI [−1.05, −1.03]). See Table 3 for ANOVA results. There were no other effects that survived multiple corrections for the frontal region.

Fig. 4.

Main effect of task-related engagement: Frontal alpha. Displays standardized frontal alpha power while participants are visually engaged with stimuli versus at rest.

Table 3.

Analysis of Variance Results for Frontal Alpha Power.

Effect	SS	NumDF	DenDF	F	p	p(FDR)	ηp²
Condition	0.01	1	409	14.04	<.001	<.001***	0.033
Gestational Weeks	0.00	1	77	0.64	0.427	0.427	0.008
Chronological Age	0.00	1	77	1.23	0.271	0.271	0.016
Hemisphere	0.00	2	398	4.20	0.016	0.070	0.021
Number of Segments	0.00	1	430	0.62	0.432	0.539	0.001
Condition × Gestational Weeks	0.00	1	400	3.19	0.075	0.225	0.008
Condition × Chronological Age	0.00	1	398	2.67	0.103	0.233	0.007
Gestational Weeks × Chronological Age	0.00	1	77	0.34	0.560	0.560	0.004
Condition × Gestational Weeks × Chronological Age	0.00	1	398	0.50	0.479	0.539	0.001

Note. SS = sum of squares; NumDF = numerator degrees of freedom; DenDF = denominator degrees of freedom; p(FDR) = false discovery rate-corrected p-value; ηp² = partial eta-squared. * p < .05, ** p < .01, *** p < .001

3.5. Central

The central model yielded a significant main effect of task-related visual engagement, F(1,409)= 14.57, p(FDR) = <.001, ηp²= 0.034, see Fig. 5. Estimated marginal means indicated lower central alpha power during task-related visual engagement (M = −1.05, SE = 0.005, 95% CI [−1.06, −1.04]) compared with the resting condition (M = −1.04, SE = 0.005, 95% CI [−1.05, −1.03]), indicating a moderate reduction in central alpha power during visual engagement relative to rest. There was also a main effect of hemisphere, F(2,398)= 5.40, p(FDR) = 0.022, ηp²= 0.026. Pairwise comparisons with FDR correction revealed that midline alpha power was significantly higher than right hemisphere alpha power, ∆M = 0.010, SE = 0.003, t(398) = 3.27, p = 0.004). No significant differences were found between left and midline hemispheres (∆M = −0.006, SE = 0.003, t(398) = -1.96, p(FDR) = 0.076) or between left and right hemispheres (∆M = 0.004, SE = 0.003, t(398) = 1.30, p(FDR) = 0.193). See Table 4 for full ANOVA results. These results suggest slightly greater central alpha power at midline relative to the right hemisphere, with no significant differences between left and either other region (Fig. 6). See Table 4 for full ANOVA results. No other effects survived FDR correction.

Fig. 5.

Main effect of task-related engagement: Central alpha. Displays standardized central alpha power while participants are visually engaged with stimuli versus at rest.

Table 4.

Analysis of Variance Results for Central Alpha Power.

Effect	SS	NumDF	DenDF	F	p	p(FDR)	ηp²
Condition	0.01	1	409	14.57	<.001	<.001***	0.034
Gestational Weeks	0.00	1	77	0.73	0.396	0.396	0.009
Chronological Age	0.00	1	77	1.35	0.249	0.447	0.017
Hemisphere	0.01	2	398	5.40	0.005	0.022	0.026
Number of Segments	0.00	1	429	0.46	0.496	0.558	0.001
Condition × Gestational Weeks	0.00	1	400	2.17	0.142	0.426	0.005
Condition × Chronological Age	0.00	1	398	1.34	0.248	0.447	0.003
Gestational Weeks × Chronological Age	0.00	1	77	0.95	0.333	0.499	0.012
Condition × Gestational Weeks × Chronological Age	0.00	1	398	0.04	0.839	0.839	0.000

Note. SS = sum of squares; NumDF = numerator degrees of freedom; DenDF = denominator degrees of freedom; p(FDR) = false discovery rate-corrected p-value; ηp² = partial eta-squared. * p < .05, ** p < .01, *** p < .001

Fig. 6.

Main effect of hemisphere: Central alpha. Displays standardized central alpha power for each hemisphere (left, midline,right), averaged across engagement condition.

3.6. Parietal

In line with the previous two models, the parietal model also revealed a significant main effect of task-related engagement, F(1,406)= 17.74, p(FDR) = <.001, ηp²= 0.042, such that the visual engagement (looking) showed significantly less alpha power compared to rest. There was also a significant main effect of hemisphere in parietal alpha, F(2,388)= 6.92, p(FDR) = .005, ηp²= 0.034. Pairwise comparisons with FDR correction showed that this effect was driven by significantly greater alpha in the left hemisphere compared to the midline, ∆M = 0.01, SE = 0.003, t(398) = 3.72, p(FDR) < .001. There were no significant differences in alpha power between the left and right hemisphere (∆M = 0.005, SE = 0.003, t(398) = 1.70, p(FDR) = 0.090) or between the midline and right hemisphere (∆M = 0.006, SE = 0.003, t(398) = -2.02, p(FDR) = 0.067), see Fig. 7. Importantly, in addition to the main effect of condition, there was a significant interaction between task-related engagement and gestational age, F(1,399)= 6.21, p(FDR) = .039, ηp²= 0.015. Post-hoc simple-slope tests revealed that gestational age did not significantly associate with alpha power across either condition: Looking (β = 0.001, SE = 0.006, t(84) = 0.13, p = .897) and Resting (β = 0.007, SE = 0.006, t(383) = 1.16, p = .249), and that the significant interaction was instead driven by the stronger and more positive in the resting versus looking condition. A second set of follow-up comparisons was conducted to probe the interaction by comparing effects of task-related engagement at different gestational age levels (-1 SD, mean, +1 SD). Results indicated that early gestation (−1 SD, 25.6 weeks) was not associated with significant differentiation in alpha power between task-related visual engagement and resting conditions, (Δ = −0.006, SE = 0.004), t(400) = -1.70, p(FDR) = .090), however, the mean gestational age (30.4 weeks), task-related engagement significantly differed, (Δ = −0.012, SE = 0.003, t(406) = -4.21, p(FDR) < .001), with a more pronounced difference at later gestational ages (+1 SD, 35.1 weeks),): Δ = -0.018 (SE = 0.004), t(405) = -4.58, p(FDR) < .001. A Johnson–Neyman analysis showed that the task-related engagement became significantly different above approximately 26.08 weeks gestation (0.89 SD), indicating that the closer toddlers were born to term, the more pronounced increases in parietal alpha power during rest relative to visual engagement. See Fig. 8 and Table 5 for the full results for the parietal model.

Fig. 7.

Main effect of hemisphere: Parietal alpha. Displays standardized parietal alpha power for each hemisphere (left, midline, right), averaged across engagement condition.

Fig. 8.

Interaction between gestational age at birth and parietal alpha power during task-related engagement. Predicted standardized parietal low-alpha power (6–9 Hz) as a function of gestational age at birth for the looking (solid line) and resting (dashed line) conditions. Shaded regions represent standard error.

Table 5.

Analysis of variance results for parietal alpha power.

Effect	SS	NumDF	DenDF	F	p	p(FDR)	ηp²
Condition	0.01	1	406	17.74	<.001	<.001***	0.042
Gestational Weeks	0.00	1	77	0.44	0.511	0.511	0.006
Chronological Age	0.00	1	77	1.91	0.171	0.386	0.024
Hemisphere	0.01	2	398	6.92	0.001	0.005	0.034
Number of Segments	0.00	1	421	0.41	0.523	0.589	0.001
Condition × Gestational Weeks	0.00	1	399	6.21	0.013	0.039	0.015
Condition × Chronological Age	0.00	1	398	0.71	0.400	0.589	0.002
Gestational Weeks × Chronological Age	0.00	1	77	0.85	0.360	0.589	0.011
Condition × Gestational Weeks × Chronological Age	0.00	1	398	0.01	0.939	0.939	0.000

Note. SS = sum of squares; NumDF = numerator degrees of freedom; DenDF = denominator degrees of freedom; p(FDR) = false discovery rate-corrected p-value; ηp² = partial eta-squared. * p < .05, ** p < .01, *** p < .001

3.7. Occipital

For the occipital alpha mixed effect repeated measures ANOVA, there was a significant main effect of task-related engagement F(1, 407) = 13.94, p < .001, ηₚ² = .033, such that alpha power was significantly reduced during looking compared to rest, Fig. 9. All other main effects and interactions were non-significant (see Table 6 for full results).

Fig. 9.

Main effect of task-related engagement: Occipital alpha. Displays standardized occipital alpha power while participants are visually engaged with stimuli versus at rest.

Table 6.

Analysis of variance results for occipital low-alpha power.

Effect	SS	NumDF	DenDF	F	p	p(FDR)	ηp²
Condition	0.01	1	407	13.94	<.001	< .001	0.033
Gestational Weeks	0.00	1	77	0.19	0.663	0.663	0.002
Chronological Age	0.00	1	77	1.31	0.257	0.647	0.017
Hemisphere	0.00	2	398	0.22	0.801	0.801	0.001
Number of Segments	0.00	1	424	0.62	0.430	0.658	0.001
Condition × Gestational Weeks	0.00	1	399	3.39	0.066	0.298	0.008
Condition × Chronological Age	0.00	1	398	1.13	0.288	0.647	0.003
Gestational Weeks × Chronological Age	0.00	1	77	0.61	0.439	0.658	0.008
Condition × Gestational Weeks × Chronological Age	0.00	1	398	0.33	0.564	0.725	0.001

Note. R² = 0.104, Adjusted R² = 0.055, F(4, 74) = 2.14, p 0.084. * p < .05, ** p < .01, *** p < .001

4. Discussion

This study examined the neural correlates of task-evoked visual engagement in a sample of preterm toddlers. Aim 1 investigated whether levels of visual engagement varied by child age and gestational age. Aim 2 assessed whether alpha power varies as a function of task-related engagement, gestational age, child age, and EEG region. Contrary to hypotheses, neither child age nor gestational age were significantly associated with visual engagement during the visual engagement task. Instead, parent socioeconomic status, based on primary caregiver education level, was most strongly correlated with behavioral indicators of attention. These results indicate that contextual factors, specifically socioeconomic conditions, may play a more prominent role in shaping early sustained visual engagement than the level of prematurity alone.

While unexpected, this finding aligns with other work showing neural and cognitive development is highly shaped by environmental experiences. Specifically, socioeconomic status, a multidimensional construct encompassing factors such as income, caregiver education, and neighborhood context, has been associated with differences in cognitive performance, visual attention, and neurodevelopment (Rakesh et al., 2025). Through mechanisms including stress exposure, variability in support, and access to enriching experiences, SES-related disparities in neural function may not always be evident by infancy but may become more apparent across childhood, particularly within neural systems supporting language and executive functioning (Brito et al., 2016). For example, preschoolers from lower-SES backgrounds exhibit reduced sustained attention and greater impulsivity (Hoyer et al., 2023). Because the present sample consisted exclusively of preterm children, incorporating a full-term comparison group in future work will be essential for clarifying the relative contributions of prematurity and socioeconomic context.

Several explanations may account for the lack of gestational age differences in overt visual engagement measured behaviorally. First, it is possible that prematurity-related neural differences may arise from underlying cortical mechanisms rather than manifesting as clear behavioral differences during toddlerhood. Second, the behavioral index for visual engagement may be too coarse to show sensitivity to gestational effects. In other words, the global measure of visual engagement (looking proportion averaged across the task) may have washed out more fine-grained engagement patterns. Time-varying measures such as changes in looking across the task or fluctuations between individual bouts, as well as distributional features such as the longest sustained look, may reveal gestational age effects not captured by overall looking. A third explanation may be that behavioral differences in visual engagement among children born preterm may not emerge until later in childhood. Prior work suggests that attention-related difficulties in premature children often become more pronounced with age, or conversely, may have already “caught up” to their gestationally older peers by toddlerhood (Finke et al., 2015), making differences difficult to detect at this developmental window. Finally, the effects of prematurity are heterogenous, with multiple co-occurring factors, such as weight, length of NICU stay, environmental context, and perinatal complications, that likely interact with gestational age to shape developmental outcomes. These intersecting influences may further contribute to the variability observed in this population (Aarnoudse-Moens et al., 2009, Walani, 2020).

Although there were no gestational age-related differences in visual engagement, the EEG scalp analyses revealed interactions between gestational age and engagement conditions. Specifically, there was a significant interaction in parietal alpha power, with toddlers born closer to term exhibiting larger differences between visual engagement and rest. These children demonstrated a reliable attenuation of alpha power during visual engagement relative to rest. Such alpha suppression is a well-established marker of cortical excitation and is commonly observed during periods of sustained attention and visual processing (Woodman et al., 2022). This pattern has been documented as early as 12 months of age in typically developing infants (Xie et al., 2018) and continues to strengthen throughout later childhood and into adulthood (Woodman et al., 2022). In contrast, toddlers born at the earliest gestational ages (<25 weeks) did not show significant alpha differences between engagement and rest. This absence of modulation suggests that extreme prematurity may disrupt or delay the maturation of this neural signature during toddlerhood, placing these children at heightened risk for atypical development of sustained attention mechanisms. This interpretation aligns with prior work showing that children born earliest in gestation exhibit the greatest attentional vulnerabilities (Baron and Rey-Casserly, 2010, Gale-Grant et al., 2022).

In addition to the gestational effects on task-related engagement on parietal alpha, there was a consistent main effect of engagement conditions in frontal, central, and occipital regions. Alpha power was greater during rest than during visual engagement. Importantly, parietal alpha also showed this same directional pattern, but the magnitude of this difference varied by gestational age, as described above. This broader pattern indicates that alpha modulation in response to visual engagement is detectable across the scalp in toddlers and may follow region-specific developmental trajectories in the context of prematurity.

The absence of age-related differences in visual engagement in this sample may reflect measurement limitations noted above (e.g., lack of time-varying gaze metrics), but it may also reflect the developmental timing of emerging attentional abilities. Early behavioral attention skills often become more overt later in childhood, whereas neural signatures, such as alpha power, may show sensitivity earlier in development. This pattern is consistent with prior work showing that early neural oscillations are predictive of later cognitive outcomes. For example, Cainelli et al. (2021) found that alpha PSD recorded in preterm neonates at 35 weeks post-conception predicted visual attention at age six, even in the absence of direct associations between gestational age and later attention outcomes. The current findings extend this pattern by demonstrating that gestational age differences can emerge in neural signatures of visual engagement during toddlerhood, even when behavioral measures remain similar. This supports the view that early neural alterations related to prematurity may precede, and in some cases normalize before, the emergence of overt behavioral differences. Evidence from Nordvik et al. (2022) showed that neonatal spectral power across EEG bands predicted global cognitive outcomes and IQ at 10–12 years of age, further underscoring the developmental significance of early neural oscillatory patterns. Although that study did not examine visual attention specifically, it reinforces the notion that neural markers in infancy and early childhood can provide an early window into later attentional and cognitive outcomes in children born preterm. Together, these findings suggest that neural signatures of attention may exhibit developmental sensitivity before behavioral measures do, helping to explain the absence of age effects in the present study despite meaningful variation in neural modulation of alpha power. The present results offer novel insight into the early maturation of neural systems supporting visual engagement in toddlers born preterm, an understudied age range in developmental cognitive neuroscience (Kerr-German et al., 2023).

Although this study provides valuable insights into the early neural development of visual attention in children born preterm, several limitations should be acknowledged. First, the indicator used to operationalize visual engagement, looking proportion, captures only a global summary of gaze. Eye gaze and looking behaviors are developmentally dynamic and meaningful precursors to attentional processes in early life. However, a single aggregated measure of total looking proportion likely obscures important behavioral variability. Time-varying measures, such as fluctuations in engagement across the task, fatigue effects, or the duration and stability of individual looking bouts, may have revealed more fine-grained patterns.

In addition, the task did not include a defined competing stimulus or a clear target stimulus (Esterman and Rothlein, 2019), which limits the inferences that can be drawn about specific attentional processes such as orienting, vigilance, or distractor suppression. For this reason, we conceptualize the behavior measured here as visual engagement rather than a more precise index of sustained attention. Similarly, neural analyses relied on mean alpha power across the entire looking period relative to rest, which removes trial-to-trial or bout-to-bout fluctuations in neural activity. These global averages also do not account for stimulus motion or spatial position, factors that may meaningfully shape how toddlers born preterm process visual information.

To further clarify the effects of prematurity on visual engagement and attentional processes in toddlerhood, future work should include full-term comparison groups. Even in normative samples, cognitive neurodevelopment remains understudied in toddlerhood, and having benchmarks of typical variability will strengthen the interpretability of findings in preterm cohorts. Additional next steps should include EEG connectivity analyses to understand how alpha activity across scalp regions operates as a network during unfolding visual engagement. Such work may shed light on the unique effects of prematurity on parietal alpha patterns and how these differences relate to the organization of early attention networks.

Together, the present findings offer novel insight into how toddlers born prematurely engage with visual stimuli at both behavioral and neural levels. Although gestational age was not associated with overt visual engagement, socioeconomic context emerged as a meaningful predictor of looking behavior, underscoring the substantial influence of environmental factors on early attentional development. In contrast, the neural results revealed early signs of gestational age–related differences: toddlers born closer to term showed clear parietal alpha attenuation during visual engagement, whereas those born extremely preterm showed little modulation, suggesting delayed or disrupted maturation of this neural signature. At the same time, alpha suppression across frontal, central, and occipital regions indicates that neural responsiveness to engagement is already broadly detectable in toddlerhood. By leveraging naturalistic variability in engagement, this study highlights the importance of ecologically grounded approaches for capturing early developmental heterogeneity in preterm children. Future work, including full-term comparison groups, more temporally sensitive behavioral indices, and network-level EEG analyses, will be critical for clarifying how prematurity and environmental context interact to shape the emergence of visual attention. Such research can ultimately inform earlier identification of risk and guide developmentally targeted interventions for children born preterm.

CRediT authorship contribution statement

Haley Laughlin: Writing – review & editing, Writing – original draft, Visualization, Investigation, Formal analysis, Data curation, Conceptualization. Anna Galvan: Visualization, Project administration, Methodology, Investigation. Andrea Ortiz-Jimenez: Project administration, Investigation, Data curation. Dana DeMaster: Resources, Project administration, Investigation, Funding acquisition. Johanna Bick: Writing – review & editing, Supervision, Resources, Project administration, Investigation, Funding acquisition. Xinge Li: Resources, Methodology, Investigation, Data curation. Marguerite Knudtson: Methodology. Kelly Barry: Resources, Methodology. Susan Landry: Project administration, Funding acquisition.

Funding

This paper was supported by funding from NIH R01HD100560 to Bick, DeMaster, and Landry.

Declaration of Competing Interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Johanna Bick, Dana DeMaster, Susan Landry reports financial support was provided by National Institute of Child Health and Human Development. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

Data will be made available on request.