Neonatal brain network growth tracks early cognitive, language and motor outcomes in critical congenital heart disease

Infants born with critical congenital heart disease face a well-documented elevated risk of abnormal brain development and later neurodevelopmental impairment, but clinicians have long lacked reliable early markers of which babies will struggle. A new prospective longitudinal cohort study published in Pediatric Research reports that the rate at which whole-brain network integration develops during the neonatal perioperative period is associated with cognitive, language and motor outcomes measured between one and four years of age. The finding comes from a team led by Lauren Harasymiw and Patrick McQuillen at the University of California, San Francisco, and suggests that the trajectory of early brain network maturation, rather than any single scan, may serve as a sensitive indicator of neurodevelopmental risk and resilience in this population.

The study enrolled neonates with critical congenital heart disease, defined as congenital heart disease requiring an operation in the neonatal period, at a single tertiary care center between November 2011 and January 2025, with follow-up extending through August 2025. Of 117 neonates enrolled, 97 met the inclusion criteria for the current analysis, which was restricted to two diagnostic groups to minimize physiological heterogeneity: 51 infants with d-transposition of the great arteries and 46 with single ventricle physiology, including hypoplastic left heart syndrome and related variants. Preterm infants born before 36 weeks, infants with confirmed or suspected genetic or malformation syndromes, and infants with contraindications to MRI were excluded. Written informed consent was obtained from parents or legal guardians, and the study followed the STROBE reporting guideline.

Each infant underwent brain MRI with diffusion imaging before cardiac surgery, typically in the first week of life, and again after surgery before hospital discharge. Eighty-five infants, or 88 percent, completed a preoperative scan and 85 completed a postoperative scan. From the diffusion data, the researchers reconstructed each infant’s structural connectome, the network of white matter pathways linking distributed brain regions. Brain regions served as network nodes, white matter pathways as edges, and the strength of each connection was weighted by fractional anisotropy. Of 170 scans from the 97 eligible infants, 41 were removed after manual review because of severe motion corruption, poor tractography quality, or inadequate atlas registration. Ultimately, 80 infants produced at least one connectome of sufficient quality, contributing 62 preoperative and 61 postoperative scans, and 43 infants contributed data from both timepoints.

From these connectomes, the team calculated graph-theoretic measures of network organization. Global efficiency, which quantifies how efficiently information can travel across the network via shortest paths between regions, served as the primary measure of network integration. Transitivity, reflecting local clustering among connected regions, and modularity, reflecting how clearly regions organize into specialized communities, captured complementary aspects of network segregation. Because the number of detected connections can influence these metrics, the researchers also tested network density and found it did not significantly change from the preoperative to postoperative scan, and that any change did not differ by brain injury status or cardiac diagnosis, reducing concern that density differences drove the results.

Structural connectivity showed clear age-related change across the perioperative period. In baseline multivariate linear mixed-effects models adjusted for gestational age, sex, and maternal education, global efficiency showed the strongest association with postmenstrual age, with a standardized beta of 0.62 and a marginal R-squared of 0.43. Among infants with paired scans, 88 percent showed an increase in global efficiency between timepoints. Transitivity also increased with age, though with a smaller standardized association of 0.42, while modularity showed only trend-level evidence of age-related increase. The median interval between preoperative and postoperative MRI was 20 days, meaning the estimated slopes reflect the rate of network integration change during roughly the first three weeks of life.

Contrary to the researchers’ original hypothesis, cardiac physiology did not significantly affect the rate of structural connectome development. Neonates with single ventricle physiology showed no slower perioperative increases in global efficiency, transitivity, or modularity compared with infants with d-transposition of the great arteries. The authors note that because the imaging window was limited to the neonatal perioperative period, cumulative effects of prolonged altered postnatal hemodynamics in single ventricle infants may not yet be detectable in structural connectomes during the first weeks of life, and later physiology-specific effects on network maturation cannot be excluded.

Brain injury told a different story. A high rate of injuries was observed both preoperatively, affecting 44.7 percent of scans, and postoperatively, affecting 35.3 percent. White matter injury, the most common form of acquired brain injury in neonates with congenital heart disease, was present in 30.6 percent of preoperative scans, and new white matter injury appeared on 24.7 percent of postoperative scans. In multivariate models, white matter injury was associated with a slower rate of increase in global efficiency and in modularity, but not transitivity. Any brain injury, including white matter injury, stroke, intraventricular hemorrhage, and hypoxic-ischemic injury, was similarly associated with slower increases in global efficiency and modularity. Stroke, which was relatively uncommon, was not significantly associated with the rate of change in any network metric, a finding the authors caution should be interpreted cautiously given the limited number of stroke cases, which may have reduced the ability to detect stroke-specific differences.

The central question was whether these early network trajectories relate to later development. To answer it, the researchers used a two-stage modeling approach. First, they estimated subject-specific trajectories of global efficiency using a Bayesian linear mixed-effects framework, which allowed infants with only one scan to be partially informed by the population-level trajectory. Second, they tested whether each infant’s standardized global efficiency slope predicted Bayley Scales of Infant and Toddler Development domain scores, analyzed as longitudinal outcomes with repeated measurements between 1 and 4 years of age. Fifty of the 61 imaging participants eligible for follow-up, or 82 percent, completed at least one Bayley assessment, and 80 percent were tested more than once. Models accounted for cumulative brain injury, maternal education, the Bayley test edition, and other covariates identified through univariate screening, and p-values were corrected for multiple comparisons across the three domains using Bonferroni correction.

The results were striking. For each one-standard-deviation increase in the predicted trajectory of global efficiency, cognitive scores rose by 8.2 points, language scores by 9.1 points, and motor scores by 5.7 points, and all three associations remained significant after correction. In post hoc sensitivity analyses, neither preoperative global efficiency alone nor postoperative global efficiency alone was significantly associated with any of the three domain scores, supporting the authors’ argument that the developmental trajectory captures information not reflected in a single imaging timepoint. The authors note that the roughly 8-point cognitive difference per standard deviation approaches the magnitude of differences associated with major perinatal risk factors such as prematurity or socioeconomic disadvantage, though they emphasize that the study was not designed to develop or validate an individual-level clinical prediction model, and the reported R-squared values reflect model fit within this cohort rather than validated prognostic accuracy.

The authors and independent observers alike should note the study’s limitations. Data came from a single center, not all infants underwent MRI at both timepoints, and missing imaging may not have been completely random, since scheduling constraints, early discharge, or clinical instability could have influenced which infants completed both scans. Subject-specific slope estimates derived from a single timepoint may be less precise and partially pooled toward the population mean. Bayley assessments during early childhood may not capture later-emerging cognitive and executive difficulties common in children with congenital heart disease, and as an observational cohort study, residual confounding from unmeasured differences in clinical severity, perioperative management, or socioeconomic context cannot be ruled out. Nevertheless, the authors argue that these findings identify early network development as a potential biomarker of neurodevelopmental risk and resilience, and that future multicenter studies with formal prediction-model development and external validation will be needed to determine whether connectome measures can inform individualized risk stratification and guide early developmental surveillance as neonatal MRI becomes increasingly incorporated into clinical care for infants with congenital heart disease.

Subject of Research: Longitudinal neonatal structural connectome development and neurodevelopmental outcomes in infants with critical congenital heart disease

Article Title: Neonatal brain network integration trajectories predict neurodevelopment in congenital heart disease

Article References: Harasymiw, L., Kuang, A., Xu, D., Scheffler, A., George, E., Peyvandi, S., & McQuillen, P. (2026). Neonatal brain network integration trajectories predict neurodevelopment in congenital heart disease. Pediatric Research. https://doi.org/10.1038/s41390-026-05546-8

Image Credits: AI Generated

DOI: 10.1038/s41390-026-05546-8

Keywords: congenital heart disease, structural connectome, global efficiency, diffusion MRI, white matter injury, Bayley Scales, neonatal brain development, neurodevelopment, graph theory, biomarker, cardiac surgery, pediatric research