Early‐Life Clinical Factors Shape Cerebellar Structure in Preterm‐Born Children

ABSTRACT

Objective

Preterm birth alters typical brain development due to complex exposures related to neonatal illness, pain management, and brain injury. Despite an increased understanding of neonatal brain injury on developmental outcomes, the association of cerebellar maturation with school‐age motor and cognitive function remains incompletely understood. The current study evaluated neonatal clinical factors associated with regionally specific cerebellar volumes and white matter pathways at 8 years of age in children born preterm (24–32 weeks of gestational age).

Methods

Neuroimaging, clinical, and behavioral data were collected from 130 children born very preterm (24–32 weeks' gestational age) at BC Women's Hospital. Cerebellar subregions were automatically segmented using CerebNet, and cerebellar white matter connectivity was assessed using tractography. The relationships between cerebellar subregion volumes and neonatal clinical factors were examined using constrained principal component analysis. Cognitive and motor neurodevelopmental outcomes related to cerebellar volume and connectivity were also investigated.

Results

Neonatal clinical factors including mechanical ventilation, number of invasive procedures, and opioid exposures were associated with reduced cerebellar volumes in specific subregions. Acute neonatal mortality risk (i.e., SNAPPE‐II) was negatively associated with cerebellar connectivity at 8 years. Reduced cerebellar volumes and structural connectivity were related to poorer cognitive and motor outcomes.

Interpretation

In very preterm children, neonatal illness severity and exposure to invasive procedures and opioids are associated with reduced cerebellar volumes, connectivity, and poorer neurodevelopmental outcomes at 8 years.

1. Introduction

Preterm birth remains the leading cause of lifelong neurodevelopmental disabilities worldwide [1]. Extremely (< 28 weeks' gestation) to very preterm infants (28–32 weeks' gestation) are born during a critical period of rapid brain development, vulnerable to white matter and cerebellar injury [2, 3, 4]. Moreover, during their neonatal intensive care unit (NICU) stay, preterm infants are frequently exposed to procedures inducing pain‐related stress (e.g., invasive skin‐breaking procedures) and treatment for neonatal morbidities, altering typical brain development [5]. Early life exposures, such as infections, neonatal pain, and analgesia and sedative medications are associated with regional changes in brain microstructure [6, 7, 8, 9]. As a result, preterm birth is often associated with abnormal brain maturation and lifelong neurodevelopmental challenges, highlighting a need for longitudinal follow‐up studies [10].

The cerebellum undergoes its most rapid period of development during the third trimester of pregnancy [11]. The co‐occurrence of this rapid cerebellar growth and foliation period and their prolonged NICU stay makes preterm infants particularly vulnerable to cerebellar injury and dysmaturation [2, 12, 13]. Smaller cerebellar volumes and altered development have been associated with supratentorial brain injury, higher procedural pain and infection, and morphine exposure in children born preterm [7, 14, 15]. Consequently, cerebellar dysmaturation and injury contribute, in part, to long‐term cognitive and behavioral dysfunctions that occur in children born preterm [13]. The extended period of postnatal cerebellar development makes post‐neonatal brain imaging critical to identifying the risk factors for impaired cerebellar maturation.

The cerebellum has long been recognized for its involvement in motor control and coordination. In addition, its significance in many cognitive aspects, including executive function and language, has become increasingly evident [16]. Given the complexity of the cerebellum, understanding specific roles of prematurity‐related factors and brain injury on cerebellar development is crucial to optimize health‐related outcomes throughout childhood. The current study evaluated neonatal clinical factors associated with regionally specific cerebellar volumes and white matter pathways at 8 years of age in children born preterm (24–32 weeks' gestational age; GA). Then, we assessed the association between cerebellar development and neurodevelopmental outcomes at 8 years of age.

2. Participants/Materials and Methods

The study was approved by the University of British Columbia and BC Women's Hospital Clinical Research Ethics Board. Parental informed consent was obtained at study enrollment and at each subsequent study visit.

2.1. Study Population

Participants were part of a larger prospective cohort study of children born very preterm (24–32 weeks GA), admitted to the level III NICU at BC Women's Hospital between 2006 and 2013. Exclusion criteria included major congenital malformations or syndrome, antenatal congenital infection, or large parenchymal hemorrhagic infarction (> 2 cm as evidenced by ultrasound) [8]. Of the 234, 130 participants completed three magnetic resonance imaging (MRI) scans (early‐life, term‐equivalent, and 8‐year follow‐up) and standardized neurodevelopmental assessments at 8 years and were included in the study (Figure 1). Comparisons between the original cohort and those followed at 8 years have been previously described [17], with no significant differences in neonatal clinical factors observed between groups.

Figure 1.

Cohort flow chart. DTI, diffusion tensor imaging.

2.2. Clinical Data Collection

Participant demographics (GA, sex, age at follow‐up) and parental information including maternal level of education, based on self‐report of the highest level of degree conferred, were reported. Systematic medical record reviews were conducted by neonatal research nurses to obtain clinical data from birth to NICU discharge or term‐equivalent age (Table 1). Neonatal clinical variables obtained included the Score for Neonatal Acute Physiology with Perinatal Extension, second edition (SNAPPE‐II), the number of invasive procedures, and average drug exposure to fentanyl, dexamethasone, hydrocortisone, morphine, and midazolam. Categorical data including whether the preterm neonate had hypotension and whether they required mechanical ventilation were also collected. Maternal education at the time of NICU admission was used as our primary measure of socioeconomic status (SES), as reported previously [18, 19].

Table 1.

Demographic and neonatal clinical characteristics.

N = 130
Prenatal clinical features
Male	69 (53.1%)
Maternal education
Primary/secondary school	17 (13.1%)
Undergraduate degree	84 (64.6%)
Postgraduate degree	27 (20.8%)
Postnatal clinical features
Gestational age, wks	28.4 [25.9–30.1]
Small for gestational age	23 (17.7%)
Apgar score at 5 min	9 [7–10]
SNAPPE‐II score at 24 h	14 [0–38.8]
No. WMI	46 (35.4%)
WMI, mm3	72.7 [33.9–181.5]
No. cerebellar hemorrhage	22 (16.9%)
Cerebellar hemorrhage, mm3	47.7 [27.1–101.5]
IVH
Grade 1	9 (6.9%)
Grade 2	86 (66.2%)
Grade 3	0 (0.0%)
Grade 4	4 (3.1%)
No. of invasive procedures	96 [55–162]
Fentanyl, mg/kg/day (n = 32)	0.01 [0.009–0.01]
Dexamethasone, mg/kg/day (n = 24)	0.1 [0.08–0.15]
Hydrocortisone, mg/kg/day (n = 27)	1.1 [0.95–1.66]
Morphine, mg/kg/day (n = 73)	0.3 [0.1–0.5]
Midazolam, mg/kg/day (n = 25)	1.3 [0.6–2.1]
8‐year follow‐up
Age, yrs	8.2 [8.1–8.6]
Total cerebral volume, cm3	1059.3 [973.2–1144.6]
Total cerebellar volume, cm3	120.9 [110.7–130.5]
Cerebellar FA	0.52 [0.50–0.54]
WASI2‐FSIQ, std. score	103 [92–115]
BRIEF GEC, std. score	49 [42–58]
Beery VMI, percentile	38 [21–55]
MABC‐2, percentile	25 [5–63]

Note: Data are presented as the median [IQR] or number (%).

Abbreviations: Beery VMI, Beery–Buktenica Developmental Test of Visual‐Motor Integration, 6th edition; BRIEF GEC, Behavior Rating of Executive Function Global Executive Composite score; IVH, intraventricular hemorrhage; MABC‐2, Movement Assessment Battery for Children, 2nd edition; SNAPPE‐II, Score for Neonatal Acute Physiology with Perinatal Extension, 2nd edition; WASI2‐FSIQ, Wechsler Abbreviated Scale of Intelligence, 2nd edition FullScale IQ; WMI, white matter injury.

2.3. Magnetic Resonance Imaging

Magnetic resonance (MR) scans were conducted using a Siemens (Berlin, Germany) 1.5T Avanto scanner with parameters previously reported [20] twice during the neonatal period: at early life when clinically stable and at term‐equivalent age (TEA). A neuroradiologist reviewed and scored the neonatal MR scans for intraventricular hemorrhage (IVH), white matter injury (WMI), and cerebellar hemorrhage severity. WMI and cerebellar hemorrhage volumes were calculated from manual segmentations on the neonatal scans (i.e., early‐life and TEA) [21]. IVH grade, maximal WMI volume, and maximum cerebellar hemorrhage volume were used in our analysis.

At 8 years of age, structural and diffusion tensor imaging (DTI) MR scans were acquired using a 3T General Electric Discovery Medical Systems Scanner (MR750) with parameters previously reported [17]. Structural images included a 3D coronal volumetric T₁‐weighted image and an axial fast spin echo T₂‐weighted image to assess structural growth. To examine white matter maturation, a single‐shot‐echo‐planar DTI sequence was acquired (TE/TR = minimum 59–69 ms/7000 ms, FOV = 220 mm × 220 mm, voxel size = 2.2 × 2.2 × 2.2 mm) with 32 diffusion‐encoding directions (b = 1000 s/mm³) and three non‐diffusion weighted images. To optimize image quality, between one and three repetitions of the DTI sequence were obtained, improving the signal‐to‐noise ratio and reducing the likelihood of motion‐related artifacts and consequently, the need for participant exclusion.

2.3.1. Volumetric Analysis

Total cerebellar volumes and total cerebral volumes were segmented semi‐automatically using MAGeT‐Brain [22], reviewed, and manually revised to ensure accurate delineation of the structures. Automated segmentations of cerebellar subregions were determined using a deep‐learning pipeline, CerebNet [23]. Briefly, CerebNet leverages the FastSurferCNN [24] approach of multiple 2D neural networks and minimal pre‐processing of T1‐weighted images to obtain boundary segmentations of the cerebellum. CerebNet segments 27 cerebellar regions. Final segmentations were checked visually prior to analysis, and all images were of good quality (Figure 2).

Figure 2.

Cerebellar subregions and DTI tracts. Top: Sample CerebNet [23] segmentation of the cerebellum and subregions of a child born preterm. Representative slices are taken along the (A) sagittal and (B) coronal planes. Bottom: Sample cerebellar white matter tracts fractional anisotropy of the middle cerebellar peduncle. Representative slices are taken along the (C) sagittal and (D) coronal planes.

2.3.2. DTI Analysis

DTI series were visually inspected for artifacts and overall scan quality. Sequences with excessive motion, defined as a mean framewise displacement greater than half the voxel width (1.1 mm), or with fewer than 35 usable volumes, were excluded. For participants with multiple DTI sequences meeting quality criteria, the sequences were concatenated and processed as a single series. Axial DTI slices were resampled at 0.86 × 0.86 mm resolution using zero‐interpolation filling, and brain masks were generated using FSL BET (FMRIB Software Library, v. 6.0.6). Data were corrected for head motion, susceptibility distortion, and eddy currents using FSL eddy. Eddy Quad [25, 26] was used to quantify eddy quality control metrics, and absolute and relative motion were compared between participants with concatenated runs (n = 96) to those with a single run (n = 34). Median absolute motion (concatenated = 0.92; single = 0.82; U(34,96) = 1304.5, p = 0.083) and median relative motion (concatenated = 0.41; single = 0.52; U(34,96) = 1294, p = 0.073) did not differ between groups.

DTI metrics, including fractional anisotropy (FA), were calculated using FSL dtifit (FMRIB Software Library, v. 6.0.6) [26, 27, 28, 29] with a weighted least squares regression. The ICBM 152 nonlinear symmetric 2009c template and CerebrA labels [30] were nonlinearly registered to both a study‐specific group FA template and to individual DTI images using Advanced Normalization Tools (ANTs, version 2.4.3) [31], enabling region‐based connectivity analyses. Tissue masks for white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF) were generated by warping the ICBM atlas to each participant's DTI space.

Deterministic tractography was performed using Diffusion Imaging in Python (DiPy, v. 1.11.0) [32]. Streamlines were seeded from voxels within the WM mask, with 16 seeds per voxel. Streamlines were propagated using the EuDX algorithm. Streamlines were excluded if they entered the CSF mask or if they did not terminate in the GM. WM tracts connecting each pair of regions defined in the CerebrA atlas [30] were segmented by identifying all streamlines intersecting both regions and cutting streamlines at the region boundaries. Structural connectivity between each pair of regions was quantified as the streamline‐weighted mean FA within the segmented tract. FA values were extracted bilaterally for the superior cerebellar peduncles (SCP) and inferior cerebellar peduncles (ICP). The middle cerebellar peduncle (MCP), being the midline structure connecting the left and right cerebellar hemispheres, was reported as a single value. These cerebellar WM tracts were quality assessed using TrackVis (http://www.trackvis.org; Figure 2).

2.4. Neurodevelopmental Outcomes

At the 8‐year follow‐up, participants completed a series of standardized neurodevelopmental assessments to quantify cognition, working memory, visual‐motor integration, and motor skills. Standardized assessments were administered by a psychologist and an occupational therapist at BC Children's and Women's Hospital. Staff were blinded to the children's imaging findings.

Standardized assessments included: (1) the Full‐Scale IQ from the Wechsler Abbreviated Scale of Intelligence, second edition (WASI2‐FSIQ); (2) Global Executive Composite Score from the Behavior Rating of Executive Function (BRIEF GEC); (3) the Beery‐Buktenica Developmental Test of Visual‐Motor Integration, sixth edition (Beery VMI); and (4) the Movement Assessment Battery for Children, second edition (MABC‐2). Apart from the MABC‐2 (mean score 10 and a standard deviation of 3), standardized tests had a mean score of 100 and a standard deviation of 15. Higher scores on the FSIQ, Beery VMI, and MABC‐2 correspond to more favorable outcomes, whereas lower scores on the BRIEF GEC reflect better functioning.

2.5. Statistical Analyses

To identify which neonatal clinical variables (GA, SNAPPE‐II, number of invasive procedures, hypotension, mechanical ventilation, exposure to fentanyl, dexamethasone, hydrocortisone, morphine, and midazolam, cerebellar hemorrhage volumes, moderate‐to‐severe WMI) were related to 8‐year cerebellar volumes and structural connectivity, we performed a constrained principal component analysis (CPCA). CPCA involves an external and internal analysis of the data. The external analysis uses a multivariate least squares multiple regression to model the dependent measures (i.e., cerebellar subregions and tract FA values) based on the independent variables (i.e., neonatal clinical factors). This step produces a matrix of predicted scores, representing the cerebellar variance explained by the neonatal clinical variables, and a residual matrix, reflecting unexplained variance. Then, the internal analysis applies a principal component analysis (PCA) on the aforementioned matrices. The resulting component structures (overall, predicted, and residual) were examined to determine which cerebellar subregional volumes and structural connectivity could be explained by the neonatal clinical variables. The number of components retained was determined by inspection of scree plots. All PCA solutions were separately rotated using varimax with Kaiser normalization. Solutions were bootstrapped 1000 times using a Monte Carlo method to produce confidence intervals and p‐values [33, 34]. CPCA computations were completed using custom scripts adapted from Takane and colleagues [33, 34] using MATLAB (MathWorks, 2024). Multiple comparisons were corrected using a 5% false discovery rate.

Generalized linear models were fit for each neurodevelopmental assessment outcome (FSIQ, GEC, Beery VMI, and MABC‐2), using CPCA component scores, sex, and maternal education as predictors, while adjusting for age at 8 years.

3. Results

Of the 164 children who returned for 8‐year follow‐up assessments, 130 completed the MRI. Demographic and neonatal clinical data are presented in Table 1.

3.1. CPCA

The external analysis of the CPCA revealed that the 14 neonatal clinical factors included accounted for 33.9% of the overall variance of the cerebellar subregional volumes and structural connectivity (i.e., FA values). Three main components (component 1, 2, 3) were extracted from the predicted solutions, which corresponded to 16.4%, 9.2%, and 4.1% of the overall variance, respectively.

Component loadings and their bootstrapped confidence intervals and p‐values for each cerebellar subregion and white matter tract are summarized in Table 2. The three components uniquely and positively loaded onto cerebellar subregions and tracts (i.e., larger volumes and greater FA). The dominant loadings on the first component were distributed bilaterally, primarily loading onto the posterior cerebellar regions including the lateral hemispheres of lobules VI‐IX, crus I, crus II, and vermis VI‐IX. Moderate contributions from lobules I‐V were also accounted for in component 1. Component 2 primarily loaded onto the anterior cerebellar subregions, including lobules I‐V and cerebellar WM. Moderate contributions included lobule VI, vermis, and inferior regions of the cerebellum onto component 2. Component 3 was comprised of the WM tracts of the cerebellum including the MCP and the right and left ICP. Additional moderate loadings from the inferior‐posterior cerebellar regions (lobules VIIb, VIIIa‐VIIIb) were also accounted for in component 3.

Table 2.

CPCA component loadings on cerebellar subregional volumes and FA in white matter tracts.

	Component 1	p‐value	Component 2	p‐value	Component 3	p‐value
White matter tracts FA
MCP	0.08 (−0.15, 0.31)	0.51	0.10 (0.28, 0.28)	0.28	0.45 (0.26, 0.64)	< 0.001
Right ICP	0.02 (−0.21, 0.25)	0.86	0.05 (0.24, 0.24)	0.60	0.33 (0.11, 0.56)	< 0.001
Left ICP	0.12 (−0.11, 0.34)	0.31	0.00 (0.20, 0.18)	0.99	0.43 (0.25, 0.61)	< 0.001
Right SCP	0.20 (−0.07, 0.47)	0.14	−0.01 (0.25, 0.21)	0.96	0.07 (−0.20, 0.34)	0.63
Left SCP	−0.01 (−0.20, 0.18)	0.94	0.12 (0.52, 0.51)	0.56	−0.07 (−0.29, 0.15)	0.53
Cerebellar volumes
Left cerebellar WM	0.46 (0.26, 0.67)	< 0.001	0.38 (0.52, 0.54)	< 0.001	0.22 (0.05, 0.39)	0.01
Right cerebellar WM	0.63 (0.48, 0.78)	< 0.001	0.39 (0.52, 0.55)	< 0.001	0.19 (0.08, 0.30)	< 0.001
Left I–IV	0.35 (0.22, 0.48)	< 0.001	0.49 (0.67, 0.68)	< 0.001	0.10 (−0.03, 0.24)	0.13
Right I–IV	0.37 (0.22, 0.52)	< 0.001	0.51 (0.70, 0.69)	< 0.001	0.07 (−0.09, 0.23)	0.39
Left V	0.39 (0.22, 0.56)	< 0.001	0.43 (0.63, 0.63)	< 0.001	0.17 (0.02, 0.32)	0.03
Right V	0.40 (0.22, 0.58)	< 0.001	0.37 (0.57, 0.57)	< 0.001	0.14 (−0.01, 0.29)	0.06
Left VI	0.41 (0.18, 0.64)	< 0.001	0.35 (0.52, 0.53)	< 0.001	0.15 (0.00, 0.30)	0.05
Vermis VI	0.37 (0.20, 0.54)	< 0.001	0.35 (0.53, 0.51)	< 0.001	0.28 (0.11, 0.45)	< 0.001
Right VI	0.63 (0.41, 0.85)	< 0.001	0.27 (0.46, 0.50)	< 0.001	0.13 (−0.03, 0.29)	0.11
Left crus I	0.37 (0.12, 0.63)	< 0.001	0.27 (0.47, 0.50)	0.01	0.06 (−0.14, 0.26)	0.57
Right crus I	0.56 (0.37, 0.74)	< 0.001	0.26 (0.46, 0.49)	0.01	0.05 (−0.12, 0.21)	0.59
Left crus II	0.53 (0.38, 0.69)	< 0.001	0.20 (0.35, 0.39)	0.01	0.08 (−0.06, 0.22)	0.28
Right crus II	0.58 (0.45, 0.71)	< 0.001	0.23 (0.36, 0.41)	< 0.001	0.03 (−0.09, 0.16)	0.62
Left VIIb	0.56 (0.37, 0.74)	< 0.001	0.18 (0.33, 0.37)	0.02	0.26 (0.11, 0.40)	< 0.001
Right VIIb	0.56 (0.38, 0.74)	< 0.001	0.18 (0.33, 0.37)	0.02	0.09 (−0.06, 0.25)	0.24
Left VIIIa	0.36 (0.11, 0.60)	< 0.001	0.17 (0.35, 0.38)	0.06	0.31 (0.11, 0.52)	< 0.001
Right VIIIa	0.52 (0.36, 0.67)	< 0.001	0.13 (0.30, 0.34)	0.15	0.27 (0.13, 0.42)	<0.001
Left VIIIb	0.44 (0.19, 0.69)	< 0.001	0.14 (0.35, 0.35)	0.17	0.23 (0.02, 0.43)	0.03
Right VIIIb	0.50 (0.35, 0.66)	< 0.001	0.20 (0.40, 0.42)	0.04	0.28 (0.12, 0.45)	< 0.001
Left IX	0.46 (0.23, 0.68)	< 0.001	0.18 (0.39, 0.43)	0.08	0.07 (−0.17, 0.30)	0.58
Vermis IX	0.34 (0.15, 0.53)	< 0.001	0.32 (0.54, 0.60)	< 0.001	0.06 (−0.12, 0.24)	0.51
Right IX	0.59 (0.41, 0.76)	< 0.001	0.19 (0.40, 0.43)	0.06	0.01 (−0.18, 0.19)	0.94
Left X	0.15 (−0.05, 0.34)	0.13	0.42 (0.61, 0.63)	< 0.001	0.23 (0.01, 0.45)	0.04
Vermis X	0.33 (0.17, 0.50)	< 0.001	0.35 (0.56, 0.58)	< 0.001	−0.02 (−0.19, 0.14)	0.77
Right X	0.14 (−0.05, 0.33)	0.16	0.41 (0.61, 0.62)	< 0.001	0.22 (0.02, 0.42)	0.03
Vermis VII	0.27 (0.06, 0.49)	0.01	0.33 (0.59, 0.54)	0.01	0.20 (−0.03, 0.42)	0.08
Vermis VIII	0.39 (0.20, 0.58)	< 0.001	0.29 (0.42, 0.44)	< 0.001	0.13 (−0.01, 0.26)	0.07

Table 3 summarizes the neonatal clinical factors that explain the variance in subregional volumes and WM integrity of the cerebellum. Predictor variables with positive loadings were associated with larger cerebellar volumes and greater FA values, while predictor variables with negative loadings were associated with smaller cerebellar volumes and lower FA values. Total brain volume positively loaded on all three components, reflecting that larger brain volume was associated with larger volumes of the cerebellum and greater WM integrity (i.e., larger FA). Gestational age was associated with larger cerebellar volumes (component 1 and 2). Component 1 reflected neonatal injury, with cerebellar hemorrhage and IVH relating to smaller posterior cerebellar volumes. Component 2 was associated with reduced cerebral blood flow, with mechanical ventilation intervention and hypotension relating to smaller anterior cerebellar volumes. Number of invasive procedures (i.e., skin‐breaking procedures) and average drug exposure, including dexamethasone, hydrocortisone, morphine, and midazolam, reflected smaller cerebellar volumes (component 1 and 2). Mechanical ventilation loaded negatively on component 1 and component 2 but positively on component 3, indicating mechanical ventilation was not associated with cerebellar subregions that loaded on both components. Thus, infants requiring mechanical ventilation in the NICU were uniquely associated with smaller volumes of the lobules I‐V and lobules VI‐IX, crus I, and crus II. These subregions loaded on component 1 and 2, explaining a larger portion of the total variance in cerebellar volumes. Severity of illness in the first 24 h of admission determined using the SNAPPE‐II was associated with smaller anterior cerebellar volumes (component 2) and reduced MCP and ICP FA values (component 3; Table S1). Component 3 also reflected supratentorial WMI relating to reduced cerebellar FA values (i.e., MCP and ICP). The correlations between prematurity‐related factors and the cerebellar subregional volumes and cerebellar structural connectivity are shown in Figure 3.

Table 3.

CPCA loadings and neonatal clinical factors explaining variance in subregional volumes and FA in white matter tracts.

	Component 1			Component 2			Component 3
	Loading	95% CI	p‐value	Loading	95% CI	p‐value	Loading	95% CI	p‐value
Total cerebral volume (mm3)	0.46	(0.29, 0.63)	< 0.001	0.40	(0.22, 0.58)	< 0.001	0.54	(0.37, 0.71)	< 0.001
GA (wk)	0.41	(0.23, 0.58)	< 0.001	0.55	(0.38, 0.72)	< 0.001	0.05	(−0.12, 0.22)	0.58
WMI (mm3)	0.08	(−0.09, 0.26)	0.44	0.02	(−0.14, 0.19)	0.77	−0.22	(−0.40, −0.05)	0.01
Cerebellar hemorrhage (mm3)	−0.81	(−0.99, −0.64)	< 0.001	0.14	(−0.04, 0.31)	0.18	−0.05	(−0.23, 0.12)	0.58
IVH	−0.39	(−0.57, −0.22)	< 0.001	−0.06	(−0.23, 0.12)	0.58	0.20	(0.03, 0.38)	0.03
SNAPPE‐II	−0.08	(−0.24, 0.09)	0.45	−0.39	(−0.56, −0.22)	< 0.001	−0.48	(−0.66, −0.30)	< 0.001
Mechanical ventilation	−0.49	(−0.66, −0.32)	< 0.001	−0.61	(−0.78, −0.44)	< 0.001	0.34	(0.17, 0.50)	< 0.001
Hypotension	−0.26	(−0.44, −0.09)	< 0.001	−0.62	(−0.80, −0.45)	< 0.001	0.03	(−0.15, 0.20)	0.77
No. invasive procedures	−0.34	(−0.51, −0.17)	< 0.001	−0.71	(−0.88, −0.54)	< 0.001	0.11	(−0.06, 0.27)	0.28
Fentanyl (mg/kg/day)	−0.06	(−0.22, 0.11)	0.58	−0.10	(−0.27, 0.07)	0.30	0.09	(−0.08, 0.27)	0.37
Dexamethasone (mg/kg/day)	−0.19	(−0.36, −0.02)	0.05	−0.55	(−0.72, −0.38)	< 0.001	0.15	(−0.01, 0.32)	0.11
Hydrocortisone (mg/kg/day)	−0.44	(−0.61, −0.28)	< 0.001	−0.25	(−0.42, −0.07)	0.02	0.06	(−0.12, 0.24)	0.58
Morphine (mg/kg/day)	−0.59	(−0.76, −0.41)	< 0.001	−0.29	(−0.46, −0.11)	< 0.001	0.14	(−0.04, 0.31)	0.17
Midazolam (mg/kg/day)	−0.47	(−0.64, −0.30)	< 0.001	−0.44	(−0.61, −0.27)	< 0.001	0.16	(−0.02, 0.33)	0.12

Note: Significant p‐values are in bold.

IV = intraventricular hemorrhage (0 = none, 1 = grade 1, 2 = grade 2, 3 = grade 3, 4 = periventricular hemorrhage).

Figure 3.

Cerebellar subregions, neonatal factors, and outcomes. Representation of cerebellar subregions with their associated components, associated neonatal factors (left), and neurodevelopmental outcomes assessed at 8 years (right) in preterm‐born children. Beery VMI, Beery–Buktenica Developmental Test of Visual‐Motor Integration, 6th edition; BRIEF GEC, Behavior Rating of Executive Function Global Executive Composite; Drug Exposure, includes dexamethasone, hydrocortisone, morphine, and midazolam; IVH, intraventricular hemorrhage; MABC‐2, Movement Assessment Battery for Children, 2nd edition; SNAPPE‐II, Score for Neonatal Acute Physiology with Perinatal Extension, 2nd edition; WASI2‐FSIQ, Wechsler Abbreviated Scale of Intelligence, 2nd edition Full‐Scale IQ; WMI, white matter injury.

Given that sex differences in cerebellar volumes have been previously reported [7, 35], independent t‐tests comparing cerebellar CPCA component loadings in males and females were conducted. Adjusting for total cerebral volumes, females had smaller anterior cerebellar subregions (component 2) compared with males (t(128) = 2.96, p = 0.004).

3.2. Neurodevelopmental Outcomes

Table 4 summarizes the associations between the three component loadings, sex, and maternal education on each of the 8‐year neurodevelopmental standardized assessments. Component 1 was associated with a decrease in the BRIEF GEC score, an increase in the Beery VMI score, and an increase in the MABC‐2 score at 8 years. Component 2 was associated with an increase in the WASI2‐FSIQ score at 8 years. Component 3 was associated with an increase in the WASI2‐FSIQ score and an increase in the MABC‐2 score at 8 years. There were no main effects of sex on the neurodevelopmental assessments. Preterm‐born children whose mothers had a postgraduate degree were associated with an increase in WASI2‐FSIQ scores compared with preterm‐born children whose mothers had an undergraduate degree. The correlations of these outcome measures and the cerebellar subregional volumes are presented in Figure 3.

Table 4.

Results from general linear models assessing the associations of the cerebellar CPCA component loadings, sex, and maternal education with 8‐year neurodevelopmental outcomes, adjusting for age at 8 years.

Neurodevelopmental assessment		Component 1		Component 2		Component 3
		B‐coefficient (95% CI)	p‐ value	B‐coefficient (95% CI)	p‐ value	B‐coefficient (95% CI)	p‐ value
WASI2‐FSIQ total score
Component loading		26.18 (−9.03, 61.40)	0.145	36.93 (3.02, 70.85)	0.033	56.22 (27.98, 84.46)	< 0.001
Male		−2.63 (−7.88, 2.63)	0.327	−3.91 (−9.46, 1.64)	0.167	−3.25 (−8.34, 1.84)	0.211
Maternal education	High school	3.54 (−3.84, 10.92)	0.347	2.92 (−4.67, 10.51)	0.451	3.80 (−3.36, 10.95)	0.298
	Undergraduate	[Ref]	—	[Ref]	—	[Ref]	—
	Graduate	5.63 (−0.329, 11.55)	0.062	6.68 (1.03, 12.34)	0.021	5.68 (0.17, 11.19)	0.04
BRIEF GEC total score
Component loading		−35.63 (−56.51, −14.75)	< 0.001	−10.73 (−33.75, 12.30)	0.361	−19.27 (−38.91, 0.36)	0.054
Male		−0.60 (−4.0. 2.80)	0.731	−0.77 (−4.52, 2.99)	0.689	−0.87 (−4.33, 2.60)	0.624
Maternal education	High school	3.41 (−2.50, 9.31)	0.258	3.25 (−2.18, 8.67)	0.241	3.02 (−2.27, 8.30)	0.263
	Undergraduate	[Ref]	—	[Ref]	—	[Ref]	—
	Graduate	−1.79 (−5.48, 1.91)	0.343	−2.24 (−6.47, 2.00)	0.300	−1.92 (−6.18, 2.35)	0.378
Beery VMI total score
Component loading		20.42 (2.37, 38.48)	0.027	6.36 (−16.04, 28.75)	0.578	15.66 (−6.08, 37.40)	0.158
Male		0.28 (−3.49, 4.04)	0.886	0.38 (−3.43, 4.19)	0.845	0.34 (−3.53, 4.21)	0.863
Maternal education	High school	0.52 (−4.14, 5.18)	0.826	0.55 (−4.19, 5.28)	0.821	0.66 (−4.09, 5.41)	0.785
	Undergraduate	[Ref]	—	[Ref]	—	[Ref]	—
	Graduate	2.56 (−2.04, 7.15)	0.276	2.85 (−1.87, 7.57)	0.237	2.64 (−2.07, 7.35)	0.272
MABC‐2 Total Score
Component loading		7.85 (1.93, 13.77)	0.009	0.45 (−7.66, 8.57)	0.913	10.78 (4.66, 16.90)	< 0.001
Male		−0.19 (−1.45, 1.08)	0.773	−0.09 (−1.43, 1.25)	0.896	−0.26 (−1.50, 0.98)	0.676
Maternal education	High school	−0.90 (−2.65, 0.85)	0.316	−0.84 (−2.61, 0.94)	0.356	−0.82 (−2.38, 0.73)	0.300
	Undergraduate	[Ref]	—	[Ref]	—	[Ref]	—
	Graduate	1.17 (−0.37, 2.71)	0.137	1.20 (−0.36, 2.75)	0.132	1.18 (−0.45, 2.81)	0.155

Note: Bold values indicate p‐value < 0.05.

Abbreviations: Beery VMI, Beery–Buktenica Developmental Test of Visual‐Motor Integration, 6th edition; BRIEF GEC, Behavior Rating of Executive Function, Global Executive Composite; MABC‐2, Movement Assessment Battery for Children, 2nd edition; WASI2‐FSIQ, Wechsler Abbreviated Scale of Intelligence, 2nd edition Full‐Scale IQ.

4. Discussion

In this prospective cohort of 8‐year‐old children born preterm, we used a data‐driven approach to understand the long‐term relationship of preterm birth with the cerebellum and subsequent neurodevelopmental outcomes. We recognize that the cerebellum is only a small part of various brain networks that contribute to motor function and cognition. Therein, we present an overview of the impact of early‐life exposures on cerebellar development. Using a CPCA, one‐third of the overall variance in cerebellar volume and structural connectivity at 8 years differentially related to early‐life exposures associated with preterm birth. Reduced cerebellar volumes at 8 years were primarily influenced by neonatal clinical factors occurring within the NICU, including mechanical ventilation, hypotension, greater number of invasive procedures, and drug exposure, while variability in structural connectivity of the cerebellum was primarily associated with WMI and early‐life predictors of illness and mortality. Consequently, altered cerebellar development due to early‐life exposures in preterm‐born children reflects below‐average cognitive, executive function, and visuo‐motor outcomes.

4.1. Neonatal Clinical Variables and Cerebellar Volumes

Total cerebral volume, GA, and male sex were positively related to cerebellar volumes in children born preterm. These findings align with previous work such that larger cerebellar volumes were associated with greater total cerebral volumes and increased GA in children born preterm [7]. As well, sex differences are consistently identified in neuroimaging studies, such that males, on average, have a greater brain size than females [36, 37]. Given the well‐established association between GA and brain size, these findings are unsurprising. Our results highlight that GA and total brain volume alone do not account for all the variability in cerebellar development. Instead, cerebellar development is also influenced by early‐life clinical exposure and brain injury.

Postnatal drug exposure, including dexamethasone, hydrocortisone, morphine, and midazolam, as well as greater number of invasive procedures, were associated with reduced cerebellar volumes at 8 years. This aligns with previous studies that reported postnatal exposure to glucocorticoids and number of invasive procedures were independent predictors of smaller cerebellar volumes [7, 14, 38]. Each of these drug exposures have been related to specific alterations in cerebellar development. Increased glucocorticoid exposure, including postnatal dexamethasone and hydrocortisone, increases apoptosis of the granule cells of the external granular layer, as evidenced by animal models [39]. Morphine exposure, from 24 weeks' gestation to term, has been suggested to impact Purkinje cell survival, resulting in smaller cerebellar growth, as evidenced in mice models [40]. Consequently, both granular and Purkinje cell vulnerability to postnatal interventions in preterm neonates may contribute to long‐term reductions in cerebellar size and function [38, 41]. Midazolam exposure has been shown to negatively impact hippocampal development, resulting in poorer outcomes in preterm neonates [42], while the number of invasive procedures has been associated with regionally specific alterations in the thalamus [43]. Given the integrative brain network linking the hippocampus, thalamus, and cerebellum, early disruptions in hippocampal or thalamic development may indirectly contribute to reduced cerebellar volumes. Collectively, these findings suggest that invasive procedures and exposure to various opioids and glucocorticoids influence cerebellar size through distinct pathways, and the timing (i.e., GA at exposure), duration, and dose of these medications likely plays a critical, yet understudied, role in shaping the molecular development of the cerebellum.

Neonatal injury, including cerebellar hemorrhage and IVH, was associated with reduced posterior cerebellar volumes at 8 years. Cerebellar hemorrhages are common complications of preterm birth, often coinciding with IVH, and result in disruptions in blood flow [13, 39, 44, 45]. Specific to the cerebellum, disruptions in blood flow of the posterior‐inferior cerebellar artery lead to smaller posterior‐inferior cerebellar volumes [46]. Postmortem examination of infants with cerebellar hemorrhage revealed neuronal loss and gliosis, particularly in the inferior olivary nuclei and dentate nucleus [47]. Thus, surviving infants with reduced neuronal loss and gliosis due to the cerebellar hemorrhage have long‐term implications for higher neurological functions. Neuronal loss and gliosis within the posterior cerebellar structures, such as the dentate nucleus, disrupts a major output pathway and may explain, in part, reduced cognitive and motor skills long‐term [46]. These findings emphasize that illness severity and hemodynamic instability are major determinants of cerebellar vulnerability; however, the current study did not localize cerebellar hemorrhage to a specific subregion, which may obscure potential differences in regional cerebellar vulnerability. Future studies incorporating region‐specific mapping of cerebellar hemorrhage could provide more detailed insight into how localized cerebellar injury contributes to long‐term neurodevelopmental outcomes in preterm‐born children.

Preterm‐born children requiring mechanical ventilation or diagnosed with hypotension, within the NICU, had smaller anterior cerebellar volumes in the present study. Previous studies have demonstrated that preterm infants requiring mechanical ventilation have reduced cortical thickness [9] and reduced hippocampal growth [42]. Our findings indicate that neonatal hypotension and mechanical ventilation intervention result in smaller cerebellar volumes in children born preterm. This suggests infants requiring mechanical ventilation or with reduced blood oxygen flow (i.e., hypotensive infants) may be unable to sufficiently supply the cerebellum with proper nutrients and oxygen, impacting long‐term development. Limperopoulos and colleagues [45] reported that more severely ill preterm infants have a higher frequency of cerebellar hemorrhage as well as an increased morbidity and mortality. Much like these findings, preterm‐born children who require neonatal intervention (e.g., opioid exposure, mechanical ventilation) due to fetal‐early neonatal illness have compromised cerebellar development.

4.2. WMI and SNAPPE‐II and Cerebellar Structural Connectivity

In preterm‐born children, greater neonatal WMI present on early‐life or term‐equivalent MRI scans was associated with reduced cerebellar structural connectivity at 8 years. Specifically, greater WMI in early life was associated with reduced FA of the middle and inferior cerebellar peduncles (i.e., MCP and ICP). The MCP and ICP are afferent pathways, relaying information between the cerebral cortices to the cerebellum (MCP) and between the spinal cord and brain stem to the cerebellum (ICP). Reduced FA of these cerebellar tracts indicates disrupted WM integrity, reflecting more isotropic water diffusion in tissue. Previous studies of very preterm infants with WMI have reported lower FA values in regions affected by the WMI, including the corticospinal tract [48], corpus callosum [49], and cingulum bundle [50]. Similar to those of Shany and colleagues [51], our results suggest that cerebral WMI may contribute to downstream microstructural alterations in the cerebellar peduncles. In other words, cerebral WMI may have a persistent relationship with the development of the cerebellum through afferent connecting pathways.

Along with WMI, the SNAPPE‐II was strongly associated with cerebellar structural connectivity. Specifically, lower FA in the middle and inferior cerebellar peduncles at 8 years was associated with increased SNAPPE‐II scores. The SNAPPE‐II score assesses neonatal health within the first 24 h of life by evaluating the efficiency of various physiological systems (i.e., cardiac, digestive, urinary), while also accounting for GA (via birth weight), Apgar score, and seizure activity [52]. Increased SNAPPE‐II scores are indicative of one or more of these physiological systems failing, suggesting reduced blood oxygen supply and an increased risk of mortality [52]. Relating to reduced FA, prenatal exposure and injury may disrupt the maturation of the oligodendrocytes lineage and impair myelination within WM tracts in preterm‐born children [53, 54]. Previous findings have reported Score for Neonatal Acute Physiology (SNAP‐II) have been associated with reduced FA in the corticospinal tracts in preterm infants at term‐equivalent age [55] and is predictive of brain abnormalities in the preterm population [56, 57]. Our study extends these findings, using an extension of the SNAP‐II (i.e., SNAPPE‐II), to suggest that preterm infants who present more critically ill in the first 24 h of life may have compromised WM structural integrity long‐term. Further, Dammann and colleagues [56] hypothesize that early‐life indicators of illness severity and mortality (i.e., high SNAPPE‐II scores) serve as markers for developmental vulnerability and long‐term brain injury. This aligns with our findings that the SNAPPE‐II may indicate prenatal exposures that impact long‐term cerebellar development, as demonstrated by reduced cerebellar volumes and structural connectivity changes within the cerebellar WM tracts at 8 years. Placental insufficiency is one of the main factors for fetal growth restrictions, preterm birth, and impacted neurodevelopmental outcomes [58, 59]. As such, a better understanding of the in utero antecedents of brain dysmaturation and blood oxygen levels is needed. Quantifying early prenatal exposures and blood oxygen levels may be a missing pathway impacting brain maturation and subsequent health outcomes in children born preterm.

4.3. Cerebellum and Neurodevelopmental Outcomes

Altered cerebellar development occurs through various pathways associated with prematurity and neonatal clinical factors. Previous studies have demonstrated an association between reduced cerebellar volumes and structural integrity with weakened motor skills in 18‐month‐old, 4.5‐year‐old, and 7‐year‐old preterm children [12, 14, 51]. Extending these findings, in our cohort of 8‐year‐old preterm‐born children, lower motor skills as determined through the MABC‐2 were associated with smaller posterior cerebellar volumes and reduced cerebellar structural connectivity. As a result, lower motor skills in children born preterm may be associated with reduced WM integrity of the MCP (i.e., lower FA), indicating less efficient communication between the cerebellum and cerebral cortex.

The cerebellum also contributes to cognition including executive function and language. Ranger and colleagues [7] found that 7‐year‐old preterm children had lower verbal comprehension and visual perception than their term‐born peers and had smaller volumes in lobules IV–VI, VIIB, VIIIA, VIIB, vermis, crus I, and crus II. Consistent with these findings [7, 60], in our cohort lower executive function and lower visual‐motor integration scores were related to smaller posterior cerebellar volumes (lobules VI–IX, crus I, crus II, and vermis VI–IX) at 8 years in children born preterm, and lower IQ was related to smaller anterior cerebellar subregions including lobules I–V and structural connectivity. Aligning with functional topography of the cerebellum, impacted posterior cerebellar regions are related to impacted cognitive skills, while reduced anterior cerebellar regions are related to poorer motor skills.

Importantly, motor and cognition are not solely associated with cerebellar structural and connectivity alterations; the total brain networks as well as external factors must be considered. IQ scores were also related to maternal education, suggesting children born preterm whose mothers have higher education have more favorable outcomes, warranting further investigation. Our findings suggest that neonatal clinical factors impacting reduced cerebellar volumes and reduced cerebellar structural integrity have long‐term consequences in neurodevelopmental outcomes, including motor, executive function, and cognition. Future studies should leverage cerebellar functional topology to quantify development in preterm‐born populations as this may improve our ability to identify children at highest risk for neurodevelopmental impairments.

5. Conclusions

The current study used a data‐driven approach to explore the impact of neonatal clinical factors on cerebellar development and subsequent neurodevelopmental outcomes long‐term. While other factors such as neonatal nutrition and occupational therapy throughout childhood were not measured and may contribute to some of the unaccounted‐for variability, our findings highlight that GA alone does not explain altered cerebellar development. Instead, illness severity and brain injury, reflected by early life clinical exposures, is a critical determinant of cerebellar development in preterm‐born children. Additionally, given the interconnectedness of the brain networks, impaired cerebellar growth is only a small piece of the overall effect of prematurity on brain and cognitive development. Although we cannot imply causation, we show associations between neonatal factors, long‐term cerebellar growth, and cognitive outcomes.

Overall, neonatal clinical factors including neonatal injury, reduced blood oxygen flow, and drug exposure all negatively impact cerebellar development long‐term in individuals born preterm. Altered cerebellar growth and WM structural integrity were related to poorer motor and cognitive outcomes of preterm‐born children at 8 years. These findings align with the current literature on prematurity and cerebellar development. Interestingly, our findings suggest prenatal hypoxic exposures, such as those arising from placental insufficiency, and early‐life brain injury may represent critical determinants of long‐term brain dysmaturation and subsequent health outcomes in individuals born preterm. Future research should continue to explore prenatal exposures associated with early‐life illness and how they relate to long‐term health outcomes in preterm‐born children. Further research exploring the environmental effects associated with prematurity (i.e., SES, maternal education, therapies) as they relate to long‐term development in preterm‐born children is warranted. Identifying modifiable risk factors such as minimizing the number of invasive procedures, decreasing exposure to neurotoxic medications, and early intervention to improve blood oxygen supply may improve brain development and long‐term outcomes in preterm‐born children.

Author Contributions

Victoria Rapos: writing – review and editing, writing – original draft, visualization, methodology, formal analysis, conceptualization. Steven Ufkes: writing – review and editing, software, resources, project administration, methodology, formal analysis, data curation, conceptualization. Cecil Chau: writing – review and editing; formal analysis. Jessie van Dyk: data curation, investigation. Ting Guo: writing – review and editing, formal analysis. Ruth E. Grunau: writing – review and editing; investigation, funding acquisition, project administration, methodology, data curation. Steven P. Miller: writing – review and editing, validation, supervision, resources, project administration, methodology, investigation, funding acquisition, conceptualization.

Conflicts of Interest

Dr. Miller reported grants from the Canadian Institute of Health Research and the Once Upon a Time Foundation: Raynor Cerebellum Project during the conduct of the study. Dr. Miller received support from the Bloorview Children's Hospital Chair in Pediatric Neuroscience and now from the Hudson Chair in Pediatric Medicine and the James & Annabel McCreary Chair in Pediatrics. No other disclosures were reported.

Supporting information

Table S1: Multiple regression analyses testing whether neonatal clinical factors predicted the MCP, R‐ICP, L‐ICP (the significant cerebellar tracts for component 3).

Data Availability Statement

Data will be made available upon reasonable request from qualified investigators. Access will be granted in accordance with institutional policies, applicable data user agreements, and ethical considerations to ensure the protection of participant privacy and confidentiality.