Higher order neurocognition in pediatric brain tumor survivors: What can we learn from white matter microstructure?

Abstract

Background:

Pediatric brain tumor survivors (PBTS) experience neurocognitive late effects, including problems with working memory, processing speed, and other higher order skills. These skill domains are subserved by various white matter (WM) pathways, but not much is known about these brain–behavior links in PBTS. This study examined the anterior corona radiata (ACR), inferior fronto-occipital fasciculi (IFOF), and superior longitudinal fasciculi (SLF) by analyzing associations among diffusion metrics and neurocognition.

Procedure:

Thirteen PBTS and 10 healthy controls (HC), aged 9–14 years, completed performance-based measures of processing speed and executive function, and parents rated their child’s day-to-day executive skills. Children underwent magnetic resonance imaging (MRI) with diffusion weighted imaging that yielded fractional anisotropy (FA) and mean diffusivity (MD) values. Independent samples t-tests assessed group differences on neurocognitive and imaging measures, and pooled within-group correlations examined relationships among measures across groups.

Results:

PBTS performed more poorly than HC on measures of processing speed, divided attention, and shifting (d = −1.08 to −1.44). WM microstructure differences were significant in MD values for the bilateral SLF and ACR, with PBTS showing higher diffusivity (d = 0.75 to 1.21). Better processing speed, divided attention, and shifting were associated with lower diffusivity in the IFOF, SLF, and ACR, but were not strongly correlated with FA.

Conclusions:

PBTS demonstrate poorer neurocognitive functioning that is linked to differences in WM microstructure, as evidenced by higher diffusivity in the ACR, SLF, and IFOF. These findings support the use of MD in understanding alterations in WM microstructure in PTBS and shed light on potential functions of these pathways.

1 |. INTRODUCTION

As of 2022, roughly 40,594 children and adolescents in the United States are estimated to have undergone treatment for a brain and/or other central nervous system (CNS) tumor.¹ An additional 5260 new diagnoses are anticipated in 2023. These tumors remain the greatest source of cancer-related deaths among children ages 0–19.¹ Fortunately, owing to decades of research and advances in treatments, overall survival rates now approach 75%, and reach or even exceed 90% for less aggressive tumors like low-grade gliomas.^1,2 However, this increase in survivorship is accompanied by growing recognition of an evolving cascade of physical, emotional, psychosocial, and cognitive effects that persist past treatment.³

One domain of late effects known to affect survivors of pediatric brain tumor is neurocognitive dysfunction.^2,4 These widely documented deficits can impact both global measures of overall functioning (e.g., intelligence quotients) as well as more specific skills, such as processing speed, working memory, inhibitory control, and behavioral regulation.^2,5 Processing speed, or the ability to process information rapidly, has been conceptualized as a foundational capacity needed for scaffolding other, more complex, cognitive skills.^6,7 Similarly, working memory, by which we hold and manipulate thoughts for organizing goal-directed behavior, is essential for other neurocognitive skills to function properly.⁸ When impacted following pediatric brain tumor, these and other executive skills (e.g., sustained and divided attention, inhibitory control), and downstream skills in day-to-day behavioral regulation, have potential to greatly affect a child’s academics and quality of life.²

Neurocognitive skills are subserved or supported by specific brain regions and the pathways that integrate them. Specifically, prominent white matter (WM) pathways are thought to support neurocognition via their role in connecting cortical regions that interact dynamically in supporting skill execution. A recent review paper summarized links among WM microstructure and executive function in healthy children and adolescents, and found that a number of projection and long-range association fibers can be linked with these higher order skills, including the bilateral anterior corona radiata (ACR), inferior fronto-occipital fasciculus (IFOF), and superior longitudinal fasciculus (SLF).⁹

The IFOF and SLF are WM pathways classified as long association fibers that connect different lobes of the brain and form prominent fiber bundles.¹⁰ The IFOF projects anteriorly from the occipital to the frontal lobe, and has been linked with neurocognitive skills such as working memory, attention, executive function, and processing speed in both pediatric cancer survivors and typically developing children.^11–14 The SLF forms a large arc (also known as the arcuate fasciculus) that sends branches bidirectionally into the frontal, parietal, occipital, and temporal lobes.¹⁰ These regions of the brain are similarly implicated in attention and working memory, as well as other executive functions like set shifting and inhibition.^15–17 Connections between the SLF and processing speed are inconsistent, with some studies supporting it¹⁸ and others failing to find an association¹⁹; this research has largely focused on adults, with few existing studies that examine children. The ACR is another WM structure that is classified as a projection fiber, and radiates across anterior cortical regions and comes together in the brain stem, connecting cortical and subcortical gray matter. Therefore, it has interconnections broadly throughout the cortex, thus communicating with many regions underlying a variety of neurocognitive processes.¹⁰

While this existing body of research provides some insight into the WM substrates of neurocognition, some important gaps remain that we aim to address here. First, neuroimaging literature in child and adolescent populations as a whole lags far beyond adult literature. This is even more so true for research focused on survivors of pediatric brain tumor who may be vulnerable to WM microstructural changes due to diagnostic and treatment factors.²⁰ Second, relative to tracts like the SLF and IFOF, others like the ACR have yet to be clearly linked to specific neurocognitive functions in survivors. Finally, nearly all of the studies described above relied solely on fractional anisotropy (FA) as a measure of WM microstructure.^21–25 FA is a very common metric that indicates the degree of diffusion anisotropy within a given voxel as a ratio of diffusion in (a) the principal direction, relative to (b) the two other orthogonal directions. An FA value of 1 reflects diffusion solely in the principal direction (e.g., directly along an axon), whereas an FA value of 0 reflects dispersed diffusion across all directions. Historically, many have ascribed to the notion that higher FA is “better” and interpreted data accordingly. However, the complexities of FA are increasingly recognized given the prevalence of crossing fibers that can reduce computed FA values.^26,27 In a recent review, Figley and colleagues²⁸ advocated for avoiding interpretation of FA, or axial (AD) or radial diffusivity (RD)—two markers of unidirectional diffusivity—in isolation; rather, diffusion metrics like mean diffusivity (MD), which integrates diffusion along the three tensor dimensions, can provide a more rounded perspective.²⁸

In the present study, we examine how anisotropy-based versus diffusion-based metrics of WM microstructure (i.e., FA vs. MD) support neurocognition in pediatric brain tumor survivors (PBTS), within the tracts of interest described above. This study extended from a larger investigation of social and behavioral outcomes of childhood brain tumors (i.e., the “parent study”) by acquiring multimodal neuroimaging. In this investigation, we analyzed WM pathway microstructure obtained through diffusion-weighted magnetic resonance imaging (MRI) in a subsample of participants who had completed data collection for the parent study. We explored links among these neuroanatomical markers and performance-based and behavioral measures of neurocognitive domains, including processing speed, executive function, and behavioral regulation. Consistent with prior research on the impact of pediatric brain tumor on neurocognition, we hypothesized that survivors would demonstrate impairment in executive function relative to healthy children, as evidenced through poor performance on measures of processing speed, working memory, set shifting, and behavioral regulation. In addition, we expected group differences in brain-based markers of WM microstructure, hypothesizing that PBTS will show reduced FA and higher MD. Finally, given our hypothesis that these brain markers support aspects of neurocognition, we further hypothesized that children who performed more poorly on performance and parent-reported measures of neurocognition would also demonstrate evidence of differences in WM microstructure, reflected in correlations among task performance and reduced FA and increased MD.

2 |. METHODS

2.1 |. Participants

The final sample for the current analyses consisted of 13 PBTS (nine males) and 10 healthy comparison peers (healthy controls [HC]; six males). As noted above, participants included a subset of participants from a parent study that assessed neurocognitive and social outcomes following pediatric brain tumor as manifested in the classroom and home settings. Eligible participants for the MRI component included those who had completed both school- and home-based data collection for the parent study and agreed to be contacted to receive more information about the MRI component. Inclusion criteria for the parent study for both groups specified that children were between 9 and 14 years of age, lived within 120 miles of the hospital, and were fluent in English, with at least one caregiver also fluent in English. Additionally, PBTS were one or more years post treatment for a primary, intracranial tumor without evidence of active, progressive disease. PBTS were excluded from the parent study if they had a diagnosis of neurological or neurodevelopmental disorder prior to their brain tumor diagnosis (e.g., neurofibromatosis) or did not receive any academic instruction in a mainstream classroom. HC were excluded if they had a diagnosed chronic medical condition (i.e., requiring subspecialty management, over 6 months in duration). Additional exclusionary criteria for the MRI component included having received inpatient treatment for a mood, behavior, or psychiatric disorder; having difficulties with motor skills, vision, or hearing that precluded completion of study measures; or having any MRI contraindications or orthodontics that could cause distortion or image artifacts. PBTS who had a ventriculoperitoneal shunt implant were excluded if it was a programmable or non-approved shunt.

Thirty-six PBTS families who completed both school and home visits for the parent study agreed to be contacted about possible participation in the MRI component. Of those, 34 (94.4%) were successfully reached for MRI recruitment and eligibility screening. Five families declined, three of whom lacked interest and two disliked MRIs. Sixteen children were ineligible due to titanium plate implants (7), orthodontic devices (6), non-approved shunts (2), or requirement of sedation (1). Thirteen (36.1%) were found eligible, agreed to enroll, and completed the MRI visit. Thirty HC, matched during parent study participation for age and sex to PBTS, agreed to be contacted for the MRI study. Twenty-three (76.7%) were successfully reached by phone and completed eligibility screening. Nine of those (39.1%) declined due to a lack of interest (7), distance from the hospital (1), and claustrophobia (1). A further four participants were deemed ineligible due to braces, leading to a total of 10 (33.3%) children who were found eligible, enrolled, and completed the MRI visit. Children included in the MRI component did not differ from others in the parent study excluded based on age, sex, race, or ethnicity. Informed consent and assent for the MRI component were obtained and procedures were approved by the institutional review board.

Demographic and socioeconomic information is provided in Table 1. PBTS and HC did not significantly differ in age at participation, sex, race, ethnicity, annual family income, or caregiver sex or marital status (Table 1). Survivors’ mean age at diagnosis was 5.30 years (SD = 3.33; range: 0.48–10.69). Mean time since diagnosis was 7.46 years (SD = 3.11; range: 2.80–13.44), with mean time off treatment of 5.98 years (SD = 2.72; range: 2.79–13.07). All survivors underwent some extent of surgical resection as part of treatment, with four (30.8%) also receiving chemotherapy and five (38.5%) receiving cranial or cranio-spinal radiation. Tumor histologies included astrocytomas (4), medulloblastomas (4), gangliogliomas (4), and glioneuronal tumor (1). These were located supratentorially (7), infratentorially (6), and in the midbrain tectal region (1).

TABLE 1.

Group comparisons of demographic and socioeconomic factors.

	PBTS (n = 13), M (SD)	HC (n = 10), M (SD)	p	d
Age in years	12.76 (2.61)	12.70 (2.20)	.951	0.03
Parent education in years	15.23 (1.83)	15.90 (2.92)	.536	−0.28
	n(%)	n(%)	p	V
Male	9 (69.2%)	6 (60.0%)	.645	.096
White	12 (92.3%)	10 (100.0%)	.370	−.187
Non-Hispanic	13 (100.0%)	9 (90.0%)	.244	−.243
Family income >$75,000	6 (50.0%)a	8 (80.0%)	.145	.311
Caregiver sex: female	13 (100.0%)	10 (100.0%)	–	–
Parent marital status: partnered	11 (84.6%)	9 (90.0%)	.704	.079

Abbreviations: HC, healthy controls; M, mean; PBTS, pediatric brain tumor survivors; SD, standard deviation.

^aOut of 12 participants for this descriptor variable due to one missing data.

2.2 |. Measures

Selected subtests of the Wechsler Intelligence Scale for Children-Fourth Edition (WISC-IV)²⁹ were administered to measure processing speed. All participants completed two subtests: Coding and Symbol Search. Both are timed processing speed subtests that require speed, accuracy, and perceptual organization. The WISC-IV has strong reliability, with internal consistency reliability for the processing speed composite at α = .88.³⁰

The Test of Everyday Attention for Children (TEA-Ch)³¹ was administered to assess multiple aspects of attention. TEA-Ch covers three domains of attention: selective attention, sustained attention, and switching/divided attention. We examined participants’ sustained attention using two subtests, Walk-Don’t-Walk and Code Transmission. All participants also completed Creature Counting, which measures switching/divided attention. Test–retest reliability for the subtests ranges from r = .71 to .78.³¹

The Behavior Rating Inventory of Executive Function (BRIEF)³² is a parent-rated measure that assesses executive function behaviors as they manifest in the daily environment. The BRIEF contains nine subscales forming two major indices, Behavioral Regulation and Metacognition, and one overall index, Global Executive Composite. For this study, the BRIEF was completed by participating mothers of PBTS and HC. The BRIEF demonstrates high test–retest reliability (r = .82) and strong internal consistency (α = .80–.98).³²

2.3 |. Image acquisition

Participants completed MRI without sedation or contrast on a 3 Tesla Siemens Prisma scanner dedicated to research (Prisma, Siemens Healthineers). Each session included acquisition of high-resolution structural images with whole-brain coverage. The T1-weighted magnetization-prepared spoiled gradient recalled echo (MPRAGE) sequence was acquired with interleaved, single-shot slice acquisition in the anterior–posterior direction (176 slices, TR = 1950 ms, TE = 4.44 ms, FOV = 256 × 256, voxels = 1 mm³, flip angle 12°). Diffusion-weighted imaging yielded measures of WM microstructure, which was assessed using a 30-direction sequence with anterior–posterior phase encoding (65 slices, TR = 8300 ms, TE = 85 ms, FOV= 288 mm, voxels = 2 mm³). Two diffusion weightings were set to a low b-value of 0 s/mm² and a high b-value of 700 s/mm².

2.4 |. MRI post-processing

Structural data were pre-processed to correct for field bias, individual variation in brain image dimension, and alignment to the AC-PC plane. T1-weighted images were adapted in native space, then co-registered for analyses and to execute protocols around screening for incidental abnormalities. For diffusion tensor imaging (DTI) analysis, we inspected diffusion-weighted imaging data and corrected for artifacts that could compromise data accuracy or reliability. Data were aligned and registered to pre-processed anatomic images described above using the Advanced Normalization Tools (ANTs) software package, then underwent tensor fitting (http://stnava.github.io/ANTs/). ANTs provide all necessary pre-processing steps consisting of well-vetted, previously published algorithms for bias correction,³³ brain extraction,³⁴ n-tissue segmentation,³⁵ template construction,³⁶ and image normalization.³⁷ Images were mapped onto the ENIGMA-DTI template³⁸ and skeletonized for region of interest processing, including extraction of FA and MD across tracts.

2.5 |. Statistical analyses

Before hypothesis testing, histograms and descriptive statistics were generated to determine if any outliers skewed the data. Outliers were assessed within rather than across groups, because we anticipated a difference in distribution based on group, and sought to avoid truncating potentially meaningful variability. No outliers were found in either group for neurocognitive assessments and behavioral measures, so parametric statistics were used for hypothesis testing.

Independent sample t-tests were used to analyze group differences on neurocognitive and behavioral measures, and on WM microstructure as measured by DTI. Due to the small group samples, effect sizes were also calculated for group comparisons; we focused on medium (Cohen’s d ≥ .50) and large effects (d ≥ .80)³⁹ in subsequent interpretation. Finally, pooled within-group correlation was used to examine the relationships among neurocognitive assessment and WM microstructure across groups, with analysis focusing on medium (≥.3) and large (≥.5) correlations.³⁹

3 |. RESULTS

3.1 |. Group differences in neurocognitive assessment

Group scores on neurocognitive measures are summarized in Table 2. PBTS performed worse on measures of processing speed than HC, with mean performance roughly a full standard deviation below their comparison peers and nearly two-thirds of a standard deviation below normative expectation. PBTS also performed significantly worse than HC on the Creature Counting subtest of the TEA-Ch, with mean performance over a standard deviation below both their healthy peers and normative expectation. PBTS’ performance on the Code Transmission subtest fell more than a standard deviation below normative expectation, but did not reach statistical significance in group comparisons, because the HC group performed more poorly on that measure than expected. Group differences on the BRIEF parent report form are also summarized in Table 2; only the Shift subscale differed significantly by group, with PBTS rated as having more difficulty than HC.

TABLE 2.

Group means on neurocognitive and behavioral measures.

	PBTS (n = 13), M (SD)	HC (n = 10), M (SD)	t	p	Cohen’s d
WISC-IV
Coding SS	6.85 (2.48)	9.60 (2.76)	−2.52	.020*	−1.06
Symbol Search SS	9.92 (2.06)	11.90 (2.28)	−2.18	.041*	−.92
PSI Composite Score	90.77 (12.36)	104.50 (11.34)	−2.74	.006**	−1.15
TEA-Ch
Creature Counting	6.85 (2.19)	10.20 (2.49)	−3.43	.002**	−1.44
Walk Don’t Walk	6.46 (4.94)	7.90 (3.11)	−0.85	.404	−.34
Code Transmission	9.08 (3.64)	10.90 (2.03)	−1.42	.170	−.60
BRIEF
Behavioral Regulation	61.17 (12.64)	59.00 (7.30)	0.46	.652	.20
• Inhibit	57.62 (12.86)	64.50 (12.62)	−1.28	.213	−.54
• Shift	64.00 (14.16)	50.78 (9.13)	2.44	.025*	1.08
• Emotional Control	56.00 (12.35)	56.30 (8.87)	−0.07	.949	−.03
Metacognition	68.38 (15.49)	65.20 (9.99)	0.57	.578	.24
• Initiate	61.92 (13.73)	59.40 (11.05)	0.47	.640	.20
• Working Memory	71.69 (15.43)	62.60 (9.40)	1.64	.116	.69
• Plan/Organize	65.69 (16.33)	62.90 (10.51)	0.50	.625	.20
• Organization of Materials	57.23 (9.93)	60.50 (6.20)	−0.91	.373	−.38
• Monitor	65.85 (13.78)	66.30 (13.38)	−0.08	.938	−.03
Global Executive Composite	66.17 (15.24)	63.11 (9.27)	0.53	.602	.23

Note: Effect sizes in italics surpassed the threshold for medium to large effect (Cohen, 1988).

Abbreviations: BRIEF, Behavior Rating Inventory of Executive Function; CT, Code Transmission; HC, healthy controls; M, mean; PBTS, pediatric brain tumor survivors; PSI, Processing Speed Index; SD, standard deviation; SS, Scaled Score; TEA-Ch, Test of Everyday Attention in Children; WDW, Walk Don’t Walk; WISC-IV, Wechsler Intelligence Scale for Children-Fourth Edition.

^*p < .05

^**p <.01.

3.2 |. Group differences in white matter microstructure

Group differences in mean FA and MD of WM microstructure are summarized in Table 3. Differences were most pronounced in MD values (see Figures 1 and 2), with PBTS showing higher MD, and seen across almost all regions (left, right, and bilateral) for the SLF and ACR (d = .52–1.02). Differences in MD in the bilateral and right IFOF were not statistically significant, but involved medium effect sizes. No differences in FA values were statistically significant, with medium to large effect sizes only found for the right (d = −.79) and bilateral (d = −.66) IFOF.

TABLE 3.

Group means in white matter tract microstructure.

	PBTS (n = 13), M (SD)	HC (n=10), M (SD)	t	p	Cohen’s d
Fractional anisotropy
Bilateral IFOF	.572 (.069)	.609 (.034)	−1.71	.104	−.66
L IFOF	.590 (.068)	.592 (.047)	−0.07	.948	−.03
R IFOF	.558 (.104)	.622 (.032)	−2.09	.054	−.79
Bilateral SLF	.559 (.025)	.560 (.029)	−0.11	.911	−.05
L SLF	.562 (.023)	.563 (.023)	−0.11	.910	−.05
R SLF	.556 (.040)	.557 (.038)	−0.09	.931	−.04
Bilateral ACR	.531 (.028)	.541 (.025)	−0.87	.397	−.36
L ACR	.530 (.026)	.542 (.022)	−1.12	.274	−.47
R ACR	.532 (.034)	.541 (.029)	−0.60	.556	−.25
Mean diffusivity
Bilateral IFOF	.0018 (.0002)	.0016 (.0003)	1.68	.108	.72
L IFOF	.0015 (.0002)	.0015 (.0002	0.69	.497	.30
R IFOF	.0019 (.0004)	.0017 (.0004)	1.55	.138	.66
Bilateral SLF	.0011 (.0002)	.0010 (.0000)	2.26	.042*	.89
L SLF	.0011 (.0002)	.0010 (.0000)	1.90	.080	.75
R SLF	.0011 (.0002)	.0010 (.0000)	2.49	.027*	.98
Bilateral ACR	.0013 (.0002)	.0011 (.0001)	2.75	.012*	1.18
L ACR	.0012 (.0001)	.0011 (.0001)	2.82	.010*	1.21
R ACR	.0013 (.0002)	.0011 (.0001)	2.16	.043*	.93

Note: Effect sizes in italics surpassed the threshold for medium to large effect (Cohen, 1988).

Abbreviations: ACR, anterior corona radiata; HC, healthy controls; IFOF, inferior fronto-occipital fasciculus; L, left; M, mean; PBTS, pediatric brain tumor survivor; R, right; SD, standard deviation; SLF, superior longitudinal fasciculus.

^*p < .05;

^**p <.01.

FIGURE 1.

Group-level differences in white matter mean diffusivity (MD), based on HC-PBTS subtraction. Warmer colors reflect greater MD in the PBTS group relative to the HC group. Images generated via FSL Randomise, with cluster-corrected thresholding, presented as t-statistics, p < .05.

FIGURE 2.

Individual exemplar white matter tractography in one pediatric brain tumor survivor (top; 13.11-year-old male, history of posterior fossa medulloblastoma with surgery, chemotherapy, and radiation) relative to a matched healthy control (bottom; 12.51-year-old male).

3.3 |. Correlations among neurocognitive measures and DTI

Next, we examined whether individual differences in WM microstructure were associated with neurocognitive performance and parentrated day-to-day executive skills (Table 4). Analyses were limited to neurocognitive domains where group differences were statistically significant, and limited to WM features where group differences met or exceeded a medium effect size, to conserve statistical power. No consistent pattern emerged for links among WM FA and neurocognition. Associations among WM MD and neurocognition, however, were more notable. Specifically, better performances on measures of processing speed on the WISC-IV were related to lower MD in the right hemisphere IFOF and bilateral SLF and ACR. Better performance on the TEA-Ch Creature Counting subtest was similarly associated with lower MD in these same regions, bilaterally. Finally, lower parent-rated difficulties with shifting on the BRIEF was also associated with lower MD in IFOF, SLF, and ACR, again bilaterally.

TABLE 4.

Correlations among neurocognitive task performance, parent ratings, and white matter microstructure.

	Coding	Symbol Search	PSI	Creature Counting	BRIEF Shift
Fractional anisotropy
Bilateral IFOF	.259	−.001	.157	.168	−.093
R IFOF	.105	−.025	.049	.131	−.004
Mean diffusivity
Bilateral IFOF	−.202	−.120	−.180	−.297	.332
R IFOF	−.317	−.236	−.311	−.358	.367
Bilateral SLF	−.399	−.437*	−.464*	−.366	.513*
L SLF	−.376	−.424*	−.443*	−.323	.499*
R SLF	−.403	−.431*	−.463*	−.391	.502*
Bilateral ACR	−.252	−.358	−.336	−.520*	.490*
L ACR	−.267	−.401	−.372	−.666*	.409
R ACR	−.205	−.277	−.263	−.356	.459*

Note: Heatmap gradient based on Cohen’s criteria for medium (r > .30) and large (r > .50) correlations; darker tones indicate larger effect.

Abbreviations: ACR, anterior corona radiata; BRIEF, Behavior Rating Inventory of Executive Function; IFOF, inferior fronto-occipital fasciculus; L, left; PSI, Processing Speed Index; R, right; SLF, superior longitudinal fasciculus.

^*p < .05.

4 |. DISCUSSION

Despite medical advancements that have led to an increase in survival rates for pediatric brain tumors, PBTS continue to experience neurocognitive late effects that negatively impact their quality of life. This study sought to characterize how differences in processing speed, working memory, set shifting, and behavioral regulation may be associated with measures of WM microstructure in several prominent regions of the brain: IFOF, SLF, and ACR.

Group comparisons of performance-based measures of neurocognition confirmed poorer performance in PBTS on tests of processing speed, set shifting, and sustained attention; however, sustained attention differences did not meet statistical significance because HC also performed more poorly than expected. Similarly, differences on most BRIEF subscales (with the exception of Shift) did not reach significance because of unexpected difficulties reportedly experienced by HC. This underperformance may have further obscured differences that would typically be present between PBTS and normative performance of healthy peers. In contrast, as expected, WM microstructure metrics showed significantly higher MD values in SLF and ACR tracts for PBTS, and nonsignificant but medium-to-large effect sizes for IFOF tracts. These findings indicate a higher degree of water diffusivity that has been associated with poorer outcomes in both pediatric brain tumor²² and other populations.^40,41

While we did not find a consistent pattern of associations among WM FA and neurocognitive outcome, lower MD in parts of the SLF was moderately correlated with better processing speed and set shifting, whereas lower MD in the ACR was moderately correlated with better switching/divided attention and set shifting. Lower diffusivity in these regions was also associated with fewer parent-reported difficulties with shifting. These findings shed light on potential functions of the ACR, whose contributions to cognitive flexibility align with its role as a projection fiber that generally helps support more complex, higher order regulatory skills needed for executive functioning.⁴²

The lack of statistical power to conduct more advanced analyses, such as mediation, is one of the several limitations that stem from the small sample size. These analyses only examined those neurocognitive domains and WM metrics that showed significant group differences or a medium effect size, in order to balance our goal of examining those links while also acknowledging the limited statistical power of the small sample and desire to balance Type I and Type II errors. This conservative approach did not incorporate all medium effect sizes for neurocognitive measures, and may have obscured additional meaningful associations. Furthermore, our analyses were limited to tract-based spatial statistics of entire WM pathways. We selected this approach in acknowledgment of limited statistical power, out of concern that using a voxel-wise approach may result in erroneous clusters and uninterpretable local maxima, given the small and heterogeneous sample. Unfortunately, using a tract-level approach does limit our ability to detect differences in subregions of the tract that may be clinically meaningful. We hope to overcome this limitation in larger future studies.

Across groups, when considering both, who declined participation and those deemed ineligible to participate, only roughly a third of potential participants completed the MRI component, highlighting the inability to generalize to the broader population of brain tumor survivors. Additionally, treatment was heterogeneous and statistical power precluded our ability to examine variability based on modality; very few participants underwent chemotherapy or radiation in addition to surgical resection. The data were also underpowered to allow for exploration of differences based on tumor histology or location. Furthermore, we were unable to consider additional treatment-related factors like exposure to anesthesia; many survivors have required sedation not just for surgical resection, but also to obtain surveillance MRI, which has been found in other population (e.g., survivors of leukemia) to contribute to both neurocognitive outcomes and WM diffusivity.⁴³ Participants were also recruited from a single site, and their racial and ethnic demographics are not representative of the surrounding area. The participants from both groups in this study were almost universally White and non-Hispanic. According to statistical reports of CNS tumor incidence rates, prevalence of tumor diagnoses is generally higher among White and non-Hispanic populations.^1,44 This may reflect broader health inequities in which historically disadvantaged or minoritized groups face decreased access to diagnostic services and treatment as a result of systemic racism during acute diagnosis and treatment.⁴⁵ For our specific sample, it may also reflect unequal follow-up rates for clinical care, resulting in outdated contact information for parent study recruitment. Regardless, future studies would benefit from larger, multisite recruitment efforts that incentivize and support participation of minoritized populations.

Other limitations include the aforementioned underperformance of HC on behavioral ratings, which may have resulted from self-selection bias. Qualitative observations by research staff noted that multiple parents of HC indicated an interest in participation due to the benefits of receiving a medically interpreted brain MRI, suggesting potential concern about their children. In examining the data from the parent study, PBTS who participated in the MRI portion were very similar to those who did not participate in the MRI portion in neurocognitive task performance and BRIEF parent ratings. In contrast, HC who participated in the MRI portion were rated by parents as having worse behavioral regulation (d = −.97), metacognition (d = −.62), and global executive function (d = −.86) relative to HC who did not participate in the MRI visit, again supporting a potential self-selection bias. Lastly, because this study is cross-sectional, we also cannot determine directionality of associations between neurocognitive and imaging measures. Future studies should consider a longitudinal design to further elucidate the cause–effect relationship between neurocognitive functioning and WM integrity.

Despite these limitations, this study provides additional evidence suggesting links among WM microstructure and late effects commonly experienced by survivors of pediatric brain tumor. Our findings support routine monitoring of neurocognitive outcome in survivorship, particularly in the area of attention and executive function; findings further highlight potential areas for targeted screening during survivorship clinic visits to monitor function between full assessments. Embedding personnel with expertise in surveillance and assessment of these domains (e.g., neuropsychology) in clinic settings would be ideal. The findings further highlight the importance of integrating multiple DTI metrics in future research in order to better capture these links, delineate heterogeneity based on diagnosis and/or treatment, and better prognosticate long-term outcomes in PBTS.

Abbreviations:

ACR

anterior corona radiata

FA

fractional anisotropy

HC

healthy control

IFOF

inferior fronto-occipital fasciculus

MD

mean diffusivity

PBTS

pediatric brain tumor survivor

SLF

superior longitudinal fasciculus

WM

white matter

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request and in accordance with institution data use agreements.