Neurobiological and psychological indicators of post-traumatic stress symptoms in children and adolescents treated for a posterior fossa brain tumour

Abstract

Background

Despite the established associations between paediatric brain tumour treatment, distress, and post-traumatic stress symptoms (PTSS), neither the treatment nor tumour pathology reliably account for symptom severity at the individual level. This study examined whether treatment-related changes in fronto-limbic white matter (WM), brain tissue thought to support emotion, is associated with additional variance in PTSS beyond the effect of treatment and endorsing having experienced a distressing event.

Methods

Thirty-six children and adolescents treated for a posterior fossa brain tumour and 17 typically developing children (TDC) completed a questionnaire assessing whether they had experienced a distressing event and measured PTSS and underwent diffusion tensor imaging. Diffusivity metrics of fronto-limbic tracts and tracts that do not support emotional functioning (control tracts) were obtained. A continuous measure of the physical, cognitive, and emotional strain of therapy (i.e., treatment burden), was estimated using the Neurological Predictor Scale. Partial Least Squares path modelling was used in an exploratory, theory-guided framework to examine statistical associations among treatment burden, distress, WM, and PTSS.

Results

Patients demonstrated greater radial diffusivity than TDC across fronto-limbic and control tracts. Fronto-limbic WM was associated with PTSS severity within the subclinical range and mediated the effects of treatment burden on PTSS. The control tract WM was not associated with PTSS. PTSS severity was also associated with endorsing a distressing event.

Conclusions

The findings support two indicators associated with dimensional variation in PTSS severity: an expected psychological indicator, through endorsing having experienced a distressing event, and a neurobiological indicator, via treatment-related changes to fronto-limbic WM. These results reflect correlational patterns in subclinical symptoms and should be interpreted as exploratory.

Highlights

• •Assessed risk for post-traumatic stress in posterior fossa paediatric brain tumour.
• •Used diffusion imaging and path modelling to test neurological indicator to post-traumatic stress.
• •White matter microstructure was directly associated with post-traumatic stress.
• •White matter may serve as a contributor to post-traumatic stress vulnerability.

1. Introduction

Among childhood cancers, brain tumours carry one of the highest treatment burdens, defined as the cumulative effects of disease and medical intervention (Powell et al., 2021). The burden of treatment reflects not only tumour characteristics but also the intensity, duration, and neurological consequences of medical care. Treatment regimens vary by the type and severity of the tumour: non-malignant tumours are often treated with gross total resection, whereas malignant tumours typically require surgery and cranial radiation therapy (CRT) with or without chemotherapy (Oyefiade et al., 2021; Pancaldi et al., 2023; Pollack et al., 2019). Accordingly, children treated for a brain tumour experience a broad range of treatment exposures, even within anatomically defined tumour groups. CRT is associated with a substantially greater treatment burden than surgery alone, including more clinic or hospital visits and invasive procedures, as well as poorer neurocognitive outcomes. CRT has also been linked to structural brain alterations, including damage to white matter (WM), the tissue that facilitates rapid neural communication essential to cognitive and emotional functions (Szentes et al., 2018; Pancaldi et al., 2023; Robinson et al., 2013; Keng et al., 2021; Fuemmeler et al., 2002; Coltin et al., 2023; Marusak et al., 2018; Ailion et al., 2017; Taïeb et al., 2003; Fields, 2008). Given these neurocognitive and neurobiological consequences, diagnosis and treatment may be distressing for children and their families, often eliciting feelings of sadness, worry, and fear (Bruce, 2006; Fuemmeler et al., 2001; Vitek et al., 2007). Patients have reported their most distressing treatment-related memories as hospital stays, injections, bone marrow aspirates, spinal taps, checkups, taking pills, and side effects such as pain, hair loss, and nausea (Kazak et al., 1996), highlighting the many ways that treatment burden is associated with distress (Siehl et al., 2018; Aleksonis et al., 2021; Wier et al., 2019a; Baron Nelson et al., 2016). In addition to these specific procedures, the unpredictability of treatment response and long-term health prospects may contribute to the persistence or intensification of distress during and after treatment (Taïeb et al., 2003; Kazak et al., 1996; Stuber et al., 2010; Hobbie et al., 2016; Sharkey et al., 2020).

Although some individuals recover over time, others show escalation to post-traumatic stress symptoms (PTSS), including intrusive memories, hypervigilance, and avoidance of emotionally evocative cues (Fuemmeler et al., 2002; Sharkey et al., 2020; Simkhaev, 2024; Shah et al., 2015; Holland and Bultz, 2007). PTSS is best conceptualized dimensionally, with substantial variability in symptom severity occurring below diagnostic thresholds. In paediatric medical populations, subclinical PTSS is nonetheless associated with functional impairment and heightened emotional vulnerability, even in the absence of a formal diagnosis. Accordingly, examining continuous variation in PTSS severity may provide critical insight into early or emerging emotional risk that is not captured by categorical diagnostic approaches (Fuemmeler et al., 2002; Sharkey et al., 2020; Simkhaev, 2024; Shah et al., 2015; Holland and Bultz, 2007). Rates of PTSS in children and adolescents who have undergone treatment for an aggressive malignant tumour, such as medulloblastoma, can reach as high as 71 %, in contrast with the 35 % observed in heterogeneous brain tumour samples and the 0–12.5 % found in mixed childhood cancer samples (Bruce, 2006; Bruce et al., 2011). However, one study found no significant association between diagnosis and PTSS (Bruce et al., 2011), and subjective experiences of illness and medical care show stronger associations with PTSS than objective medical variables such as diagnosis or treatment type (Taïeb et al., 2003; Bruce, 2006). These findings indicate that the severity of symptoms does not correlate exclusively with tumour diagnosis; rather, it reflects the manner in which treatment-related events are experienced and appraised. Importantly, PTSS results from one or more distressing or traumatic experiences; consequently, variability in symptom severity is inherently contingent upon whether and how individuals perceive treatment-related events as distressing. This underscores the importance of simultaneously considering treatment exposure and subjective distress when evaluating PTSS outcomes. Nonetheless, the factors that distinguish individuals who develop heightened PTSS following treatment remain insufficiently understood. One hypothesis posits that PTSS reflects not only subjective distress, but also neurobiological effects associated with treatment variations. In this context, WM microstructure serves as a potential neurobiological correlate for PTSS severity (Wier et al., 2019; Stuber et al., 2010).

Work in typically developing children (TDC) shows associations between distress and WM microstructure, including reduced fractional anisotropy (FA) and increased radial diffusivity (RD) in fronto-limbic tracts commonly implicated in emotional processes (see reviews by Daniels et al., 2013; and Siehl et al., 2018). In TDC, PTSS has been associated with dysregulation in brain regions involved in emotional processing, including the amygdala, hippocampus, and thalamus. Communication between these limbic regions and the cortex is supported by fronto-limbic WM tracts, including the cingulum bundle (CG), inferior fronto-occipital fasciculus (IFO), and uncinate fasciculus (UF). FA reflects fibre organization, whereas RD is sensitive to changes related to demyelination; lower FA and higher RD are generally interpreted as markers of WM injury (Marusak et al., 2018, 2019; Harnett et al., 2020; Admon et al., 2013). Alterations in FA and RD in fronto-limbic tracts have been consistently observed among trauma-exposed TDC relative to non-distressed controls (Daniels et al., 2013; Siehl et al., 2018).

To date, findings from brain tumour populations remain mixed. Some studies report associations between fronto-limbic WM microstructure and a higher likelihood of endorsing a distress event (Aleksonis et al., 2021; Wier et al., 2019). Another study reported WM microstructural differences between patients and their sibling controls, but found no apparent association with PTSS (Baron Nelson et al., 2016). Still, another found no association between fronto-limbic WM microstructure and a dichotomized, parent-reported PTSS measure (Leenders et al., 2024). For posterior fossa tumour populations, evidence suggests that treatment-related injury to cerebellar midline structures may be related to affective regulation in addition to fronto-limbic injury (Micklewright et al., 2008; Soto et al., 2023; Moxon-Emre et al., 2019). This raises the possibility that multiple neural systems affected by treatment may be jointly associated with PTSS, thereby potentially accounting for inconsistent findings in the literature. These inconsistencies may also reflect methodological limitations: sibling controls may underestimate group differences, and parent reports often diverge from child self-reports, particularly in brain tumour populations (Schulte and Barrera, 2010; Alderfer et al., 2010; Urso and Mastroyannopoulou, 2018). Moreover, treating PTSS as a dichotomous variable limits sensitivity to meaningful variation in symptom severity (De Bellis and Zisk, 2014), and failing to account for treatment burden may obscure critical individual differences (Powell et al., 2021). Posterior fossa tumour populations are also inherently heterogeneous with respect to tumour pathology and treatment exposure, which may contribute to variability in reported associations. Moreover, most evidence is cross-sectional, making the directionality of the relationship among WM and distress unclear: WM injury may be associated with greater vulnerability to distress, distress may be associated with differences in WM development, or both processes may occur in parallel (Daniels et al., 2013; Siehl et al., 2018). Together, these limitations highlight the need for models that simultaneously consider distress exposure, treatment burden, and neurobiological variation when examining PTSS severity (Urso and Mastroyannopoulou, 2018; De Bellis and Zisk, 2014). In posterior fossa tumour populations, these factors may converge due to shared anatomical location and overlapping exposure to treatment-related neurotoxicity, despite heterogeneity in tumour pathology.

This study investigates whether fronto-limbic WM microstructure is associated with variability in PTSS severity within the subclinical range, beyond having experienced a distressing event, in understanding how treatment burden is associated with PTSS in children and adolescents treated for a brain tumour (Marusak et al., 2019; De Bellis, 2001; Marsac et al., 2014). Treatment burden is analyzed through both categorized treatment exposure (CRT versus non-CRT, with TDC serving as a non-treatment comparison group) and a continuous cumulative score. We compare fronto-limbic WM microstructure and PTSS among patients who received CRT, those who did not (non-CRT), and TDC. The non-CRT group serves as a critical treatment-exposed comparison, capturing variability in surgical and medical management in the absence of cranial radiation. We acknowledge that this design does not permit complete separation of tumour-specific biological effects from treatment-related processes; addressing this limitation will require larger, tumour-homogeneous samples in future work. If group differences in WM are observed, we will conduct exploratory Partial Least Squares Path Modelling (PLS-PM) to examine associations among treatment burden, WM microstructure, and PTSS, while accounting for endorsement of distressing events. This theory-guided approach evaluates direct and indirect associations without assuming temporal, causal, or predictive relationships, and is well-suited to small samples common in paediatric brain tumour research (Sarstedt et al., 2021; Griffiths-King et al., 2025). To address limitations of prior studies, we account for sibling status, use a continuous self-report measure of PTSS, and quantify treatment burden continuously using a validated index that reflects cumulative treatment exposure and related complications (Oyefiade et al., 2021; Micklewright et al., 2008; Soto et al., 2023; Moxon-Emre et al., 2019). The study is not intended to identify clinical PTSS or model disorder onset, but to characterize correlational patterns underlying individual differences in symptom severity within a predominantly subclinical sample.

We hypothesize that PTSS will differ between the patient and TDC groups, but not between the CRT and non-CRT groups. We hypothesize that fronto-limbic WM microstructure will vary across the groups, with the most pronounced differences observed in the CRT group, followed by the non-CRT group, compared to TDC. We hypothesize that fronto-limbic WM microstructure and endorsement of distressing events will statistically mediate the relationship between treatment burden and PTSS. Aligned with the idea that PTSS depends on experiencing distress, mediation models specifically include endorsement of a distressing event, rather than viewing PTSS as separate from such experiences. A secondary model will replace fronto-limbic tracts with control tracts (arcuate fasciculus, optic radiation) to assess anatomical specificity; we hypothesize that WM microstructure in the control tracts will not be associated with PTSS.

2. Methods

2.1. Participants

Data were analyzed from 36 children and adolescents treated for a posterior fossa brain tumour at The Hospital for Sick Children (SickKids; Toronto, Canada) and 17 TDC (see Table 1 for a summary of participant characteristics). Patients were categorized into two treatment groups: CRT (n = 19) and non-CRT (n = 17). The CRT group comprises patients with malignant tumour pathology who received cranial radiation as part of standard treatment. The non-CRT group comprises patients with benign tumour pathology who did not receive cranial radiation (see Table 2 for a summary of medical variables).

Table 1.

Participant demographics.

	TDC n = 17	non-CRT n = 17	CRT n = 19	p
	Mean (SD; Range)
Age at testing (years)	12.22 (2.49)	12.74 (3.14)	14.15 (2.21)	.030
	nmale,nfemale (%)
Sex	6, 11 (64.71)	12, 5 (30.4)	11, 8 (39.1)	.111

Note. An overview of the participant sample demographics, including age at assessment, and sex distribution across three groups.TDC = Typically Developing Children; CRT = history of cranial radiation therapy. non-CRT = no history of cranial radiation therapy.patients treated with CRT (CRT group) The p-values indicate group comparisons for each demographic variable.

Table 2.

Patient characteristics.

	non-CRT n = 17	CRT n = 19	p
	Mean (SD; Range)
Age at diagnosis (years)	6.23 (3.65; 1.8–15.4)	7.43 (2.59; 3.0–12.2)	.26
Time since diagnosis (years)	6.66 (2.97; 1.4–11.3)	6.70 (2.97; 1.3–11.4)	.97
Tumour size (mm2)a	2124.29 (1208.86; 506–4800)	1883.44 (916.86; 624–3996)	.54
Treatment-related events	24.2 (13–55)	38.4 (17–70)	p < .001
	Total Number
Gross total resection (>95 % of tumour resected)	13	16	.43
Mutism following surgeryb	3	7	.18
Tumour type
Medulloblastoma (average risk)	1	11	<.001
Medulloblastoma (high risk	1	4
Ependymoma	1	4
Pilocytic astrocytoma	13	0
Cribriform neuroepithelial Tumour	1	0
Tumour location within the posterior fossa
Midline	9	16	.08
Left hemispheric	2	0
Right hemispheric	5	2
Unavailable	1	1
Hydrocephalus
No Hydrocephalus	6	1	.03
Hydrocephalus with no treatment (resolved)	5	5	.56
Hydrocephalus requiring CSF diversion (EVD, shunt, ventriculostomy)	6	12	.09
Neurological complications
Cranial nerve deficit	1	5	.12
Hearing Loss	2	8	.37
Meningitis	0	2	.27
Motor deficits (ataxia, dysmetria, dysdiadochokinesia)	6	19	.02
Visual impairment (Nystagmus; diplopia)	7	10	.36
Recurrence
0	15	14	.45
1	2	4
2	0	1
Number of surgeries
1	16	14	.24
2	1	4
3+	0	1
Radiation Type
None	17	0	<.001
Focal (5400–5940 cGy)	0	3
Reduced dose Cranial-Spinal (1800–2340 cGy) + TB Boost (3240 cGy)	0	11
Standard dose Cranial-Spinal (2340–3600 cGy) + TB or PF Boost (1800–3240 cGy)	0	5
Chemotherapy
None	12	5	.01
ACNS0332 (carboplatin, cyclophosphamide, vincristine, cisplatin, isotretinoin)	0	1
COG9961 (vincristine, lomustine, cisplatin)	0	2
COG99703 (thiotepa, carboplatin)	2	1
Head Start II (vincristine, cisplatin, cyclophosphamide, etoposide, methotrexate)	1	0
POG9631 (etoposide, cisplatin, cyclophosphamide, vincristine)	0	1
SJMB96 & SJMB03 (vincristine, cisplatin, cyclophosphamide)	0	9
Vinblastine monotherapy	2	0

Note. An overview of patient characteristics, including tumour type, tumour size, surgical details, neurological complications, and treatments. TDC = Typically Developing Children; CRT = history of cranial radiation therapy. non-CRT = no history of cranial radiation therapy.

^aTumour size was unavailable for 6 patients (3 non-CRT, 3 CRT).

^bPatients were classified as having mutism if they had diminished speech output, linguistic difficulties or dysarthria following surgery. Mutism is a transient dysfunction that had resolved in all participants by baseline assessment.

All patients were assessed at least 12 months after treatment completion. Patients were recruited through mailed invitations or routine clinic visits. TDC were recruited from the community, as siblings of patients (n = 7), or as family members of SickKids staff. Siblings were included in the TDC group due to recruitment feasibility and convenience. However, we acknowledge that siblings may experience elevated family-level stress, and sensitivity analyses excluding siblings were conducted for all analyses (see Section 2.7).

Patients with diffuse brainstem gliomas or those in palliative care were excluded. Exclusion criteria for all participants included premorbid learning disabilities, neurological disorders, traumatic brain injury, metal implants, or the need for sedation during magnetic resonance imaging (MRI). All participants were native English speakers or had completed at least two years of education in the English language. The SickKids Research Ethics Board approved this study. Before participation, an experimenter explained the study procedures, safety measures, and participant rights. Participants answered knowledge-check questions to assess their understanding of the material. Individuals with accurate comprehension provided informed self-consent; others had their parents provide written consent, and participants provided written assent. This participant sample has been previously reported by Moxon-Emre et al. (2019; 2020) and Igoshina et al. (2025).

Siblings of patients were included in the TDC group as a convenience sample to facilitate recruitment in a rare clinical population, rather than as a targeted comparison group. Their inclusion increased feasibility while also providing a comparison sample with similar access to healthcare resources and shared environmental context. This approach is commonly used in paediatric neuroimaging research to reduce variability related to socioeconomic and contextual factors that are difficult to match in unrelated community samples. At the same time, we acknowledge that siblings may experience indirect psychosocial exposure associated with having a medically ill sibling, which could introduce shared family-level stress. To address this potential confound, all primary analyses were conducted both with and without siblings in the comparison group, and sibling status was additionally included as a covariate in the PLS-PM analyses. Sensitivity analyses were used to evaluate the robustness of the findings to variations in comparison-group composition.

2.1.1. Sample size determination

The current sample was drawn from an ongoing programme of research on posterior fossa brain tumours at SickKids and includes participants who completed both an MRI scan and the study-relevant questionnaires. Sample size was constrained by the low incidence of posterior fossa brain tumours, strict MRI eligibility requirements, and the travel and time demands placed on families participating in research.

2.2. Magnetic resonance imaging

MRI data were acquired using a Siemens 3T Prisma Fit scanner with a 20-channel head and neck coil. Diffusion-weighted imaging used a single-shot spin-echo diffusion tensor sequence (30 directions, b = 1000 s/mm², TE/TR = 90/9000 ms, 70 slices, voxel size = 2 mm isotropic, interpolated to 1 × 1 × 2 mm). Additionally, a T1-weighted 3D MPRAGE GRAPPA 2 protocol was used (TE/TR = 3.83/2300 ms, 160 slices, voxel size = 1 mm isotropic).

2.3. Questionnaire measure of PTSS

The Child Post-Traumatic Stress Scale assessed the experience of a distressing event and PTSS (The Child PTSD Symptom Scale-Foa et al., 2001). Participants spontaneously identified their most distressing event and when it occurred, without being prompted about disease or treatment-related experiences. The distressing events reported were not limited to treatment-related experiences and could include other experiences (e.g., accidents, interpersonal events). Those who reported a distressing event were classified as experiencing distress and completed the PTSS assessment, which included raw subscale scores for symptom clusters (avoidance, re-experiencing, arousal). Distress was operationalized as a dichotomous variable (1 = distress reported, 0 = no distress), hereinafter referred to as a distressing event. The PTSS scale was only administered to those who reported a distressing event. Participants who did not complete the PTSS scale due to the absence of a reported distressing event were assigned a score of zero for PTSS, reflecting no symptoms according to the scale's structure.

2.4. Treatment burden

Treatment burden was quantified using the Neurological Predictor Scale (NPS), a validated composite index that captures cumulative neurological and treatment-related burden (Micklewright et al., 2008). The NPS incorporates indicators of hydrocephalus, neurological complications, tumour recurrence, number of surgical resections, radiation exposure, and chemotherapy protocol. The frequency of these components is influenced by both tumour type and treatment plan, and the NPS captures this variation in a single total score.

Surgical exposure was operationalized as the number of tumour-directed neurosurgical procedures and categorized as no surgery, biopsy or biopsies only, one tumour-directed neurosurgery, or more than one tumour-directed neurosurgery. Chemotherapy exposure was coded dichotomously based on whether chemotherapy was administered as part of treatment. Radiation exposure was categorized as no radiation, focal radiation to the tumour site, whole-brain or craniospinal radiation, or whole-brain or craniospinal radiation with an additional focal boost.

To capture post-operative neurological morbidity not fully represented by treatment exposure alone, individual post-surgical indicators were additionally coded. These included hearing and visual impairments, motor deficits, and cranial nerve deficits identified on postoperative neurological examination. Motor deficits encompassed ataxia, dysmetria, or hemiparesis, while visual impairment included diplopia or nystagmus. Each post-surgical indicator was coded as absent or present.

Hydrocephalus requiring cerebrospinal fluid diversion was coded separately as absent or present based on the requirement for surgical intervention, such as ventriculoperitoneal shunt placement or endoscopic third ventriculostomy. All clinical variables were coded such that higher values reflected greater treatment intensity or neurological impact. A descriptive summary of NPS components and post-surgical indicators, including their distribution within the sample, is provided in Table 3.

Table 3.

Distribution of Neurological Predictor Scale components and their contribution to total NPS scores.

Neurological Predictor Scale components	Score contribution	n (%)
Post surgical complicationsa
Hearing impairment
None	0	26 (72.22 %)
Present	1	10 (27.78 %)
Visual impairment
None	0	19 (52.78 %)
Present	1	17 (47.22 %)
Motor deficits
None	0	16 (44.44 %)
Present	1	20 (55.56 %)
Cranial nerve deficit
None	0	30 (83.33 %)
Present	1	6 (16.67 %)
Hydrocephalus requiring cerebrospinal fluid diversionb
None	0	7 (19.44 %)
Present	1	29 (80.56 %)
Surgery countc
None	0	0 (0 %)
Biopsy	1	3 (8.33 %)
Neurosurgery	2	27 (75.00 %)
More than one biopsy and surgery	3	6 (16.67 %)
Chemotherapy exposure
None	0	17 (42.22 %)
Received	1	19 (52.78 %)
Radiation exposured
None	0	17 (42.22 %)
Focal	1	0 (0 %)
Whole brain or craniospinal	2	0 (0 %)
Whole brain or craniospinal with boost	3	19 (52.78 %)

Note. Treatment burden was indexed using components of the Neurological Predictor Scale (NPS; Micklewright et al., 2008) and post-surgical indicators derived from medical record review.

^aPost-surgical complications include hearing impairment; visual impairment (diplopia or nystagmus); motor deficits (ataxia, dysmetria, or hemiparesis); and cranial nerve deficits (cranial nerve palsy) identified on neurological examination.

^bHydrocephalus requiring cerebrospinal fluid diversion includes ventriculoperitoneal shunt placement or endoscopic third ventriculostomy.

^cSurgery count reflects the number of tumour-directed neurosurgical procedures (none, biopsy or biopsy only, one neurosurgery, more than one neurosurgery).

^dRadiation exposure was coded as none, focal radiation, whole-brain or craniospinal radiation, or whole-brain or craniospinal radiation with an additional focal boost.

2.5. Procedure

After providing informed consent or assent, participants completed the imaging protocol, followed by a questionnaire administered by the experimenter.

2.6. Imaging data preprocessing and processing

2.6.1. Preprocessing

Diffusion-weighted images were denoised to reduce thermal and scanner noise, corrected for Gibbs ringing to remove edge artifacts, undistorted to address susceptibility and eddy-current distortions, brain masked to eliminate non-brain voxels, smoothed to enhance the signal-to-noise ratio, and Rician bias corrected to account for b1 field inhomogeneities using PyDesigner, an automated pipeline for diffusion tensor imaging preprocessing and tensor estimation (Dhiman et al.; Dhiman et al., 2022). The PyDesigner code is available at https://github.com/m-ama/pydesigner.

2.6.2. Processing

TractSeg, a WM tract segmentation approach, was used to identify fronto-limbic (CG, IFO, UF) and control (AF, OR) tracts (Wasserthal et al., 2018, 2019, 2020). This is achieved through a deep learning-based approach that directly segments WM tracts from diffusion-weighted images (Wasserthal et al., 2018, 2019, 2020). The TractSeg code and training data are openly available at https://github.com/mic-dkfz/tractseg. FA and RD metric maps were generated from preprocessed images using PyDesigner (Dhiman et al.; Dhiman et al., 2022). Subsequently, tract segmentation masks (from TractSeg) were co-registered with metric maps (from PyDesigner) to obtain average scalar FA and RD metric values for each tract using MRtrix3 (Tournier et al., 2019). All tracts were visually quality checked for anatomical accuracy before analysis.

2.7. Data analysis

Hypothesis 1

Analysis of variance (ANOVA) was conducted to evaluate group differences in PTSS among groups (TDC, non-CRT, CRT). Analyses were conducted using the aov function from the R stats package. (R Core Team, 2024)

Hypothesis 2

Two repeated measures multivariate analyses of variance (rmMANOVA) were conducted to assess group differences in WM microstructure (FA, RD) across fronto-limbic (CG, IFO, UF) and control (OR, AF) tracts among groups (TDC, non-CRT, CRT). Analyses were conducted using the manova function from the R stats package. (R Core Team, 2024)

As a follow-up to ANOVA/MANOVA, planned pairwise comparisons were conducted where indicated, with the Holm correction applied to control the family-wise error rate. Analyses for Hypotheses 1 and 2 were conducted both with and without siblings included in the TDC group to assess whether their inclusion influenced the outcomes.

Hypothesis 3

PLS-PM was conducted using 5000 bootstrapped resamples. Three models were specified: a hypothesized model, a null model, and a comparison (control tract) model. In the hypothesized model, we examined the effect of treatment burden on PTSS, mediated by fronto-limbic WM and experience of a distressing event (yes, no). The Child PTSD Symptom Scale measures (i.e., having endorsed a distressing event or not and PTSS) were included separately in the model. Treatment burden, quantified by the NPS, served as the primary exogenous variable, with age at testing and sibling status included as covariates. Because PLS-PM requires complete data for all indicators, mean imputation was applied to six missing WM values (three CRT, three non-CRT) before model estimation. This approach was chosen to preserve sample size, in line with best-practice recommendations for small samples with minimal missing data, without invoking predictive aims. (Sarstedt et al., 2022; Enders, 2012) Because the study is cross-sectional and the sample is modest, the PLS-PM analyses are interpreted as exploratory and hypothesis-generating, and paths are treated as statistical associations rather than predictive or causal effects. To assess the added contribution of WM, we compared a null model, which included only treatment burden and the experience of a distressing event as correlates of PTSS, to the hypothesized model. This comparison tested whether excluding fronto-limbic WM resulted in similar or lower explained variance in PTSS. To assess the specificity of fronto-limbic WM explained variance in PTSS, the control tract model replaced fronto-limbic WM with WM metrics from the AF and OR.

Model Fit and Evaluation. Model fit was assessed across all models using variance inflation factor (VIF), composite reliability, Cronbach's alpha, average variance extracted (AVE), and heterotrait-monotrait (HTMT) ratios (Hair et al., 2021, 2022; Becker et al., 2018). Variables with VIF values > 5 were excluded to address multicollinearity. Internal consistency was confirmed when composite reliability, Cronbach's alpha, and rho A exceeded .70. Convergent validity was supported by AVE >.50, and discriminant validity was confirmed with HTMT <.90. Model robustness was evaluated using the coefficient of determination (R²) and effect size (f²). R² indicates the proportion of variance in a dependent variable that is explained by the independent variables in a model and was interpreted as low (<.20), moderate (.20–.50), or strong (>.50). R² values are reported for all endogenous variables, including binary outcomes such as the experience of a distressing event (yes, no), consistent with standard PLS-PM estimation procedures. For binary variables (i.e., the experience of a distressing event), these represent pseudo-R (Oyefiade et al., 2021) values. Effect sizes followed Cohen's (1988) guidelines: small (f² = .02), medium (f² = .15), and large (f² = .35) (Cohen, 1992). Path coefficients (β) were considered significant if their 95 % bootstrap confidence intervals (CI₉₅) excluded zero. Construct accuracy was confirmed through indicator loadings (ranging from .70 to .95), ensuring that latent variables were adequately represented by their observed measures. Cross-loadings were examined to ensure that indicators loaded more strongly on their intended construct than on others (Hair et al., 2021, 2022).

Model Comparison and Out of Sample Evaluation. To evaluate the performance of the hypothesized model, we compared it with that of the null and control tract models (Becker et al., 2018; Shmueli et al., 2016). Given the modest sample size, this evaluation is treated as exploratory and descriptive rather than confirmatory. We applied a 5-fold cross-validation, given the sample size limitations, using the R function PLSpredict to estimate out-of-sample performance (Q²). Performance errors, including Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), were compared to those from a linear regression benchmark model. A model was considered to show better out-of-sample performance than the benchmark if the average Q² was positive, indicating that the model's out-of-sample estimates outperformed a benchmark estimation based on the mean of the observed values (Hair et al., 2021, 2022; Schuberth et al., 2023).

Assessing the Unique Role of Fronto-limbic WM. To evaluate the specific contribution of fronto-limbic WM to PTSS, we compared R² and f² values from the hypothesized model and the control tract WM model. Greater R² and larger f² values in the hypothesized model would indicate stronger explanatory power of fronto-limbic WM relative to control tracts. Additionally, we assessed the significance of β between WM and PTSS in both models. A significant β in the hypothesized model, but not in the control tract model, would further support the unique statistical association of fronto-limbic WM with PTSS (Hair et al., 2021, 2022; Becker et al., 2018).

Therefore, the best model for the data should satisfy all the fit criteria outlined above and demonstrate a strong R², meaningful f², low RMSE and MAE, and a Q² that indicates out-of-sample performance relative to the benchmark in this sample.

All path analyses were conducted using the R SEMINR and cSEM packages.

All analyses were conducted in R version 4.4.2 (Koutná et al., 2022).

3. Results

Participant demographics and patient characteristics. Participant demographics did not differ between groups (TDC, non-CRT, CRT), except for age at testing (see Table 1). To account for this, age at testing was included as a covariate in the analyses. Participants treated with CRT experienced more hydrocephalus than those treated without CRT (p = .03; see Table 2).

Hypothesis 1. An ANOVA comparing PTSS across participant groups was not significant, F(1,51) = 1.707, p = .197, indicating no differences in PTSS between groups. No participants reported clinical levels of PTSS that would signal a need for a post-traumatic stress disorder assessment (all scores below <30; TDC: mean (M) = 8.33, standard deviation (SD) = 9.40, range = 0–26; non-CRT: M = 5.32, SD = 7.90, range = 0–24; CRT: M = 7.32, SD = 7.19, range = 0–20, one outlier with a maximum PTSS score of 51 was removed).

The results remained consistent when siblings of patients were excluded from the TDC group, with no significant group differences in PTSS, F(1,44) = 2.407, p = .128. Although mean PTSS scores did not differ significantly across groups, the distribution of scores provides important context. Most participants reported none to minimal PTSS (TDC: N = 12; Surgery: N = 15; CRT: N = 11), and a subset in each group reported moderate to severe symptoms (TDC: N = 6; Surgery: N = 4; CRT: N = 5). This suggests that, although overall group means were low, individual variability in symptom severity was present across groups. Notably, seven participants in the TDC group were siblings of children treated for a brain tumour. Of these, four reported none to minimal symptoms, while three reported moderate to severe symptoms. When siblings of patients were considered separately from the TDC group, their average PTSS score (range = 0–25; M = 8.14) was comparable to that of the TDC group.

Notably, all PTSS scores fell below established clinical thresholds, indicating that observed variability reflects subclinical symptom expression rather than clinical PTSS. Accordingly, subsequent analyses focus on individual differences in symptom severity rather than diagnostic classification.

Hypothesis 2. WM microstructure (FA, RD) was evaluated across fronto-limbic tracts (CG, IFO, UF) and control tracts (AF, OR) between groups. Significant multivariate effects were observed for group, V = 0.112, F(2, 4) = 14.783, p < .001, and tract, V = 0.879, F(4, 8) = 97.719, p < .001. Significant two-way interactions were found for group × tract, V = 0.120, F(8, 16) = 3.964, p < .001, and tract × side, V = 0.197, F(4, 8) = 13.631, p < .001.

A follow-up simple effects analysis showed that the CRT and non-CRT groups exhibited greater RD in the CG and IFO compared to TDC (p = .020 and < .001, respectively). FA in the CG was lower in CRT and non-CRT groups than in the TDC (p = .027 and p = .048, respectively). For control tracts, greater RD in the OR was present in the CRT group compared to non-CRT and TDC groups (p = .018 and p < .001, respectively). In the OR, the CRT participants exhibited lower RD than the non-CRT (p = .016), and the non-CRT group showed greater RD compared to the TDC group (p = .018). In the AF, non-CRT participants exhibited higher RD than TDC (p = .036). No group differences were found for the UF in any WM metrics or for FA in the IFO (see Fig. 1). As a result, UF RD, UF FA, and IFO FA were excluded from subsequent PLS-PM analyses.

Fig. 1.

Panels depict group differences in white matter microstructure measures of radial diffusivity (RD) and fractional anisotropy (FA) across specific tracts. A: RD values for the inferior fronto-occipital fasciculus (IFO), cingulum bundle (CG), uncinate fasciculus (UF), arcuate fasciculus (AF), optic radiations (OR). Significant differences are marked with p-values between groups. B: FA values for the same tracts as in (A), showing significant between-group differences. Error bars represent standard errors of the mean. Asterisk above bars indicate statistically significant differences between groups, as indicated by p-values (’∗‘: p ≤ .05; ‘∗∗‘: p ≤ .01; ‘∗∗∗‘: p ≤ .001).

When siblings of patients were excluded from the TDC group, WM microstructure (FA, RD) was again evaluated across fronto‐limbic tracts (CG, IFO, UF) and control tracts (AF, OR) between groups. Significant multivariate effects were observed for group, V = 0.069, F(4, 714) = 6.400, p < .001, and tract, V = 0.795, F(8, 714) = 58.900, p < .001. Significant two‐way interactions were found for group × tract, V = 0.099, F(16, 714) = 2.325, p = .002, and tract × side, V = 0.172, F(8, 714) = 8.420, p < .001.

A follow-up simple effects analysis, with siblings excluded from the TDC group, showed a pattern consistent with the primary analysis. The CRT and non-CRT groups exhibited greater RD in the CG and IFO compared to the TDC group (all ps < .05). FA in the CG was lower in the CRT and non-CRT groups than in the TDC (all ps < .05). For control tracts, greater RD in the OR was present in the CRT group relative to both the non-CRT and TDC groups (all ps < .05). Within the OR, CRT participants differed from non-CRT participants, and the non-CRT group showed greater RD compared to the TDC group (all ps < .05). In the AF, non-CRT participants exhibited higher RD than the TDC group (p < .05). No group differences were observed for the UF in any WM metric or for FA in the IFO.

Hypothesis 3. Although preliminary analyses indicated no significant group differences in PTSS across categorical participant groups (CRT, non-CRT, TDC), PTSS was retained in the PLS-PM based on prior literature indicating that group-based comparisons may fail to capture the full range of individual variability in PTSS outcomes. Given that treatment burden is best conceptualized along a continuum as described in the introduction, we anticipated that modelling treatment burden as a continuous variable using the NPS would yield a more sensitive and ecologically valid estimation of its relationship with PTSS.

During model construction, we initially included all tracts showing significant group differences, namely FA and RD in the CG and IFO, as well as RD in the AF. However, subsequent assessments of construct reliability and validity indicated that FA contributed minimally to the WM construct across tracts. Similarly, RD in the AF demonstrated low indicator loadings and did not meet reliability thresholds. As a result, both FA (across all tracts) and RD in the AF were excluded from all models. Following the removal of these indicators, construct reliability and validity were adequate across the hypothesized, null, and control tract models.

Composite reliability exceeded 0.7 for all constructs, and AVE values were above 0.50, supporting strong internal consistency and convergent validity. Discriminant validity was confirmed via HTMT ratios and cross-loadings (Sarstedt et al., 2021; Hair et al., 2021, 2022). The hypothesized model demonstrated strong out-of-sample performance, yielding lower or comparable RMSE and MAE values, as well as a higher Q² value for PTSS, compared to a linear regression baseline model. In contrast, the null model and control tract model exhibited lower out-of-sample performance relative to the benchmark, as indicated by their negative average Q². These results suggest that the hypothesized model fits the sample data well and shows better out-of-sample performance than the null model within this dataset, as evidenced by the highest R² for PTSS (see Table 4 for model fit comparisons). (Sarstedt et al., 2021; Hair et al., 2021, 2022) For a complete report of model fit statistics and comparisons, see the Supplementary Materials Tables S1 through S9.

Table 4.

Model comparison.

Model	PTSS R2	PTSS f2	Q2() Average
A: Hypothesized	.438	NPS = .011 WM = .114 DISTRESSING EVENT = .431	0.067
B: Null	.374	NPS = .000 DISTRESSING EVENT = .428	−0.086
C: Comparison	.407	NPS = .008 WM = .056 DISTRESSING EVENT = .373	−0.109

Note. A: Hypothesized model in which treatment burden (Neurological Predictor Scale; NPS) is associated with post-traumatic stress symptoms (PTSS) both directly and indirectly via two mediators: (1) the experience of a distressing event and (2) fronto-limbic white matter (WM) microstructure. WM is indexed by radial diffusivity (RD) values from the left and right cingulum bundle and inferior fronto-occipital fasciculus. PTSS indicators include avoidance, re-experiencing, and arousal. Covariates include whether a typically developing control was a sibling of a patient (sibling status) and age at testing. B: Comparison model excluding WM tests whether treatment burden and distress alone are associated with PTSS as effectively as when WM is included. C: Anatomical control model testing whether the effect of WM on PTSS is specific to fronto-limbic tracts. WM is represented by RD values from control tracts, the left and right optic radiation. The best-fitting model was identified based on a combination of out-of-sample relevance (Q²) for the final endogenous variable, PTSS. Superior models demonstrated higher explained variance (R²), higher effect size (f²), and positive Q². Model comparisons were conducted according to PLS-PM guidelines outlined by Hair et al. (2017, 2022).

The hypothesized PLS-PM model (Fig. 2A), incorporating fronto-limbic WM (RD in the CG and IFO) as a mediator between treatment burden and the experience of a distressing event to PTSS, explained 43.8 % of the variance in PTSS (R² = .438), 24 % in WM microstructure (R² = .240), and 15.2 % in the likelihood of experiencing a distressing event (R² = .152). Higher NPS scores, reflecting greater treatment burden, were positively associated with the likelihood of reporting distress (β = 0.372, CI₉₅[0.094; 0.658]). In turn, and as expected, endorsing a distressing event had a strong positive effect on PTSS (β = 0.614, CI₉₅[0.414; 0.841]). WM microstructure was also significantly associated with PTSS (β = 0.337, CI₉₅[0.037; 0.669]); higher RD in the CG and IFO was associated with more severe PTSS. NPS scores were positively associated with WM microstructure (β = 0.328, CI₉₅[0.023; 0.6015]), indicating more WM microstructural change as treatment burden increased. Experiencing a distressing event was not significantly associated with WM microstructure (β = 0.089, CI₉₅[−0.185; 0.370]), indicating no effect of such an experience on brain structure. The indirect effect of NPS on PTSS through experiencing a distressing event and WM microstructure was significant (β = 0.350, CI₉₅[0.123; 0.652]), suggesting that treatment burden is associated with PTSS not only through the expected psychological indicator, experiencing a distressing event, but also through a neurobiological indicator, via increased RD in fronto-limbic tracts. Age at testing was negatively associated with WM microstructure (β = −0.328, CI₉₅₌[−0.568; −0.120]), with older participants showing reduced RD, consistent with the expected neurodevelopmental trajectory (Lebel and Deoni, 2017; Lebel et al., 2012, 2019). Being a sibling of a brain tumour patient was associated with endorsing a distressing event (β = 0.319, CI₉₅[0.070; 0.547]). For a detailed discussion of this finding, see Igoshina et al. (2025), and for a complete report of these results, see Supplementary Materials Table S8a.

Fig. 2.

Partial least squares path modelling of hypothesized indicators to post-traumatic stress symptoms (PTSS). A: Hypothesized model in which treatment burden (Neurological Predictor Scale; NPS) is associated with PTSS both directly and indirectly via two mediators: (1) the experience of a distressing event and (2) fronto-limbic white matter (WM) microstructure. WM is indexed by radial diffusivity (RD) values from the left and right cingulum bundle (CG) and inferior fronto-occipital fasciculus (IFO). PTSS indicators include avoidance (Av), re-experiencing (Re), and arousal (Ar). Covariates include whether a typically developing control was a sibling of a patient (sibling status) and age at testing. B: Comparison model, excluding WM, testing whether treatment burden and distress alone are associated with PTSS as effectively as when WM is included. C: Anatomical control model testing whether the effect of WM on PTSS is specific to fronto-limbic tracts. WM is represented by RD values from control tracts, the left and right optic radiation (OR). The overall model structure and covariates remain the same. For all models, standardized path coefficients (β) are presented; significant paths are denoted by ‘∗‘, indicating p < .05. R² values indicate the variance explained for endogenous variables. Lambda (λ) values represent factor loadings for observed indicators.

The model substituting fronto-limbic tracts with control tracts (RD in the OR) explained 40.7 % of the variance in PTSS (R² = .407), 21.8 % in WM microstructure (R² = .218), and 15.2 % in the likelihood of experiencing a distressing event (R² = .152). Age was not significantly associated with WM microstructure in this model, suggesting no age-related differences in RD. Similar to the hypothesized model discussed above, NPS was positively associated with WM microstructure (β = 0.426, CI₉₅[0.157; 0.660]), indicating greater WM microstructural changes as treatment burden increased. Also consistent with the findings above, the experience of a distressing event was not significantly associated with WM (β = 0.127, CI₉₅[−0.102; 0.341]). However, unlike the hypothesized model, control tract WM microstructure was not significantly associated with PTSS (β = 0.233, CI₉₅[−0.224; 0.604]), providing no evidence of an effect of control tract WM on PTSS. For a complete report of results, see Supplementary Materials Table S8c.

4. Discussion

This study evaluated whether WM microstructure provides additional explanatory value beyond experiencing a distressing event in understanding how treatment burden is associated with PTSS in children and adolescents treated for a brain tumour. As we hypothesized, patients exhibited greater RD in two fronto-limbic tracts, the CG and IFO, as well as in control tracts, AF and OR, compared to TDC. These findings align with prior research demonstrating injury to WM microstructure following brain tumour treatment. However, differences between the CRT and non-CRT groups were minimal, suggesting that non-CRT treatments also contributed to WM damage in this cohort (Oyefiade et al., 2021; Ailion et al., 2017). This pattern emphasizes that WM alterations are not solely attributable to CRT exposure but may also represent broader treatment-related effects that span various levels of treatment intensity. The absence of group differences in the UF may reflect tract-specific vulnerability, as the prolonged developmental trajectory of the UF could make it less susceptible to early injury (Pancaldi et al., 2023; Lebel and Deoni, 2017; Lebel et al., 2012, 2019). Importantly, we demonstrate that treatment-related WM alterations are not fully explained by CRT exposure alone. Instead, the observed WM injury may vary at the individual level, potentially reflecting differences in cumulative treatment burden not captured by categorical groupings. Together, these findings support the rationale for examining treatment burden both categorically (CRT vs non-CRT) and dimensionally using a cumulative index. Building on this foundation, we move beyond traditional group-based comparisons by combining a dimensional measure of treatment burden with objective neurobiological data to identify early structural indicators of emotional vulnerability, rather than tumour-specific biological effects, offering an approach to understanding PTSS following paediatric brain tumour treatment.

Contrary to our hypothesis, PTSS did not differ significantly across the CRT, non-CRT, and TDC groups. Notably, all PTSS scores in this sample were subclinical, and our findings therefore reflect variability within the non-clinical range rather than clinical PTSS per se. Given that the severity of symptoms in this sample predominantly remained within the subclinical range, the current findings primarily reflect dimensional variation in PTSS rather than the existence or absence of a clinical disorder. This distinction is vital for interpreting both null group differences and observed individual-level associations. Consequently, the present results indicate individual differences in symptom severity rather than diagnostic status. This may reflect emotional recovery or resilience in many individuals, despite their exposure to distressing medical experiences (Phipps et al., 2006, 2009, 2014; Howard Sharp et al., 2015; Barakat et al., 2006; Koutná et al., 2022). Alternatively, the lack of group differences may reflect individual variability in PTSS within each group, indicating that group-level averages may not accurately represent the meaningful symptoms reported by a subset of participants. Accordingly, the null findings should be interpreted with caution. Indeed, closer examination of symptom distributions reveals that a subset of participants within each group reported moderate to severe PTSS. This within-group heterogeneity highlights the importance of looking beyond mean severity scores to understand how distressing experiences and treatment burden interact to influence PTSS outcomes. Moreover, it emphasizes the importance of considering contextual factors, such as indirect exposure to treatment-related stress experienced by siblings, when comparing PTSS outcomes between TDC and patients. Although siblings offer a sociodemographic comparison group, they may differ from unrelated controls due to shared familial stress linked to paediatric illness. Importantly, sensitivity analyses excluding siblings produced similar distributions of PTSS and did not change the interpretation of the group-level results, demonstrating that the findings were resilient to the choice of comparison group composition. Though further research is needed, these findings align with prior literature indicating that diagnosis and treatment type alone may not adequately capture the impact that paediatric brain tumour treatment has on emotional functioning (Bruce et al., 2011). as exploratory and descriptive, rather than as evidence of stable risk markers or predictive pathways, given the exploratory nature of the analyses.

The strong performance of our hypothesized PLS-PM model supports a multidimensional framework in which treatment burden, objective neurobiological injury, and a subjective distressing experience are jointly associated with PTSS in children and adolescents treated for a brain tumour (Marusak et al., 2019; De Bellis, 2001; Marsac et al., 2014). Considering the range of subclinical symptoms and the cross-sectional study design, the model is meant to identify association patterns rather than to delineate risk pathways, mechanistic processes, or predictive trajectories. Notably, this framework emphasizes associative patterns rather than mechanistic or causal relationships. PTSS severity appears to be associated with two distinct yet interrelated indicators: an expected psychological one, through having endorsed the experienced a distressing event, and an additional neurobiological one, through alterations to WM microstructure in fronto-limbic tracts. As expected, the endorsement of the experience of a distressing event mediated the relationship between treatment burden and PTSS, consistent with prior findings linking CRT to an increased likelihood of experiencing a distressing event and identifying treatment as the source of that distress (Igoshina et al., 2025). Higher treatment burden may increase the risk of perceiving treatment as threatening or overwhelming, which can amplify emotional salience and, in turn, elevate vulnerability to PTSS (Johnson-Laird et al., 2006; Ehring and Quack, 2010; Gross et al., 2007). These findings underscore the importance of examining objective treatment exposure and subjective appraisal together, rather than in isolation.

Our findings newly demonstrate, in support of our hypothesis, that fronto-limbic WM microstructure mediated the relationship between treatment burden and PTSS in our exploratory PLS-PM model. Specifically, increased RD in fronto-limbic tracts, which are associated with emotional processing and regulation (Ahmed et al., 2015), was linked to more severe PTSS. Notably, this neurobiological indicator emerged independently of the expected psychological indicator, suggesting that treatment-related alterations in WM microstructure may contribute to PTSS vulnerability even when a distressing event is not explicitly identified. The lack of association between PTSS and control tract WM further reinforces the regional specificity of this effect and aligns with existing research implicating fronto-limbic circuits in emotional dysregulation and stress-related psychopathology (Wier et al., 2019). Our findings indicate that specific disruptions in fronto-limbic WM microstructure are linked to PTSS, consistent with research on trauma-exposed TDC, suggesting that injury to these tracts may be associated with difficulties in emotion regulation that increase vulnerability to PTSS (Marusak et al., 2018). The predominance of RD effects over FA may indicate RD's higher sensitivity to myelin disruption, which is particularly important in developing WM tracts vulnerable to injury related to treatment (Fields, 2008; Daniels et al., 2013; Winklewski et al., 2018). Previous null findings investigating the relationship between fronto-limbic WM and PTSS in children and adolescents treated for a brain tumour may have resulted from methodological limitations, such as dichotomized symptom scores or exclusive reliance on caregiver reports; however, additional research is needed to verify this interpretation (Powell et al., 2021). Together, these results underscore the value of a multidimensional approach to evaluating PTSS severity, incorporating both subjective interpretations of distressing experiences and objective neurobiological measures.

Although we expected to observe an association between having experienced a distressing event and WM (Daniels et al., 2013; Siehl et al., 2018), this was not supported in our sample. This suggests that treatment-related neurobiological injury may play a more prominent role in shaping WM development than distressing experiences in the paediatric brain tumour population, or that distressing experience-related changes to WM microstructure are too subtle and difficult to detect cross-sectionally, especially given that a high proportion of TDC also reported experiencing a distressing event (see Igoshina et al., 2025). Moreover, given that WM continues to mature throughout childhood and adolescence (Lebel and Deoni, 2017; Lebel et al., 2012, 2019), distress-related effects may vary across developmental stages, thereby shaping vulnerability to PTSS. Importantly, although age at testing was included as a covariate in all analyses, this statistical adjustment does not fully account for the complexity of WM maturation across childhood and adolescence. WM development follows a prolonged, non-linear trajectory that varies across tracts and individuals, with ongoing myelination, fibre reorganization, and pruning continuing into early adulthood. As a result, diffusion metrics such as FA and RD are inherently developmentally sensitive, and variation in these measures may reflect differences in maturational timing rather than fixed or static structural alteration. In the present cross-sectional sample, observed differences in fronto-limbic WM microstructure should therefore be interpreted as deviations from expected developmental patterns at a single time point, rather than as indicators of persistent injury or altered developmental course. Treatment exposure may further interact with age-dependent maturation processes, particularly in tracts with extended developmental windows, such as fronto-limbic pathways. It remains unclear whether the observed associations reflect delayed maturation, altered developmental trajectories, or transient disruption that may normalize over time. Because these possibilities cannot be disentangled in a cross-sectional design, caution is warranted when interpreting diffusion findings as enduring neurobiological differences. Longitudinal follow-up will be necessary to determine how treatment-related factors intersect with normative WM development to shape PTSS vulnerability over time. Most available evidence linking stress exposure to WM differences in typically developing youth is cross-sectional, and the directionality of these associations remains unclear. Tracking these developmental changes over time is crucial for distinguishing transient disruptions from lasting structural alterations associated with having experienced a distressing event.

As part of supplementary analyses, we examined whether tumour location or intracranial volume (ICV) contributed to variability in PTSS or fronto-limbic WM microstructure (see Supplementary materials for a complete report of results). Tumour location (midline, hemispheric, peduncular) was not significantly associated with PTSS or WM; however, the sample was predominantly midline tumours, limiting anatomical variability. Accordingly, these analyses were not intended to test tumour-specific biological effects. Although cerebellar–limbic indicators are thought to be implicated in affective regulation (Schmahmann and Sherman, 1998; Riva and Giorgi, 2000), our sample lacked sufficient anatomical diversity to evaluate this cerebellar contribution directly. Future studies with more balanced tumour distributions will be necessary to determine whether cerebellar–limbic circuits confer additional emotional vulnerability. These null findings should therefore be interpreted cautiously. Similarly, group differences in ICV were no longer significant after adjustment for age at testing, suggesting that age accounted for most variation in head size. Nonetheless, future work should continue to examine the combined influence of age, ICV, and treatment on WM maturation, as these factors may interact in ways that smaller samples cannot adequately capture. To directly assess whether the inclusion of siblings affected neurobiological outcomes, all analyses of WM were repeated with siblings excluded from the comparison group. The removal of siblings from the sample did not modify the pattern or direction of findings related to fronto-limbic or control tracts, and comparable estimates were observed across various analytical techniques. These findings imply that the observed WM modifications are indicative of treatment-related neurobiological differences rather than artifacts arising from the composition of the comparison group. Together, these supplementary findings underscore the methodological challenge of disentangling age-related maturation, anatomical factors, family-level context, and treatment effects in paediatric neuroimaging and highlight the importance of longitudinal designs for interpreting WM changes in relation to emotional outcomes.

Several limitations should be noted. First, participants who did not endorse a distressing event were assigned a score of zero on the PTSS measure, as the assessment required identification of a specific distressing event. This approach may underestimate symptom burden among individuals who experienced distressing aspects of treatment or other significant stressors but did not interpret or report them as a single identifiable event, leading to potential misclassification. We addressed this limitation through targeted sensitivity analyses, which indicated that the primary patterns of association were robust to alternative scoring and sampling approaches. This limitation highlights the importance of assessing PTSS multidimensionally by incorporating reports of a distressing event with tools that do not require identifying a single distressing event, such as interviews or questionnaires. Self-report measures of distressing event endorsement and PTSS may be influenced by age and cognitive abilities, particularly in younger participants (Pancaldi et al., 2023; Schulte and Barrera, 2010b; Mulhern et al., 2004). Future studies should incorporate additional measures of cognitive and social functioning, family context, and peer relationships to better characterize PTSS trajectories across multiple dimensions (Marusak et al., 2019; De Bellis, 2001; Marsac et al., 2014; Hocking et al., 2015; Kemala et al., 2023; Kearney et al., 2015; Steele et al., 2015; Moscato et al., 2022).

Although including siblings within the TDC group limits the extent to which our control sample reflects a true control group, we accounted for this confound by conducting all group analyses, including and excluding siblings, and by including sibling status as a covariate in our PLS-PM analyses. Future research with larger samples may benefit from distinguishing siblings from unrelated controls to estimate the effects of family-level stress exposure more precisely. These considerations underscore the broader psychosocial impact of paediatric brain tumour treatment, extending to family members, and highlight the need for family-wide support strategies (Alderfer et al., 2010; Urso and Mastroyannopoulou, 2018). At the same time, robustness to sibling exclusion should not be interpreted as evidence that familial or contextual influences are absent. Instead, these analyses demonstrate that the primary findings are not driven by sibling inclusion. Shared psychosocial exposure remains an important consideration when interpreting emotional outcomes in paediatric brain tumour populations and highlights the broader family-level impact of childhood illness.

The sample included a range of posterior fossa tumour types, which was unavoidable given the rarity of specific diagnoses in the paediatric population (Ailion et al., 2017; Griffiths-King et al., 2025). Accordingly, tumour variability is recognized as a limitation, and interpretations are confined to treatment-related patterns rather than tumour-specific effects. Our use of the NPS, therefore, aimed to capture variability in overall treatment burden across diagnostic categories rather than treating diagnosis as the primary variable. Additionally, the small sample size, although common in the paediatric brain tumour population (Griffiths-King et al., 2025), may limit the generalizability of the findings. Larger, multi-centre cohorts will be important for clarifying the robustness of these associations. Finally, the cross-sectional design of this study limits the ability to establish causal relationships. Longitudinal research is essential for tracking WM over time and evaluating how age-related maturation, treatment effects, and endorsement of distressing events jointly contribute to PTSS.

An additional limitation of the present study concerns developmental variability in WM maturation. Although age at testing was included as a covariate, this approach cannot capture non-linear, tract-specific, or individual differences in neurodevelopmental trajectories. Diffusion metrics reflect dynamic biological processes that evolve across childhood and adolescence, and age adjustment alone does not distinguish between delayed maturation, altered developmental timing, or persistent structural differences. Consequently, the observed associations between treatment burden, WM microstructure, and PTSS should be interpreted within a developmental framework, recognizing that they represent snapshots of ongoing neurodevelopment rather than stable or final outcomes. These constraints underscore the importance of longitudinal designs that track within-person change in WM across development. Such approaches are essential for determining whether treatment-related differences persist, resolve, or diverge over time, and for clarifying how developmental timing may influence emotional vulnerability following paediatric brain tumour treatment.

Our findings support two indicators associated with PTSS in survivors of a paediatric brain tumour: a psychological indicator through endorsement of a distressing experience, and a neurobiological indicator via treatment-related fronto-limbic WM alterations. Specifically, increased RD in fronto-limbic tracts was associated with more severe PTSS, independent of the psychological indicator. Consistent with prior findings linking CRT to distress, we extend this work by demonstrating that treatment burden is similarly associated with the likelihood of experiencing a distressing event. Critically, treatment burden was associated with PTSS indirectly through these indicators rather than directly, suggesting the importance of intermediate psychological and neurobiological processes. These findings suggest that vulnerability to PTSS may not always be evident through self-report alone, hihglighting the value of neurobiological indicators for identifying children and adolecents who may be vulnerable even when they do not explicitly report distressing experiences. The association between treatment burden and distressing event endorsement suggests treatment-related distress may represent a modifiable target, presenting opportunities for early psychological intervention during active treatment to support emotion regulation. Collectively, the current findings indicate that significant variability related to PTSS can be detected even in the absence of clinical-level symptoms, supporting the value of dimensional approaches in identifying early emotional vulnerability within diverse paediatric brain tumour populations, while also highlighting the necessity for longitudinal studies to examine the progression of subclinical symptoms over time. Ultimately, longitudinal research will be essential to disentangle treatment-related structural changes from developmental and distress-related effects, informing targeted, developmentally appropriate strategies to support resilience in this vulnerable population.

CRediT authorship contribution statement

Elizaveta Igoshina: Writing – review & editing, Writing – original draft, Visualization, Validation, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Iska Moxon-Emre: Writing – review & editing, Data curation. Suzanne Laughlin: Writing – review & editing. Julie Tseng: Writing – review & editing, Data curation. Donald J. Mabbott: Writing – review & editing, Supervision, Resources, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is the Supplementary data to this article:

Data availability

The data that has been used is confidential.