Personalized survival prediction in pediatric glioblastoma using a machine learning-powered web tool

Introduction

Pediatric glioblastoma (p-GBM) represents one of the most formidable challenges in neuro-oncology due to its highly aggressive biology and limited therapeutic options, which contribute to persistently poor survival outcomes (1-3). Although GBM is among the most common primary brain tumors in adults, p-GBM is relatively rare, accounting for only approximately 6% of all central nervous system tumors in children (4). To date, exposure to ionizing radiation remains the only environmental factor consistently linked to p-GBM. Inherited cancer predisposition syndromes, including Li-Fraumeni syndrome, constitutional mismatch repair deficiency (CMMRD) syndrome, neurofibromatosis type I, and Turcot syndrome, have also been associated with increased risk (5-7). Distinct from adult GBM, pediatric cases display unique molecular profiles and clinical courses (1,2,8-10). Nevertheless, population-level data on incidence patterns, prognostic determinants, and dynamic survival trajectories are limited. As a result, clinicians and families frequently face considerable uncertainty in predicting individual outcomes, complicating both treatment planning and long-term follow-up.

Traditional survival analyses provide static estimates at diagnosis, which often fail to reflect the evolving risk profile of patients who survive initial treatment periods (11-14). Conditional survival (CS) analysis addresses this limitation by recalculating survival probabilities contingent on time already survived, offering a dynamic perspective that is particularly relevant for high-risk pediatric populations (15). However, few studies have leveraged CS to quantify the shifting prognosis in children with GBM, leaving a critical knowledge gap in personalized prognostic assessment.

In this study, we harnessed the Surveillance, Epidemiology, and End Results (SEER) database to examine p-GBM from 2000 to 2021, providing a comprehensive assessment of incidence trends, survival outcomes, and prognostic determinants. We performed Kaplan-Meier and multivariable analyses to identify key clinical and treatment-related factors, and applied CS analysis to capture the dynamic evolution of prognosis over time. Leveraging least absolute shrinkage and selection operator (LASSO) regression, we constructed a CS-nomogram for individualized risk prediction and developed an accessible online tool to facilitate real-time prognostic estimation (16). By integrating population-level epidemiology, dynamic survival modeling, and interactive risk prediction, this work aimed to provide a novel framework for understanding and managing pediatric GBM in a more personalized and clinically actionable manner. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0098/rc).

Methods

Data source and study population

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Data were extracted from the SEER database, which covers approximately 28% of the U.S. population. Patients diagnosed with p-GBM between January 2000 and December 2021 were identified from the International Classification of Diseases for Oncology, Third Edition (ICD-O-3) using the histology code 9440/3, restricted to tumors located in the brain. Inclusion criteria were: (I) age ≤18 years at diagnosis; (II) histologically confirmed GBM as the first primary malignancy; and (III) availability of survival and treatment data. Patients with incomplete demographic or survival information were excluded.

Study variables

Demographic and clinical variables included age at diagnosis, sex, race, primary tumor site, tumor extension, tumor size, surgical resection, radiotherapy (RT), chemotherapy (CT), and county-level median household income. Survival time and vital status were obtained from SEER.

Incidence trends

Age-adjusted incidence rates were calculated using SEER*Stat and expressed per 100,000 person-years, standardized to the 2000 U.S. standard population. Temporal trends in incidence were analyzed with Joinpoint Regression Program. The annual percent change (APC) and 95% confidence intervals (CIs) were estimated.

Survival analysis

Overall survival (OS) was defined as the time from diagnosis to death from any cause. Kaplan-Meier methods were used to estimate survival curves, and log-rank tests compared differences among subgroups. CS probabilities were calculated as the probability of surviving an additional five years given survival for a specified number of years post-diagnosis. CS was calculated as:

CS(y|x)=S(x+y)/S(x)[1]

where x is the number of years already survived, and y is the additional survival duration of interest. Cox proportional hazards regression was conducted to identify independent prognostic factors.

Model development and internal validation

The cohort then was randomly divided into training (70%) and internal validation (30%) sets. In the training cohort, LASSO regression was applied to prevent overfitting and select the most informative predictors, with penalty parameter λ determined via 10-fold cross-validation using the minimum criteria. Selected variables were incorporated into a multivariable Cox model to construct a CS-based nomogram. CS probabilities were derived post-hoc from the fitted Cox proportional hazards model. The CS for an additional y years given x years already survived was calculated as S(x+y)/S(x). While the nomogram provided baseline survival estimates at the time of diagnosis, the CS-nomogram was developed to allow for the real-time calculation of CS probabilities, providing updated prognostic information for patients at various clinical follow-up stages.

Regarding model stability, our cohort of 597 patients includes 500 events. Given the 10 candidate predictors, the events-per-variable (EPV) ratio is 50, which far exceeds the standard requirement of 10. This high EPV, combined with LASSO regularization, ensures the model’s robustness and effectively minimizes the risk of overfitting. Model performance was assessed using several approaches. Calibration was evaluated by plotting predicted versus observed survival probabilities. Discrimination was quantified by time-dependent receiver operating characteristic (ROC) curves and the corresponding area under the curve (AUC). Clinical utility was evaluated using decision curve analysis (DCA).

Risk stratification

For each patient, a risk score was derived from the final model. Patients were stratified into risk groups according to the optimal cutoff values determined by maximally selected rank statistics. Kaplan-Meier curves were generated to compare survival between risk groups, with log-rank tests used to assess differences.

Web-based dynamic tool

To facilitate individualized clinical application, an interactive web-based calculator was developed using the final nomogram model. The tool provides real-time estimates of CS probabilities and dynamic risk stratification for pediatric GBM patients.

Statistical analysis

All statistical analyses were performed using R software, and all tests were two-sided, with a P value <0.05 considered statistically significant.

Results

Incidence trends

The age-adjusted incidence of pediatric GBM increased modestly during 2000–2021; however, the trend was not statistically significant (APC =1.52, P>0.05). As shown in Figure 1, the incidence rate increased from 0.091 per 100,000 person-years in 2000 to 0.109 per 100,000 person-years in 2021. No joinpoints were identified, indicating a consistent linear increase throughout the study period.

Figure 1 Age-adjusted incidence of p-GBM in the United States from 2000 to 2021. Joinpoint regression analysis demonstrates a modest but non-significant linear increase in incidence over two decades, highlighting the rarity and stable population-level occurrence of this malignancy. APC, annual percent change; p-GBM, pediatric glioblastoma.

Basic clinical features

Table 1 summarizes baseline characteristics of 597 p-GBM patients, stratified into training (n=417) and internal validation (n=180) sets. The majority of patients were aged 10–18 years (60.3%) and male (57.8%). Most cases were White (74.0%) with tumors located supratentorially (63.7%), and three-quarters presented with localized disease. Regarding management, 82.2% underwent surgical resection, and 75.9% received radiotherapy, while 69.3% were treated with chemotherapy. More than half of the cohort (58.0%) resided in areas with a median household income below $85,000.

Survival analysis

In CS analysis, with increasing survival time from 0 to 4 years, the 5-year CS rate of p-GBM patients showed a substantial upward shift (Figure 2). At diagnosis (0 year), the probability of surviving an additional 5 years was only 14%. For patients who had already survived 1 year, the 5-year CS improved to 27%. Among 2-year survivors, this probability further increased to 48%, while 3-year survivors achieved a 5-year CS of 71%. Notably, patients who survived 4 years demonstrated a markedly favorable outlook, with the 5-year CS reaching 88% (Figure 2). These results emphasized that prognosis became progressively more optimistic as patients surpass critical early survival milestones.

Figure 2 Five-year CS probabilities of p-GBM patients. CS markedly improves with each additional year survived post-diagnosis, illustrating the dynamic evolution of prognosis and the value of CS in providing time-adapted survival estimates. CS, conditional survival; p-GBM, pediatric glioblastoma.

Kaplan-Meier survival analysis of the entire cohort revealed substantial heterogeneity in outcomes based on clinical and treatment-related factors (Figure 3). Significant differences in survival were observed according to tumor site, extent of disease, tumor size, surgical intervention, and chemotherapy, whereas other factors did not appear to have a clear impact (Figure 3). Multivariable Cox analysis showed that infratentorial location [hazard ratio (HR) 1.867, 95% CI: 1.431–2.437, P<0.001], tumor size ≥50 mm (HR 1.333, 95% CI: 1.067–1.666, P=0.01), and regional/distant extension (HR 1.489, 95% CI: 1.199–1.848, P<0.001) predicted worse survival. Gross total resection (HR 0.486, 95% CI: 0.368–0.643, P<0.001) and chemotherapy (HR 0.529, 95% CI: 0.430–0.652, P<0.001) were associated with improved survival, whereas other factors, including age, sex, race, radiation, and income, were not significant (Figure 4).

Figure 3 Kaplan-Meier survival curves stratified by key clinical and treatment-related variables including age (A), sex (B), race (C), site (D), size (E), extension (F), surgery (G), RT (H), and CT (I). CT, chemotherapy; GTR, gross total resection; NOS, not otherwise specified; RT, radiotherapy; STR, subtotal resection.

Figure 4 Multivariable Cox proportional hazards regression analysis of overall survival. Hazard ratios and 95% confidence intervals identify infratentorial tumor location, tumor size ≥50 mm, and regional/distant disease as adverse prognostic factors, while gross total resection and chemotherapy confer survival benefits. CI, confidence interval; CT, chemotherapy; GTR, gross total resection; HR, hazard ratio; NOS, not otherwise specified; RT, radiotherapy; STR, subtotal resection.

CS prediction model

Given the observed heterogeneity in survival across clinical and treatment-related factors, we proceeded to construct a predictive model for the cohort. Rather than relying solely on the previously identified multivariable Cox model, we employed LASSO regression in the training set to enhance variable selection and minimize overfitting. Using the LASSO minimum criteria, tumor site, extent of disease, surgical intervention, and chemotherapy were identified as the most informative predictors (Figure 5A,5B). These variables were then incorporated to successfully establish a nomogram for individualized survival prediction (Figure 6). The performance of the nomogram was first assessed using calibration curves, which demonstrated good agreement between predicted and observed 1-, 3-, and 5-year survival probabilities in both the training and internal validation cohorts (Figure 7A,7B). Time-dependent ROC analysis further confirmed the model’s discriminative ability (Figure 7C,7D): in the training set, the 1-, 3-, and 5-year AUCs were 0.775 (95% CI: 0.729–0.820), 0.729 (95% CI: 0.671–0.788), and 0.674 (95% CI: 0.597–0.751), respectively; in the internal validation set, the corresponding AUCs were 0.710 (95% CI: 0.633–0.786), 0.661 (95% CI: 0.558–0.764), and 0.641 (95% CI: 0.515–0.767). Finally, DCA demonstrated notable clinical utility of the nomogram across both cohorts, supporting its potential value for individualized prognosis assessment (Figure 7E,7F).

Figure 5 Variable selection using LASSO regression. (A) LASSO coefficient profiles for all candidate predictors; (B) Cross-validation plot showing the optimal lambda (λ) that minimizes mean cross-validated error, leading to selection of the most informative variables for the CS-based nomogram. CS, conditional survival; LASSO, least absolute shrinkage and selection operator.

Figure 6 Conditional survival nomogram for individualized 1-, 3-, and 5-year survival prediction in pediatric GBM. The model integrates tumor site, extent, surgical intervention, and chemotherapy, enabling dynamic, patient-specific prognostic estimation. CS, conditional survival; GBM, glioblastoma; NOS, not otherwise specified; OS, overall survival.

Figure 7 Performance evaluation of the CS-nomogram in training and validation cohorts. (A,B) Calibration plots demonstrating agreement between predicted and observed survival probabilities. (C,D) Time-dependent ROC curves assessing discriminative ability. (E,F) Decision curve analysis highlighting clinical utility. (G,H) Kaplan-Meier curves for risk-stratified subgroups, confirming the nomogram’s capacity to distinguish high- and low-risk patients. AUC, area under the curve; CI, confidence interval; CS, conditional survival; ROC, receiver operating characteristic.

Risk stratification and online nomogram tool

To assess the practical utility of the nomogram, a risk score was generated for each patient based on the selected variables. Using the optimal cutoff derived from the training cohort, patients were classified into high- and low-risk groups. Kaplan-Meier analysis revealed that patients in the high-risk group had markedly poorer survival compared with those in the low-risk group in both the training and internal validation cohorts (P<0.001 for both, Figure 7G,7H), highlighting the strong discriminative power of the risk stratification. Furthermore, an interactive online version of the nomogram was created to enable personalized survival prediction and enhance its accessibility for clinical application (https://feyneurosurgery.shinyapps.io/pGBM/, Figure 8).

Figure 8 Web-based interactive version of the CS-nomogram. Clinicians can input patient-specific clinical data to obtain real-time, individualized conditional survival predictions, facilitating dynamic risk assessment and personalized follow-up planning. CS, conditional survival; OS, overall survival.

Discussion

This population-based analysis of p-GBM spanning two decades offered new insights into incidence trends, prognostic determinants, and the evolving nature of survival probabilities when assessed through CS. Our findings reinforced the aggressive biological behavior of p-GBM while highlighting the heterogeneity in outcomes according to tumor and treatment-related characteristics. Importantly, the application of CS analysis provided a more nuanced understanding of prognosis, which may inform both clinical decision-making and family counseling.

First, although our study identified a modest linear increase in the incidence of p-GBM during the study period, the change was not statistically significant. This finding aligned with prior epidemiological reports indicating that p-GBM remains a rare malignancy with relatively stable population-level incidence (17). The lack of significant temporal variation suggested that neither diagnostic practices nor environmental exposures had substantially shifted the burden of disease, and inherited cancer predisposition syndromes remained the most clearly established risk factors. Nevertheless, even subtle changes in incidence should prompt ongoing surveillance, as improved molecular diagnostics may refine case ascertainment in the future.

Second, our analysis underscored the dynamic prognostic landscape of p-GBM when viewed through the lens of CS. At diagnosis, long-term survival probabilities were extremely poor, with a 5-year CS of only 14%. However, for patients surpassing early survival milestones, prognosis improved substantially, with the 5-year CS reaching nearly 90% at four years post-diagnosis. This striking gradient suggested that early survivorship acted as a strong surrogate marker for favorable tumor biology and effective treatment response. These results paralleled observations in other pediatric malignancies where CS analyses had revealed progressively improving outlooks for long-term survivors (18). For p-GBM, CS provided a more realistic reflection of prognosis than static Kaplan-Meier estimates and may help mitigate the uncertainty faced by families after initial diagnosis.

Third, we identified several independent prognostic determinants that could refine risk stratification. Tumor location, disease extension, and size were consistently associated with adverse outcomes, underscoring the importance of anatomical and biological constraints in therapeutic management. Conversely, gross total resection and chemotherapy conferred significant survival benefits, consistent with the paradigm that maximal safe resection combined with adjuvant therapy remains the cornerstone of management (3,19,20). Interestingly, radiotherapy did not emerge as a significant prognostic factor in multivariable analysis, which may reflect either its near-universal use in eligible patients or its limited efficacy in altering long-term outcomes for this disease. Consistent with our findings, Liu et al., in their cohort of 1,173 patients with p-GBM, reported no survival advantage associated with radiotherapy (7). The lack of survival benefit from radiotherapy in p-GBM may partly be attributed to the considerable proportion of children younger than 3 years, in whom radiotherapy is often deferred or delivered at reduced doses owing to the heightened vulnerability of the developing central nervous system (7,21). Future studies should therefore analyze the impact of radiotherapy with age-stratified or propensity score-based approaches, conduct sensitivity analyses excluding patients under 3 years of age, and explore potential interactions with surgical extent and chemotherapy to more accurately delineate its survival effect. Additionally, demographic variables, including age, sex, race, and socioeconomic indicators, did not independently predict survival, suggesting that tumor- and treatment-related factors exert stronger influence than host-related variables in this context.

Fourth, the development and internal validation of a CS-based nomogram represents a key contribution of this study. By integrating tumor site, extent, surgical intervention, and chemotherapy, the model demonstrated robust calibration, moderate discrimination, and meaningful clinical utility across training and internal validation cohorts. Importantly, the incorporation of CS into the nomogram offered clinicians a dynamic, patient-specific prognostic tool that adjusts survival probabilities over time rather than relying on static estimates at diagnosis (11,13,14). Risk stratification based on model-derived scores further distinguished patients with divergent survival trajectories, underscoring its potential role in tailoring follow-up intensity, therapeutic strategies, and survivorship planning. The translation of this model into an interactive web-based tool enhanced accessibility, promoting its adoption in clinical practice as a decision-support system.

The clinical and prognostic landscape of what was historically termed “pediatric GBM” has been redefined by the 2021 World Health Organization (WHO) classification (22). Most tumors previously labeled as p-GBM are now recognized as distinct molecular entities under the umbrella of diffuse pediatric-type high-grade gliomas. This reclassification reflects the understanding that p-GBM is biologically distinct from adult GBM. Our finding that clinical factors like tumor location (infratentorial vs. supratentorial) and extension are significant predictors likely mirrors the underlying molecular distribution of these new entities—for instance, the poorer prognosis associated with infratentorial locations often correlates with the H3 K27-alteration. While our nomogram provides a robust clinical framework based on available SEER parameters, its results should be interpreted as a reflection of the broad “histological GBM” phenotype which remains a common clinical starting point in many centers before molecular results are finalized.

Nevertheless, several limitations warrant discussion. First, as with all SEER-based studies, the absence of molecular data limited our ability to incorporate genomic predictors such as H3K27M mutations, IDH status, or MGMT methylation, which are increasingly recognized as pivotal in pediatric glioma biology. Therefore, it should be considered a complementary epidemiological tool for risk estimation rather than a substitute for molecularly informed prognostic assessment. Future studies integrating large population data with molecular profiles will be important for developing more precise and clinically applicable prognostic models. Second, treatment details in SEER are not fully comprehensive, and information regarding chemotherapy regimens, radiotherapy dose, and molecularly targeted or experimental therapies was unavailable. Third, while the nomogram was internally validated, external validation in independent cohorts, ideally with molecular annotations, is required before broad clinical adoption. Another potential limitation of this study is the observed statistical imbalance in tumor size between the training and internal validation cohorts (P=0.02). However, this variable was excluded during the model selection phase. Consequently, the final nomogram’s performance and generalizability are derived solely from well-balanced prognostic factors, ensuring the stability of the validation metrics. Finally, although CS provides valuable prognostic recalibration, it cannot fully capture the quality of survivorship, including neurocognitive, functional, and psychosocial outcomes that remain central to pediatric oncology care.

Conclusions

In conclusion, this study provided a comprehensive population-level assessment of p-GBM, demonstrating stable incidence trends but highly variable survival outcomes shaped by tumor biology and treatment. By applying CS analysis and constructing a dynamic prognostic nomogram, we offered a novel framework for individualized risk assessment that evolves over time. These findings highlighted the importance of integrating dynamic survival modeling into clinical practice and underscored the urgent need for molecularly informed, multimodal therapeutic strategies to improve outcomes for children afflicted by this devastating disease.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0098/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0098/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0098/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Njonkou R, Jackson CM, Woodworth GF, et al. Pediatric glioblastoma: mechanisms of immune evasion and potential therapeutic opportunities. Cancer Immunol Immunother 2022;71:1813-22. [Crossref] [PubMed]
2. Das KK, Mehrotra A, Nair AP, et al. Pediatric glioblastoma: clinico-radiological profile and factors affecting the outcome. Childs Nerv Syst 2012;28:2055-62. [Crossref] [PubMed]
3. Nikitović M, Stanić D, Pekmezović T, et al. Pediatric glioblastoma: a single institution experience. Childs Nerv Syst 2016;32:97-103. [Crossref] [PubMed]
4. Korshunov A, Schrimpf D, Ryzhova M, et al. H3-/IDH-wild type pediatric glioblastoma is comprised of molecularly and prognostically distinct subtypes with associated oncogenic drivers. Acta Neuropathol 2017;134:507-16. [Crossref] [PubMed]
5. Lam S, Lin Y, Zinn P, et al. Patient and treatment factors associated with survival among pediatric glioblastoma patients: A Surveillance, Epidemiology, and End Results study. J Clin Neurosci 2018;47:285-93. [Crossref] [PubMed]
6. Tamber MS, Rutka JT. Pediatric supratentorial high-grade gliomas. Neurosurg Focus 2003;14:e1. [Crossref] [PubMed]
7. Liu M, Thakkar JP, Garcia CR, et al. National cancer database analysis of outcomes in pediatric glioblastoma. Cancer Med 2018;7:1151-9. [Crossref] [PubMed]
8. Fontebasso AM, Liu XY, Sturm D, et al. Chromatin remodeling defects in pediatric and young adult glioblastoma: a tale of a variant histone 3 tail. Brain Pathol 2013;23:210-6. [Crossref] [PubMed]
9. Uppar AM, Sugur H, Prabhuraj AR, et al. H3K27M, IDH1, and ATRX expression in pediatric GBM and their clinical and prognostic significance. Childs Nerv Syst 2019;35:1537-45. [Crossref] [PubMed]
10. Brassesco MS, Roberto GM, Delsin LE, et al. A foretaste for pediatric glioblastoma therapy: targeting the NF-kB pathway with DHMEQ. Childs Nerv Syst 2023;39:1519-28. [Crossref] [PubMed]
11. Huang H, Tunje RC, Xia J, et al. Building a dynamic web calculator for individualized conditional survival estimation in brainstem ependymoma. Sci Rep 2025;15:27703. [Crossref] [PubMed]
12. Xu L, Yang Z, Chen H, et al. Conditional survival and changing risk profile in patients with gliosarcoma. Front Med (Lausanne) 2024;11:1443157. [Crossref] [PubMed]
13. Sun C, Yang Z, Gu Z, et al. Conditional survival estimates for ependymomas reveal the dynamic nature of prognostication. Discov Oncol 2024;15:460. [Crossref] [PubMed]
14. Zheng G, Yang Z, Qian H, et al. Conditional survival of patients with primary bone lymphoma of the spine: how survival changes after initial diagnosis. Front Oncol 2024;14:1356947. [Crossref] [PubMed]
15. Zabor EC, Gonen M, Chapman PB, et al. Dynamic prognostication using conditional survival estimates. Cancer 2013;119:3589-92. [Crossref] [PubMed]
16. Balachandran VP, Gonen M, Smith JJ, et al. Nomograms in oncology: more than meets the eye. Lancet Oncol 2015;16:e173-80. [Crossref] [PubMed]
17. Ostrom QT, Cioffi G, Waite K, et al. CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States in 2014-2018. Neuro Oncol 2021;23:iii1-iii105. [Crossref] [PubMed]
18. Mertens AC, Yong J, Dietz AC, et al. Conditional survival in pediatric malignancies: analysis of data from the Childhood Cancer Survivor Study and the Surveillance, Epidemiology, and End Results Program. Cancer 2015;121:1108-17. [Crossref] [PubMed]
19. Pervez W, Bakhshi SK, Mirza FA, et al. Management of paediatric glioblastoma. J Pak Med Assoc 2021;71:385-7. [PubMed]
20. Adams H, Adams HH, Jackson C, et al. Evaluating extent of resection in pediatric glioblastoma: a multiple propensity score-adjusted population-based analysis. Childs Nerv Syst 2016;32:493-503. [Crossref] [PubMed]
21. DA W. Prognostic Factors and Survival Prediction of Pediatric Glioblastomas: A Population-Based Study. Turk Neurosurg 2021;31:873-9. [PubMed]
22. Louis DN, Perry A, Wesseling P, et al. The 2021 WHO Classification of Tumors of the Central Nervous System: a summary. Neuro-Oncology 2021;23:1231-51. [Crossref] [PubMed]