Analysis of survival in pediatric high-grade brainstem gliomas: A population-based study.

BACKGROUND: The purpose of this study was to use the National Cancer Institutes' Surveillance, Epidemiology, and End Results (SEER) database to perform a large-scale analysis of brainstem anaplastic astrocytoma (AA) and glioblastoma multiforme (GBM). Use of the SEER database gave us a larger sample size of this rare tumor type, allowing for the analysis of the relationship between prognostic factors and survival.
MATERIALS AND METHODS: We selected pediatric patients (<18 years old) from the SEER database with histologically confirmed diagnoses of primary high-grade gliomas (World Health Organization Grade III/IV) of the brainstem. In univariate and multivariate analysis, we analyzed the relationship between demographic (age, gender, race, diagnosis date), histologic (AA, GBM), and treatment (surgery, radiation) factors on survival.
RESULTS: In our cohort of 124 patients, those with AA had a median survival of 13 months and those with GBM 9 months. Higher-grade tumors were associated with statistically significantly increased mortality (hazard ratio [HR]: 1.74, confidence intervals [CIs]: 1.17-2.60). Surgical intervention was associated with a significantly lower mortality, either alone (HR: 0.14, CI: 0.04-0.5) or in combination with radiation (HR: 0.35, CI: 0.15-0.82). Radiation therapy alone was significantly associated with decreased mortality within the first 9 months after diagnosis but not with overall mortality. No demographic characteristics were significantly associated with mortality.
CONCLUSIONS: Outcome remains poor in the pediatric high-grade brainstem glioma population. Survival is correlated with lower-grade tumor histology, radiation therapy only in the first 9 months after diagnosis, and surgical resection.

Introduction

Brainstem tumors compose 10-20% of pediatric central nervous system tumors. In the US, approximately 150-300 pediatric cases are diagnosed annually.[1234] Brainstem tumors are classified as diffuse, focal, exophytic, and cervicomedullary.[567] About 60-85% of pediatric brainstem tumors are diffuse pontine gliomas.[3489] Usually infiltrative and high-grade on histological examination, they carry a dismal median survival of <1 year.[1011] Focal, exophytic and cervicomedullary tumors, usually well-defined, indolent in course, with low-grade glioma growth patterns, are more amenable to surgical resection and carry much better prognoses, with median survival >5 years.[39] A standard treatment for diffuse brainstem gliomas is external beam fractionated radiation. Notwithstanding numerous clinical investigations, chemotherapy's efficacy, either alone or combined with radiotherapy, remains to be proven.[12] Additionally, resection is often considered infeasible in diffuse infiltrative lesions.[13] Since no treatment has yet been shown to prolong overall survival in brainstem gliomas and given the dismal prognosis for this neoplasm, having a better understanding of the prognostic factors associated with a favorable outcome is vital.[4141516]

Due to high-grade pediatric brainstem tumors’ relative rarity, most studies have been small clinical trials or single-institution studies.[316171819] Here, we used the National Cancer Institutes’ (NCI) Surveillance, Epidemiology, and End Results (SEER) database to perform a large-scale analysis of brainstem anaplastic astrocytoma (AA) and glioblastoma multiforme (GBM).

Materials and Methods

Study population

Data for this analysis were obtained from the SEER program (1973-2008) of the NCI. This cancer registry includes data from 17 geographic areas in the United States and represents approximately 26% of the US population (http://seer.cancer.gov/data/ciatation/html). For the purposes of the present analysis, cases from Louisiana were not utilized due to noncontinuous reporting of data from the impact of Hurricanes Katrina and Rita in the Gulf Coast region. Site and histology codes of the International Classification of Disease for Oncology (ICD-O-3) were used to identify cases. Subjects classified as having histopathologically confirmed AA (ICD-O-3:9401) or GBM (ICD-O-3:9440, 9441, 9442) located in the brainstem (primary site code C71.7) were included in this study. Cases diagnosed by autopsy or death certificate were excluded. We included only individuals for which AA or GBM was their only or first primary tumor. Subsequent tumors or recurrences were not considered for this analysis. Patients with a survival time of 0 month (n = 6) were also excluded. SEER data is publicly available; therefore, exempt review status was received from our Institutional Review Board.

Surgical procedure data

Trained coders extracted surgical procedure codes from the SEER database according to established guidelines in the medical records review to determine the extent of resection. The variable containing surgery codes for the cases diagnosed between 1998 and 2008 was named “RX Summ-Surg Prim Site (1998+)” and was based on the American College of Surgeons Commission on Cancer's Facility Oncology Registry Data Standards.[20] Field codes for cases diagnosed between 1983 and 1997 were included in the “Site Specific Surgery (1983-1997)” variable. Any case diagnosed before 1983 was coded using a nonspecific scheme (yes/no/unknown). In order to create a new variable for analysis that included cases from all time periods, we recoded the surgical procedures for all years into six previously established categories.[21] These categories included: No surgery (pre-1998 codes 00, 01, 03, 04, 07; for 1998+, code 00); biopsy (Bx, pre-1998 code 02; for 1998+, code 20); partial resection (pre-1998 codes 20, 40; for 1998+, code 40); gross total resection (pre-1998 codes 30, 50, 60; for 1998, code 55); surgery-not otherwise specified (NOS) (pre-1998 codes 10, 90; for 1998+, codes 10, 90); and surgery status unknown (pre-1998 codes 05, 06, 09, 80; for 1998+, code 99).

Covariates

Categorical age at diagnosis (0-6, 7-12, 13-18 years), sex, race (white, black, American Indian, Asian/Pacific Islander and other/unknown), Hispanic ethnicity (yes/no/unknown), surgery (no surgery/biopsy/partial resection/gross total resection/surgery - NOS/unknown), radiation therapy (yes/no/unknown), and the combination of surgery and radiation (surgery only, radiation only, both surgery and radiation, neither surgery nor radiation) were evaluated in this analysis. For the binary surgery (surgery/no surgery) variable, no surgery and biopsy were collapsed into “no surgery”; partial resection, gross resection, and surgery NOS were collapsed into “surgery”; and unknowns retained their own category. Detailed patient- and treatment-related factors, such as chemotherapeutic regimens, radiotherapy technique, and comorbid conditions, were unaccounted for in the SEER database and were not evaluated in the present analysis.

Statistical analysis

Descriptive analyses were conducted to evaluate the distribution of patient- and tumor-related characteristics. The Kaplan–Meier method was used to estimate overall survival, as well as 10-, 5-, 2-, and 1-year survival. The association of variables with survival was quantified using hazard ratios (HRs) estimated from Cox proportional hazards models. Univariate Cox proportional hazards models were used to calculate HRs and 95% confidence intervals (CIs) in order to estimate the influence of each variable of interest on the hazard function. Some multivariate Cox proportional hazards models were conducted to estimate the hazard function utilizing variables identified for analysis based on a priori assumptions of associations with survival. Where specified, stepwise multivariate Cox proportional hazards models were also constructed using backward stepwise selection; predictor variables were dropped when their P value in the multivariate analysis was higher than 0.1. Log-rank statistics were used to determine differences between survival curves. A P < 0.05 was considered as statistically significant. Statistical analyses were performed using STATA version 12 (Stata Corporation, College Station, TX, USA).

Results

The SEER database included 124 patients who met the selection criteria [Table 1]. An equal number of males and females were included in the study, with a mean age of diagnosis of 7.8 years. A majority of the patients (84%) were under the age of 13 years. Most (74%) were white and non-Hispanic (87%). All patients had histologically confirmed high-grade gliomas: About 48% were AA and 52% were GBM. SEER did not contain data about the focality or specific location of these tumors within the brainstem. A majority (83%) of the patients received radiation therapy. About half (49%) of the patients received either no surgery or a biopsy, over a third (35%) had some surgical resection, and the rest (16%) had an unknown surgery status [Table 1].

Table 1

Demographics

Univariate analyses using a Cox proportional hazards model [Table 2] found the following factors to be significant (P < 0.05): Tumor histology with GBM was associated with higher mortality compared to AA (HR: 1.77, CI: 1.19-2.63). Radiation treatment was not associated with overall mortality; however, in the analysis limited to the first 9 months after diagnosis, radiation treatment was associated with a statistically significantly decreased mortality (HR: 0.45, CI: 0.24-0.86). Surgery type with partial resection was associated with decreased mortality compared to no surgery (HR: 0.5, CI: 0.29-0.88); other surgery subtypes failed to reach significance. Finally, any surgical resection, either with or without radiation, was significantly associated with decreased mortality in comparison to receiving neither radiation nor surgery: Radiation and surgery (HR: 0.40, CI: 0.17-0.93) and surgery alone (HR: 0.18, CI: 0.05-0.61). Unknown radiation and surgery status were also associated with decreased mortality [Table 2]. Tumor histology of glioblastoma (HR: 1.61, CI: 1.07-2.45, P = 0.02) and extent of resection being partial resection (HR: 0.47, CI: 0.27-0.86, P = 0.01) were confirmed as significant predictors of survival by multivariate analyses [Table 3].

Table 2

Univariate analysis of risk factors for mortality

Table 3

Multivariate analysis of risk factors for mortality

Multivariate analysis using a Cox proportional hazards model with specific surgical subtypes [Table 3] identified the following factors to be independent prognostic factors significant at a P < 0.05 level: Tumor histology with GBM was associated with increased mortality as compared to AA (HR: 1.61, CI: 1.07-1.63), and surgery type partial resection was associated with a decreased mortality compared to no surgery (HR: 0.47, CI: 0.27-0.86). In a stepwise multivariate analysis using a Cox-proportional hazards model adjusted for age, sex, race, date of diagnosis, tumor histology, and presence or absence of radiation and/or surgery, the following factors were found to be independent prognostic factors: Tumor histology with GBM was associated with an increased mortality relative to AA (HR: 1.74, CI: 1.17-2.60). Both receiving surgery alone without radiation and receiving surgery and radiation were protective against mortality compared to getting neither surgery nor radiation (surgery and radiation [HR: 0.35, CI: 0.15-0.82] and surgery alone [HR: 0.14, CI: 0.04-0.5]). Unknown radiation/surgery status was also protective (HR: 0.36, CI: 0.15-0.86), though this uncertainty in coding is a limitation of the SEER database and thus should be interpreted with caution [Table 4].

Table 4

Stepwise multivariate analysis of risk factors for mortality

Median survival among this group of pediatric patients with high-grade brainstem glioma was 10 months [Figure 1]. Stratified by tumor histology [Figure 2], median survival was 13 months for patients with AA and 9 months for patients with GBM; this difference was significant (P < 0.05) by log-rank test. Two-year survival for patients with AA was 34%, and 9% for patients with GBM.

Figure 1

Ten-year Kaplan–Meier survival estimates of pediatric patients with high-grade brainstem anaplastic astrocytoma. Median survival: Ten months; total n = 124. Survival at 1 year was 40%, at 2 years was 21.4%, at 5 years was 18.6%, and at 10 years was 15.3%

Figure 2

Ten-year Kaplan–Meier survival estimates by tumor histology. Median survival for patients with anaplastic astrocytoma (n = 60) was 13 months, and for patients with glioblastoma multiforme (n = 64) 9 months. Log-rank test for equality of survivor functions found a statistically significant difference between the two groups (χ2 = 8.96, P = 0.003)

Stratified by radiation treatment, no significant difference existed in survival among those who received radiation and those who did not [Figure 3]. However, worth nothing is that if the analysis is limited to the first 9 months after diagnosis, radiation is associated with lower mortality by the log-rank test for equality of survivor functions. Stratified by surgical treatment, receiving any surgical resection was significantly associated with decreased mortality; median survival and 2-year survival was 9 months and 12.6% for those who did not receive surgical resection, and 11 months and 31.2%, respectively, for those who did [Figure 4].

Figure 3

Ten-year Kaplan–Meier survival estimates by radiation type. Median survival: Five months for pediatric patients who did not receive radiation (n = 19) and 11 months for patients who did (n = 103). Log-rank test for equality of survivor functions did not find a statistically significant difference between the two groups

Figure 4

Ten-year Kaplan–Meier survival estimates by the presence of surgical treatment. Median survival: Nine months for those who did not get surgical resection (n = 61, no surgery or biopsy only) and 11 months for those who did (n = 43). Log-rank test for equality of survivor functions did find a statistically significant difference between the two groups (χ2 = 4.46, P = 0.03)

Discussion

This is a population-based study of 124 pediatric patients diagnosed with primary high-grade brainstem glioma. Data from the SEER registry between the years of 1973 and 2008 were reviewed to analyze the relationship between demographic, tumor histology, treatment modality factors, and survival outcomes. This study found a statistically significant correlation between lower-risk for mortality and tumor grade (Grade 3 compared to Grade 4), and lower-risk for mortality with surgical intervention alone or in combination with radiation therapy (compared to no treatment). We found that radiation was significantly associated with decreased mortality in the first 9 months after diagnosis, but not with mortality overall. Patients who received radiation therapy alone had 77.6% and 56% mean survival at 6 and 9 months compared to 47.3% and 36.8%, respectively, for those without treatment (P < 0.05).

No demographic factors, such as age, sex, or race, were found to be significantly associated with survival outcomes. Various case series of pediatric brainstem gliomas evaluating age as a prognostic factor have reported conflicting results.[317] In corroboration with our findings, Ueoka et al. performed a retrospective analysis of 86 patients bearing brainstem gliomas. The mean age was 14.2 years, and the median survival was 9 months. They found no difference between early and late progressors concerning age at onset or gender.[17] On the other hand, Sun et al. found older age to be a positive prognostic indicator of survival in a case series of 33 patients (P = 0.008).[3] Other comparable studies have not found age or other demographic factors to be significant prognostic indicators.[2182223] It is important to note that most reports describing age as a prognostic factor include a heterogeneous population of brainstem glioma patients with both low- and high-grade tumors. In these studies, no statistical analysis compared if the older patients with longer survival were the ones with low-grade focal tumors, which are known to be more amenable to surgery and have better prognoses. This current study specifically analyzed a homogeneous population of pediatric patients with high-grade gliomas.

Higher-grade tumor histology (GBM vs. AA) was found to be significantly associated with a worse survival outcome. This is consistent with previously published findings on pediatric brainstem gliomas.[2425] However, most studies of pediatric brainstem tumors emphasize focality, location, and clinical symptoms rather than tumor histology as prognostic indicators of survival.[11426] In addition, duration of symptoms has also been suggested to be a significant prognostic factor.[1727] Kaplan et al. have reported that longer duration of symptoms was positively correlated with better survival outcomes in a series of 119 pediatric patients with brainstem tumors.[27] SEER does not offer such granularity of data for analysis, and we acknowledge the lack of clinical details for further analysis. It is described that diffuse neoplasms tend to be high-grade gliomas (World Health Organization Grade III/IV), with a median survival of less than a year, in comparison with more focal gliomas (typically low-grade gliomas), with a median survival of over 5 years.[3892526] In addition, there is a paucity of reports analyzing exclusively high-grade brainstem glioma patients. Most studies compare prognostic factors that influence survival in focal versus diffuse lesions or low- versus high-grade tumors.[317]

In this study, radiation treatment was found to be associated with improved survival in the first 9 months after surgery, but not with overall survival. This is consistent with published studies reporting that, in cases of diffuse brainstem glioma, radiation can provide a transient improvement in neurological symptoms and a progression-free survival period; nevertheless, it is not shown to affect overall survival as cancer progresses after 6-9 months.[24252628] Although no randomized controlled trials comparing radiotherapy against no treatment for children with brainstem gliomas have been performed, radiation is the only treatment that has shown any beneficial effect in this population and thus is considered as the standard of care.[29142225262829] Previous studies have revealed that up to 70% of children bearing brainstem gliomas experience some neurological improvement after radiation therapy.[3031] The improvement following irradiation is noted on magnetic resonance imaging (MRI) in approximately 40-70% of cases.[32] In the present study, radiation was found to be associated with improved survival in the first 9 months both in univariate analysis and multivariate Cox regressions controlling for age, sex, diagnosis date, and type of surgery. However, such a positive response is only transient, and decline in survival was noted to occur 9 months after initial radiation therapy.[133] SEER does not provide specific data on the radiation regimen adopted, though a randomized control trial comparing the outcomes of 128 pediatric patients diagnosed with diffuse intrinsic brainstem tumors treated with either hyperfractionated or standard radiotherapy did not show a difference in survival between these two groups.[8]

Partial surgical resection, with or without radiation, was found in this study to be associated with an improved median survival of 2 months. As reported in the literature, in cases of brainstem neoplasms, surgical resectability is correlated with increased survival, both because surgical resection decreases local tumor burden and mass effect and because resectability often indicates the presence of a focal or exophytic tumor, which are generally lower-grade tumors.[123142425] Importantly, the greatest reduction in HRs, here, was with partial resection surgery alone. Possibly, the positive correlation between partial resection and better survival outcomes in both univariate and multivariate analyses may be due to selection bias in our studied population, since most of the patients in the surgical group may have had tumor morphology more amenable to resection. These tumors may represent more benign histology and association with better outcomes.[34] We also showed, through univariate analysis, an improved survival of patients treated with both radiation and surgery (P = 0.03). The literature provides conflicting information regarding this approach.[317] While some authors show improved survival compared to surgery alone, others do not describe any improvement. However, Sun et al. point out that lack of observed effect on survival could be due to a limited number of patients, which significantly decreases the statistical power of the analysis.[3]

Due to their infiltrative growth pattern, diffuse neoplasms are traditionally considered not amenable to surgery.[11525] In our study, 7% of the patients underwent gross total resection; however, such a technique was not correlated with better survival outcomes. Similarly, Epstein and McCleary published an account of radical resection in pediatric patients with diffuse brainstem neoplasms (n = 22). They reported that while it was possible to achieve such resection in the brainstem, they did not find it to be an effective treatment for the neoplasm,[26] as it did not address the underlying biology of the disease. In addition, 12% of our patients underwent diagnostic biopsy. Whether or not to perform stereotactic biopsy in a high-grade brainstem glioma is currently a heated discussion. While some authors assert that such a procedure should not be warranted in most cases due to the attendant risk of morbidity and lack of therapeutic value,[3536] others claim that it should be performed since the complication rates are low and the information provided could potentially alter treatment pathways.[3738] In our results, biopsy was not associated with improved survival either in the univariate or multivariate analyses. However, the information provided by such a procedure could open doors for the development of newly improved targeted therapies that hopefully could improve patients’ overall survival and quality of life. Finally, why unknown radiation/surgery status was associated with increased survival remains unclear; unknown radiation and surgery status was most common in the patients diagnosed before 1988; however, this variable is independently predictive of a better outcome even after adjusting for diagnosis date. This is a notable limitation in SEER data and should be interpreted with caution.

Study limitations include lack of data about the focality, specific location, and clinical symptoms of the brainstem gliomas, which in other studies have been shown to be important prognostic indicators of survival outcome.[2223] Moreover, there were ambiguities in coding that we could not confirm independently with the original clinical data. Extent of surgical resection may be biased in a few ways in this data: As the data span almost four decades, including the pre-MRI era, extent of resection would have been confirmed with computed tomography, which may overestimate the amount of tumor removed. Furthermore, extent of resection could not be confirmed by volumetric analysis in this study. Additionally, SEER does not contain data about chemotherapeutic regimens, which are often used in the treatment of brainstem neoplasms. However, to date, no single chemotherapeutic agent or combination of agents has been shown in clinical trials to significantly prolong survival in pediatric diffuse brainstem gliomas.[41415] This is thought partly to be as a consequence of poor access of the drugs to the tumor site due to the presence of a highly selective blood-brain barrier, poor intratumoral drug distribution, and the intrinsic resistance of the tumor cells.[13339] Similarly, SEER does not contain data about the type or dose of radiation therapy used. Nevertheless, it is important to note that, to date, no variation in type, dose of radiotherapy or combination with chemotherapy has been associated with increased survival in any clinical trials.[48151640] Finally, as with all population-based studies, these data are observational in nature and cannot be used to deduce causality or to influence clinical decision-making.

As to strengths, with 124 cases, this is one of the largest cohorts of pediatric brainstem high-grade gliomas reported until now. The strengths of SEER data include the high-quality control of data performed by the NCI's SEER program, a large study cohort, and long follow-up. These data may help to give more information for prognosis and influence future studies with timing and type of radiation and surgery that could impact clinical management. Because of the rarity of these tumors, continued support of and enrollment in the pediatric brain tumor consortia and databases are vital to amassing knowledge critical for understanding the biology of the disease and its optimal treatment.

Conclusion

We found that higher-grade tumors (glioblastoma) were statistically significantly associated with higher mortality. In addition, radiation therapy alone was associated with an increased survival in the first 9 months’ (P < 0.05) postinitial treatment as compared to no treatment. However, radiation therapy alone was not significantly correlated with overall survival. Partial surgical resection was significantly associated with increased survival. Demographic factors, such as sex, age, or race, were not independent predictors of survival.