Relative survival after diagnosis with a primary brain or other central nervous system tumor in the National Program of Cancer Registries, 2004 to 2014

Abstract

Background

The majority of reported cancer survival statistics in the United States are generated using the National Cancer Institute’s publicly available Surveillance, Epidemiology, and End Results (SEER) data, which prior to 2019 represented 28% of the US population (now 37%). In the case of rare cancers or special subpopulations, data sets based on a larger portion of the US population may contribute new insights into these low-incidence cancers. The purpose of this study is to characterize the histology-specific survival patterns for all primary malignant and nonmalignant primary brain tumors in the United States using the Centers for Disease Control and Prevention’s National Program of Cancer Registries (NPCR).

Methods

Survival data were obtained from the NPCR (includes data from 39 state cancer registries, representing 81% of the US population). Relative survival rates (RS) with 95% CI were generated using SEER*Stat 8.3.5 from 2004 to 2014 by behavior, histology, sex, race/ethnicity, and age at diagnosis.

Results

Overall, there were 488 314 cases from 2004 to 2014. Overall 5-year RS was 69.8% (95% CI = 69.6%-69.9%). Five-year RS was 35.9% (95% CI = 35.6%-36.1%) for malignant and 90.2% (95% CI = 90.1%-90.4%) for nonmalignant tumors. Pilocytic astrocytoma had the longest 5-year RS (94.2%, 95% CI = 93.6%-94.6%) of all glioma subtypes, whereas glioblastoma had the shortest 5-year RS (6.1%, 95% CI = 6.0%-6.3%). Nonmalignant nerve sheath tumors had the longest 5-year RS (99.3%, 95% CI = 99.1%-99.4%). Younger age and female sex were associated with increased survival for many histologies.

Conclusions

Survival after diagnosis with primary brain tumor varies by behavior, histology, and age. Using such a database that includes more than 80% of the US population may represent national survival patterns.

Brain and other CNS tumors are the most commonly occurring cancer among children ages 0 to 14 years, and the eighth most common cancer among adults older than 40 years.¹ Brain and other CNS tumors are highly heterogeneous, varying significantly in tissue of origin, molecular genetic alterations, affected populations, and average outcome.¹ They can be broadly grouped into malignant (which invade adjacent tissue and are frequently fatal) and nonmalignant tumors,² and are classified and graded based on histologic and molecular characteristics. The most common malignant tumors are glioma (including glioblastoma, the most frequent primary malignancy), whereas the most common nonmalignant tumor is meningioma.

Population-based cancer survival rates are one of the critical statistics for evaluating progress in cancer screening and treatment over time, as well as the overall mortality burden due to these diseases. The majority of reported cancer survival statistics in the United States—including those reported in the Central Brain Tumor Registry of the United States (CBTRUS) annual statistical reports¹—have been generated using the National Cancer Institute’s (NCI’s) publicly available Surveillance, Epidemiology, and End Results (SEER) data, which prior to 2019 represented 28% of the US population (now 37%).^3,4 These registries are selected via a competitive application process and receive funding to support the collection of active survival follow-up information.

Central cancer registries funded by CDC’s National Program of Cancer Registries (NPCR) also collect outcomes information and are included as part of the US Cancer Statistics (USCS). The method for collecting outcomes varies by registry but is usually passive and achieved by linking with other vital statistics databases. NPCR maintains a database of survival statistics for registries that meet USCS Publication Criteria,⁵ which now represent 81% of the US population. Prior analyses of the NPCR survival data have found these data to be high quality and to produce reliable survival estimates.^6,7 NPCR registries also participate in the North American Association of Central Cancer Registries’ (NAACCR) annual Call for Data, which includes data needed for survival analyses. Data from most, but not all NPCR registries, were included in the 2018 Annual Report to the Nation on the Status of Cancer and NAACCR’s Cancer in North America.^8,9 Additionally, NPCR survival data have been included in CONCORD, a large international survival study.¹⁰ These publications report overall survival statistics for brain and other CNS tumors, but do not report histology-specific estimates. In the case of rare cancers, such as specific brain and other CNS tumor histologies, or special subpopulations, data sets based on a larger portion of the US population may allow for detailed analyses and contribute new insights into cancers with low incidence.

The overall goal of this analysis is to characterize the histology-specific survival patterns for malignant and nonmalignant primary brain tumors in the United States from 2004 to 2014 using the NPCR survival database.

Methods

This study was approved by the University Hospitals Cleveland Medical Center Institutional Review Board. Survival data were obtained from NPCR for 2004 to 2014,¹¹ which includes data from 39 of the 51 central cancer registries (funded by CDC’s NPCR), representing 81% of the US population. Survival time was generally calculated by file linkage, particularly with national or state death indexes, depending on the registry (Supplementary Table 1). Relative survival rates (RS) (1-year, 5-year, and 10-year) with 95% CI were generated by site, histology, age at diagnosis, sex, and race/ethnicity in SEER*Stat 8.3.5¹² using the actuarial method based on US life tables for expected survival from 1970 to 2014. Tumors were included only if they were the first malignant or nonmalignant reportable tumor for an individual. Cumulative expected survival rates were calculated using the Ederer II method, which calculates expected survival by matching cancer patients with equivalent individuals in the population and considers these matched individuals to be at risk until the cancer patient dies or is censored.¹³ Survival curves between strata over the entire period were compared using a z test as described by Brown¹⁴ and implemented in SEER*Stat, with 2-sided P values estimated using a normal distribution in R 3.5.¹⁵ Figures were created in R using ggplot2 and SEER2R statistical packages.^15–17

The CBTRUS classification is based on the 2000 revision to the World Health Organization (WHO) Classification of CNS Tumors, which is the most recent standard to be fully implemented in US cancer registration for the years covered in this article. Beginning January 1, 2018, tumors included in the 2007 and 2016 WHO Classification are included in collection practices,¹⁸ but as implied these data are not available for analyses. CBTRUS defines brain and other CNS tumors as all tumor morphologies located within the Consensus Conference site definition (International Classification of Diseases [ICD]-O-3 site codes C70-C72 and C75.1-C75.3), including lymphoma and other hematopoietic histologies, as well as olfactory tumors of the nasal cavity (for ICD-O-3 histology codes 9522-9523 only) (see Supplementary Table 2 for an overview of the classification).¹⁹ In comparison with analyses presented by CBTRUS, those of NPCR, SEER, and NAACCR exclude lymphoma and leukemia histologies (ICD-O-3 histology codes 9590-9989).

All brain and other CNS tumors are broadly classified into nonmalignant (also called benign, ICD-O-3 behavior code of/0 (benign) or/1 [uncertain]) and malignant (ICD-O-3 behavior code of/3). Though pilocytic astrocytoma is clinically considered and classified as a nonmalignant tumor with a WHO grade of 1, these tumors have historically been reported as malignant tumors for the purposes of cancer registration reports within North America and by the International Agency for Research on Cancer and International Association of Cancer Registries.^20,21 Therefore, this practice is followed by CBTRUS in its reporting of malignant tumors of the brain and other CNS locations, including its annual statistical reports, unless otherwise stated.

The inclusion of brain and other CNS categories within the NPCR survival set is based on the NPCR definition of brain and other CNS tumors, and therefore does not perfectly overlap with the survival data published in the annual CBTRUS statistical report. For this analysis, histologic groups were defined based on the CBTRUS histologic grouping (see Supplementary Table 2).¹

Results

There were 488 314 tumors included in the survival data set (Supplementary Table 3). The overall 5-year RS was 69.8% (95% CI = 69.6%-69.9%). For malignant brain and other CNS tumors, the 5-year RS was 35.9% (95% CI = 35.6%-36.1%), and for nonmalignant tumors it was 90.2% (95% CI = 90.1%-90.4%).

Though pooled survival was shorter on average for malignant tumors and longer for nonmalignant tumors, survival estimates varied significantly by histology. Glioblastoma was the most common type of malignant brain tumor (N = 99 576, 44.3% of individuals included in malignant analyses) and was the histology with the shortest 5- and 10-year survival (5-year RS = 6.1%, 95% CI = 6.0%-6.3%, and 10-year RS = 4.0%, 95% CI = 3.9%-4.2%) (Fig. 1A, Supplementary Table 4). The most incident–predominately nonmalignant histology is meningioma (N = 219 424), which had a 5-year RS of 88.0% (95% CI = 87.8%-88.0%). Survival for malignant meningioma was notably poorer than nonmalignant (including atypical) meningioma (malignant 5-year RS = 66.5%, 95% CI = 64.2%-68.8% vs nonmalignant 5-year RS = 88.3%, 95% CI = 88.1%-88.5%).

Fig. 1.

Relative Survival by A, Selected Histologies, and B, Relative Survival for Selected Glioma Histologies by Age at Diagnosis, National Program of Cancer Registries/US Cancer Statistics, Centers for Disease Control and Prevention, 2004 to 2014

Primary CNS lymphoma (5-year RS = 34.3%, 95% CI = 33.4%-35.3%, and 10-year RS = 26.0%, 95% CI = 24.9%-27.1%) and anaplastic astrocytoma (5-year RS = 29.6%, 95% CI = 28.6%-30.5%, and 10-year RS = 21.6%, 95% CI = 20.6%-22.6%) also had short survival outcomes.

The histology within the broad category of glioma with the highest 5- and 10-year survival was pilocytic astrocytoma (5-year RS = 94.2%, 95% CI = 93.6%-94.6% and 10-year RS = 92.5%, 95% CI = 91.8%-93.1%). The second highest 5- and 10-year survival for glioma histologic groups was observed in ependymal tumors (5-year RS = 90.1%, 95% CI = 89.3%-90.8% and 10-year RS = 86.1%, 95% CI = 84.8%-86.1%).

Among predominantly nonmalignant histologies, the histology with the lowest 5- and 10-year survival was craniopharyngioma (5-year RS = 82.0%, 95% CI = 78.7%-84.9%, and 10-year RS = 76.2%, 95% CI = 70.9%-80.7%) (Fig. 1A, Supplementrary Table 4). Nerve sheath tumors had the highest survival rates at all time intervals (5- and 10-year RS = 99.2%, 95% CI = 99.1%-99.3%). Mesenchymal tumors (5-year RS = 92.7%, 95% CI = 90.8%-94.2%, and 10-year RS = 86.8%, 95% CI = 82.6%-90.1%), hemangioma (5-year RS = 94.3%, 95% CI = 93.4%-95.0%, and 10-year RS = 91.9%, 95% CI = 90.0%-93.5%), and unique astrocytoma variants (5-year RS = 95.7%, 95% CI = 93.5%-97.1%, and 10-year RS = 90.5%, 95% CI = 85.2%-94.0%) were all associated with higher survival outcomes.

In general, RS decreased with increasing age, but this finding varied by histology. Within specific glioma histologies, younger age (ages 0-14 at diagnosis) was associated with higher survival, whereas individuals ages 65 and older at diagnosis had the poorest survival across all histologies (Fig. 1B).

Overall 5-year RS for children ages 0-14 years was 74.9% (95% CI = 74.2%-75.5%) for malignant tumors and 97.0% (95% CI = 96.6%-97.5%) for nonmalignant tumors (Supplementary Table 3). Glioblastoma was the histology with the lowest 5- and 10-year survival in this age group (5-year RS = 20.4%, 95% CI = 17.2%-23.8%, and 10-year RS = 17.2%, 95% CI = 13.9%-20.8%) (Supplementary Table 5). Among predominately nonmalignant tumors, the lowest 5-year survival was observed in craniopharyngioma (5-year RS = 91.4%, 95% CI = 85.7%-94.9%). Within specific embryonal histologies among children ages 0-19 years only, survival was generally highest in those ages 5-14 years, whereas children younger than 1 year at diagnosis had the poorest survival outcomes (Fig. 2).

Fig. 2.

Relative Survival for Selected Embryonal Histologies by Age at Diagnosis, National Program of Cancer Registries/US Cancer Statistics, Centers for Disease Control and Prevention, 2004 to 2014

Overall 5-year RS for adolescents and young adults (AYAs) ages 15-39 years was 71.9% (95% CI = 71.3%-72.4%) for malignant tumors and 97.4% (95% CI = 97.2%-97.6%) for nonmalignant tumors (Supplementary Table 3). Glioblastoma was the histology with the poorest survival in this age group (5-year RS = 25.0%, 95% CI = 23.4%-26.5% and 10-year RS = 16.5%, 95% CI = 14.8%-18.3%), whereas the highest survival for a glioma histology was observed in pilocytic astrocytoma (5-year RS = 95.0%, 95% CI = 93.9%-95.8% and 10-year RS = 92.9%, 91.3%-94.2%). Among predominantly nonmalignant histologies, craniopharyngioma was associated with the lowest survival in AYAs (5-year RS = 90.2%, 95% CI = 84.7%-93.8%, and 10-year RS = 83.2%, 95% CI = 72.7%-90.0%), whereas nerve sheath tumors were associated with the highest long-term survival (10-year RS = 97.8%, 95% CI = 97.2%-98.3%).

Overall 5-year RS for adults ages 40 years and older was 20.8% (95% CI = 20.6%-21.0%) for malignant tumors and 88.6% (95% CI = 88.4%-88.8%) for nonmalignant tumors (Supplementary Table 5). Glioblastoma was the histology with the shortest survival in this age group (5-year RS = 5.5%, 95% CI = 5.3%-5.7%, and 10-year RS = 3.4%, 95% CI = 3.2%-3.6%), and ependymal tumors were the glioma histology with highest longer-term survival (5-year RS = 90.4%, 95% CI = 89.3%-91.4%, and 10-year RS = 87.5%, 95% CI = 85.4%-89.3%). The predominately nonmalignant histology with the highest survival rates for adults was nerve sheath tumors (5- and 10-year RS both were 99.3%, 95% CI = 99.1%-99.4%).

Longer-term (10-year) survival was significantly higher in female patients compared with male patients overall, as well as within specific histologies (Supplementary Table 4). Within malignant germ cell tumors, survival was slightly but statistically significantly higher in men (5-year male RS = 88.3%, 95% CI = 86.2%-90.1% vs female RS = 84.1%, 95% CI = 79.8%-87.6%, P = .049). Among predominately nonmalignant tumors, the greatest female survival advantage was observed in nonmalignant meningioma (5-year male RS = 83.8%, 95% CI = 83.3%-84.3% vs female RS = 89.8%, 95% CI = 89.5%-90.1%, P < .001).

Overall survival after diagnosis with malignant tumors was higher in white Hispanics at all time periods, with 10-year RS = 44.9% (95% CI = 43.9%-45.9%) compared with 27.6% (95% CI = 27.3%-28.0%) in white non-Hispanics and 36.5% (95% CI = 35.3%-37.6%) in African Americans (Supplementary Table 3). White Hispanic individuals also had the highest RS at all time points for nonmalignant tumors, with 10-year RS = 91.1% (95% CI = 90.0%-92.1%) compared with 87.1% (95% CI = 86.7%-87.5%) in white non-Hispanics and 81.6% in African American individuals (95% CI = 80.5%-82.7%). This pattern was observed in many malignant histologies with the exception of malignant ependymal tumors, for which survival was highest among white non-Hispanics, and several histologies for which there was no substantial difference by race/ethnicity (Supplementary Table 6). White Hispanics had the highest long-term RS in most nonmalignant histologies, with the exceptions of nonmalignant ependymal tumors and craniopharyngioma (for which survival was highest in white non-Hispanics), and several histologies for which survival was approximately equal, including germ cell tumors.

Survival after diagnosis with embryonal tumors was nonstatistically significantly lower in African Americans compared with white non-Hispanics, but this pattern varied by specific embryonal histologies. There was no significant difference by race/ethnicity in medulloblastoma or primative neuroectodermal tumors (PNET), whereas African Americans had statistically significantly worse survival in atypical teratoid/rhabdoid tumors (ATRT, P = .0002).

Tumors of the cranial nerves had the highest 10-year survival of any site (RS = 99.4%, 95% CI = 99.2%-99.5%), followed by tumors of the spinal cord and cauda equina (RS = 91.7%, 95% CI = 90.6%-92.6%) (Supplementary Table 7). Parietal lobe tumors had the poorest 5- and 10-year RS, with 5-year RS = 27.1% (95% CI = 26.4%-27.7%) and 10-year RS = 22.1% (95% CI = 21.3%-23.0%).

Discussion

The NPCR survival data set includes 81% of the US population from 2004 to 2014, which is larger than the population covered by other population-based data sets commonly used for research, including the American College of Surgeons’ hospital-based National Cancer Database (~70%) and the NCI’s SEER program (37% as of 2019), with which the NPCR survival data set overlaps.^3,22 As of the 2019 data release, the NPCR survival data set will include 43 cancer registries for a total of 93% of the US population from 2001 to 2015.²³ As a result, these data are particularly well suited for evaluation of survival patterns in rare brain and CNS tumor histologies overall and within subpopulations. Prior analyses of these data showed robustness of results in comparison to other sources of population-based survival data in ability to produce reliable survival estimates.^6,7The present analysis represents the most comprehensive evaluation of survival patterns by specific brain tumor histologies from the largest US population-based survival data set available for research.

Survival statistics for brain and other CNS tumors are often presented overall, which leads to statistics that can be biased by including patients with the most common tumor types (eg, glioblastoma) and individuals among whom brain and CNS tumors are more common (eg, non-Hispanic whites, older individuals). The large sample size of the NPCR data set allows for comparisons between demographic groups, even within relatively rare tumor types, such as some pediatric tumor types, and allows for comparison by sex, race/ethnicity, and smaller age strata. The larger sample size of this data set means that these analyses are well powered to identify significant differences between groups that are small and may not be clinically relevant.

A statistically significant female survival advantage in glioblastoma has been previously reported in analyses of case series and population-based data sets.^24,25 The results of this analysis demonstrate that this difference exists at the US population level, especially for longer-term outcomes (5-year and 10-year RS, P < .001). This analysis also identifies a female survival advantage in less commonly occurring tumor types, including malignant and nonmalignant ependymal tumors (P < .001 and P = .008, respectively) and both malignant and nonmalignant meningioma (both P < .001) (Supplementary Table 4). This analysis identifies a male survival advantage in malignant germ cell tumors (P < .001), which is contrary to prior publications that have reported no sex difference in survival in smaller study populations.²⁶ Though this analysis identified statistically significant differences in survival by sex, many of these differences are generally small and may not be clinically meaningful. Sex-based survival differences may provide researchers insight into various factors that could contribute, in part, to increased survival overall.

This analysis confirms the results of prior studies that found a shorter comparative survival for non-Hispanic whites across many glioma histologies.^27–29 Prior studies have found significant survival differences by race/ethnicity in many glioma histologies (especially glioblastoma), but they have not previously found racial/ethnic differences in survival for ependymoma.^28,30,31 This analysis also confirmed a white non-Hispanic survival advantage both for malignant (compared both to Hispanics, P = .001 and African Americans, P = .001) and nonmalignant tumors (compared to African Americans only, P = .001) (Supplementary Table 6). Prior studies have had mixed results as to whether medulloblastoma survival varies by race/ethnicity,^32,33 but this analysis found no significant difference. Racial/ethnic differences in survival within ATRT and PNET (currently referred to as CNS embryonal tumors by the WHO) are novel and have not been identified in prior analyses of survival in these rare tumor types. Though this analysis identified statistically significant differences in survival by race/ethnicity, many of these differences are generally small and may not be clinically meaningful. These differences may point, in part, to disparities in access to diagnosis and treatment, as well as “true” differences that may be attributable, in part, to various risk factors.

There are several limitations to the NPCR survival data. One significant limitation to use of these data for clinically relevant survival estimates is the classification of pilocytic astrocytomas as malignant tumors when these tumors are clinically treated as nonmalignant tumors. To address this, we calculated survival statistics by behavior category both with the current pilocytic astrocytoma cancer registry reporting classification and with the more clinically appropriate classification (Supplementary Table 3). The reclassification has little effect on survival estimates for nonmalignant tumors, but leads to a decrease in overall survival for malignant tumors. This effect is largest in patients at younger ages, for whom these tumors represent a larger proportion of all brain and other CNS tumors. As is the case in most population-based cancer registry survival data sets, there is no central pathology review for these cases, and histology codes are based on the diagnosis assigned by the treating institutions’ pathologist. With the revision of the WHO classification of CNS tumors, the diagnostic criteria for brain and CNS tumor subtypes now depend on specific molecular markers. In particular, the incorporation of isocitrate dehydrogenase 1/2 mutation status, 1p/19q codeletion status, and histone H3 mutations into the diagnostic criteria for diffuse gliomas, as well as the molecular subtyping of medulloblastomas, have had a large effect on the way these tumors are diagnosed.² These molecular markers are not currently included in the public data releases for the years covered by this analysis, but are now collected by central cancer registries for cancers diagnosed from January 1, 2018, forward.

There are multiple methods for estimating net survival for cancer patients from population-based databases without the use of individual cause of death information,^13,34 all of which have strengths and limitations. RS estimates rate death due to cancer alone by adjusting for population mortality, and, as a result, are not directly comparable to relative survival estimates in other populations for which the underlying mortality rate may vary. The RS estimates presented in this report have been adjusted using estimated survival based on the Ederer II method and have not been age standardized. This could result in overestimation of the survival probability in pooled age estimates because older individuals’ contribution to long-term survival estimates is underestimated because of excess hazard of death from other causes. Because many brain tumors are more common in older age groups, this may result in some inflation of long-term survival estimates.

Prior analyses of other data sets have identified that treatment patterns may also vary by sex,²⁵ but this could not be assessed in this data set because treatment variables are not currently available. This also precludes other survival analyses that include treatment information. However, treatment data may be available in the future as these data are collected by central cancer registries when recorded in patient records. Source of data for survival follow-up varied within the data set (Supplementary Table 1), with the majority of survival outcome data coming from file linkages or state death indexes. More nonmalignant tumors were missing follow-up information compared with malignant tumors; hence survival estimates may be biased if missing survival data are correlated with certain patient characteristics (eg, facility type, age, race/ethnicity).

Conclusions

Survival after diagnosis with a primary brain tumor varies by behavior, histology, sex, race/ethnicity, and age at diagnosis. The worst 5-year survival outcomes were in glioblastoma, whereas nonmalignant nerve sheath tumors had the best survival outcomes. Younger age and female sex were associated with increased survival for many histologies. Survival outcomes from the NPCR survival data, a database that includes more than 80% of the US population, likely accurately represents national survival patterns overall and by histology for all primary brain and other CNS tumors.

Funding

This work was supported by a Research Training Grant from the Cancer Prevention and Research Institute of Texas [grant number RP160097T to QTO]. Funding for the Central Brain Tumor Registry of the United States was supported by the Centers for Disease Control and Prevention [CDC, contract No. 2016-M-9030], the American Brain Tumor Association, The Sontag Foundation, Novocure, Abbvie, the National Cancer Institute [NCI, contract No. HHSN261201800176P], the Musella Foundation, National Brain Tumor Society, the Children’s Brain Tumor Foundation, the Uncle Kory Foundation, and the Zelda Dorin Tetenbaum Memorial Fund, as well as private and in-kind donations. The findings and conclusions in this report are those of the authors and do not necessarily reflect the official position of the CDC or the NCI.

Conflict of interest statement. None declared.