Trends in Pediatric Central Nervous System Tumor Incidence in the United States, 1998–2013

Discussion

In this large, population-based study covering 60% of the U.S. population, we did not observe a statistically significant change in the incidence rates of all malignant CNS cancers combined between 1998 and 2013. We observed a statistically significant increase in the incidence rate of glioma (0.77% per year) and of pilocytic astrocytoma (0.89% per year) and a statistically significant decrease in incidence of embryonal cancer (−0.88% per year). We estimated that there were 120 excess cases of glioma, 72 fewer embryonal cancers, and 94 excess cases of pilocytic astrocytoma diagnosed U.S.-wide in 2013 compared with the expected had 1998 rates remained stable.

Childhood CNS tumor incidence trends in the United States have rarely been reported with the granularity presented here, making comparisons of our findings with others challenging. Consistent with our results, however, CBTRUS (using 52 NAACCR registries) reported a borderline significant increase of 0.6% per year for all malignant CNS tumors among 0- to 14-year-olds between 2000 and 2014 and of 1% per year among 15- to 19-year-olds from 2000 to 2008 (14, 27). Between 2000 and 2014, they estimated an increase in gliomas among 0-to 14-year-olds of 1.5% per year (14). The Centers for Disease Control and Prevention (using 47 NAACCR registries) reported a decrease in embryonal tumors from 2000 to 2010 of 0.6% per year, also consistent with our results (27). The compatible findings within the United States are not surprising given the shared data sources. By restricting to only high-quality registries, we decreased the likelihood of data quality influencing our findings. By including more narrowly defined histologic subgroups, we have shown diverging patterns in subtype-specific incidence rates underlying a higher level, average trend. Our results are consistent with recent trends in Europe showing, for example, increasing incidence of pilocytic astrocytoma and ATRT and decreasing incidence of PNET and embryonal cancers overall (28, 29).

The trends we report are likely to be the result of a combination of changing classification, reporting, diagnosis, and true changes in CNS cancer incidence. We observed increasing incidence rates of radiographically confirmed diagnoses of malignant glioma and pilocytic astrocytoma. We hypothesize there are at least three ways for shifting patterns in method of confirmation to influence incidence trends: (i) as imaging becomes more powerful and accessible, an increasing share of diagnoses have been made through imaging rather than biopsy leading to more general diagnosis categories; (ii) not previously detectable cancers are now being detected radiographically; and (iii) more frequent surveillance of patients with tumor predisposition syndromes and diagnoses by radiologists have led to increased reporting to registries by radiologists. The latter two factors could lead to the appearance of increasing incidence in the absence of a true increase in CNS cancer. A detailed analysis of subtype-specific trends in mortality and survival over the same period has not been conducted to our knowledge but may provide insight into whether detection of potentially indolent cancers is contributing to the observed trends.

In addition to changes in method of detection, some subtype trends are likely explained by the increasing availability and understanding of molecular pathology. For example, we found that “other” embryonal tumors (mostly ATRT) increased by 6.99% per year and PNET incidence decreased 5.67% per year. Up until the late 1990s, PNET and ATRT were difficult to distinguish based on clinical and pathologic features and most tumors were classified as PNET. Since then, increasing recognition of the distinct genetic signature of ATRT has increased the frequency of its diagnosis (30, 31). Similarly, tumors previously identified as astrocytoma NOS, which after molecular testing show H3F3A mutations, are now being diagnosed and classified as glioma NOS based on evolving understanding of the etiology and prognosis of cancers displaying these mutations (22, 32). This is reflected in our findings; astrocytoma NOS incidence decreased by 2.4% per year and glioma NOS incidence increased by 3.13% per year.

To address the possibility that subtype trends reflect classification shifts rather than true net changes, we reported trends in broader, aggregated groups that are less subject to classification patterns. Even trends in the aggregated groups, however, can be subject to practice changes, if, for example, new technology identifies cancers that would not have previously been reported. In the future, large biobanking efforts such as the Children's Brain Tumor Tissue Consortium (33), may allow for the retrospective classification of tumors using uniform standards that reflect our present day understanding of etiology. For this to contribute to trends research, however, accurate estimates of the size of the population contributing to the consortium would be needed as rate denominators.

Aside from classification and diagnosis, changes in risk factor prevalence could have contributed to changing incidence of pediatric CNS tumors. Ionizing radiation is the strongest known modifiable risk factor for pediatric CNS tumors (7). The dose–response relationship is strongest for meningiomas, which are nonmalignant and excluded from this analysis, but a significant association has also been found with gliomas (9). Steeply increasing use of CT scans could contribute to increasing incidence of gliomas among children aged 5 and older (younger children would have insufficient latency; ref. 34). Recent guidelines to identify pediatric patients at emergency departments who may not benefit from CT (35), and the increasing adoption of the “as low as reasonably achievable (ALARA)” principle for balancing dose with image quality, may decrease current and future dose from pediatric CT scans (36). Given the uncertain association between RF-EMF and malignant CNS tumors (37), it is unclear if and to what extent increases in exposure contribute to the glioma trends.

High birth weight is associated equally with increased risk of gliomas, astrocytomas, and embryonal tumors (5% increase in risk per 500-g increase; ref. 38). However, it is unlikely that high birth weight is driving increases in glioma and astrocytoma, given the U.S. prevalence of high birth weight has decreased over recent decades (39). Cancer predisposition syndromes, such as neurofibromatosis type 1 and 2, Li-Fraumeni syndrome, and Cowden syndrome are estimated to account for approximately 9% of all pediatric CNS cancers (40, 41) and prevalence of these inherited syndromes occurs over generations, making them unlikely drivers of the observed trends over the 15 years of this study. As brain tumor etiology is poorly understood, changes in another unknown exposure could have contributed to the changes in CNS tumor incidence we observed. Further research on the causes of pediatric brain tumors is merited.

The primary strength of this work is the coverage, size, and quality of the registries. Despite being the leading cause of childhood cancer–related death (2), pediatric CNS cancers are rare. The 29 geographically dispersed registries yielded sufficient cases to estimate recent trends by narrow subtype, age, sex, and race/ethnicity. The contributing registries have high completeness and standards that minimize misclassification and missingness (42, 43). Unlike previous studies (14, 27, 44, 45), we restricted to registries that received NAACCR certification to ensure the highest data quality. This is especially critical for CNS cancers, given the rarity of these tumors, the diversity of subtypes, and the varying classification systems.

The data does have inherent limitations. First, as has been outlined, trends in classification, diagnosis, or reporting could be mistaken for true trends in underlying rates. In light of this, we have shown results for all subtypes at specific and aggregated levels, and been cautious in our interpretation. Second, the collection of demographic variables, both in cancer registries (numerators) and in the census (denominators), can impact the estimation of rates and trends in subpopulations defined by these variables. For example, NAACCR introduced a Latino identification algorithm in 2003. Prior to that, each registry used its own approach (46). In addition, the 2010 census used different methods than the 2000 census to assign race/ethnicity to those reporting “some other race” or multiple races (47). The decreasing incidence rates of all malignant CNS cancers and gliomas among Latinos were not observed in the other race/ethnic groups and should be interpreted cautiously, given the changes in race/ethnicity classification in the registry and the census and distinct Latino migration patterns (48, 49).

In sum, we present recent trends in pediatric CNS tumor incidence rates in the United States, overall and stratified by subtype, age, sex, and race/ethnicity. Our data demonstrated that the overall incidence rate of malignant CNS tumors has not changed significantly between 1998 and 2013. We observed increasing rates of glioma and pilocytic astrocytoma and decreasing rates of embryonal cancer. Absolute increases were small (equivalent to 94 excess pilocytic astrocytoma and 120 excess glioma cases in a year) and partially offset by decreases in embryonal cancers (72 cases). Continued surveillance of pediatric CNS tumors should remain a priority, given their significant contribution to pediatric cancer–related deaths. The development and promotion of a historically compatible, clinically relevant, and widely accepted classification scheme for pediatric CNS tumors would improve the feasibility, comparability, and usefulness of trends analyses.