Perinatal and Familial Risk Factors for Brain Tumors in Childhood through Young Adulthood

Brain tumors are the most common solid tumors and the leading cause of cancer mortality in childhood (age <15 years; ref. 1). Established risk factors, including ionizing radiation, neurofibromatosis 1, and other rare genetic syndromes, account for only a small minority of cases, whereas the etiology of most brain tumors remains unknown (2). Because brain tumors are a relatively common cancer in childhood, perinatal factors may be etiologically important, and their identification may provide insights into mechanisms. Case–control studies have reported that high birth weight is associated with the most common brain tumor subtypes (astrocytomas and medulloblastomas), possibly through growth factor pathways involving insulin-like growth factors (IGF; refs. 3, 4). However, most previous studies have focused on birth weight without examining its specific components, gestational age at birth and fetal growth; hence, the specific contributions of these factors are still uncertain. In addition, most of these studies have examined brain tumor risk only in childhood, and not longer-term risk into adolescence or adulthood. These issues have rarely been addressed in large population-based cohort studies, which have the potential to provide more robust and generalizable risk estimates and to clarify susceptible population subgroups.

We conducted the largest population-based cohort study to date to examine perinatal and familial risk factors for brain tumors and the most common subtypes among approximately 3.5 million individuals born in Sweden during 1973–2008. Detailed information on perinatal and familial factors and brain tumor incidence were obtained from birth and cancer registries that are nearly 100% complete nationwide. Our aims were to examine whether fetal growth, gestational age at birth, and other perinatal and familial factors are associated with brain tumors and specific subtypes in a large national cohort.

We identified 3,595,055 individuals in the Swedish Birth Registry who were born from 1973 through 2008. We excluded 10,424 (0.3%) individuals who had missing information for birth weight, and 7,702 (0.2%) others who had missing information for gestational age at birth. To remove possible coding errors, we also excluded 5,334 (0.1%) others who had a reported birth weight more than four SDs above or below the mean birth weight for gestational age and sex based on a Swedish reference growth curve (5). A total of 3,571,574 individuals (99.3% of the original cohort) remained for inclusion in the study. This study was approved by the Regional Ethics Committee of Lund University in Sweden.

The study cohort was followed up for brain tumor incidence from birth through December 31, 2010 (maximum attained age was 38 years). All incident brain tumors were identified using the International Classification of Diseases (ICD), revisions 7, 8, 9, and 10, in the Swedish Cancer Registry (codes 193.0 in ICD-7, 191 in ICD-8 and ICD-9, and C71 in ICD-10). This registry includes all primary incident cancers in Sweden since 1958, with compulsory reporting nationwide. Brain tumor subtypes were classified according to Systemized Nomenclature of Medicine (SNOMED) codes since 1993 and synonymous definitions provided by the World Health Organization prior to this period (6, 7), and were categorized as pilocytic astrocytomas (the most common brain tumor in childhood; ref. 8), other (nonpilocytic) astrocytomas that did not occur in sufficient numbers for separate analysis by histologic subtype (e.g., anaplastic astrocytomas, glioblastoma multiforme, and oligoastrocytomas), medulloblastomas, ependymomas, primitive neuroectodermal tumors (PNET), and other or unspecified subtypes.

Perinatal and familial characteristics that may be associated with brain tumors were identified from the Swedish Birth Registry and national census data, which were linked using an anonymous personal identification number (9). The following variables were examined as predictors of interest or adjustment variables: sex (male or female), birth year (modeled simultaneously as a continuous variable and a categorical variable by decade to allow for a nonlinear effect), fetal growth [a standardized variable defined as the number of SD from the mean birth weight for gestational age and sex based on a Swedish reference growth curve (5), modeled alternatively as a categorical (<−1; −1 to <1; ≥1 SD) or continuous variable], gestational age at birth [based primarily on maternal report of last menstrual period in the 1970s, at which time ultrasound estimation was gradually introduced until it was used exclusively starting in the 1990s; modeled alternatively as a categorical (<37, 37–41, ≥42 weeks) or continuous variable], birth weight [modeled alternatively as a categorical (<2,500, 2,500–3,999, ≥4,000 g) or continuous variable], birth length [crown-heel length in cm, modeled alternatively as a categorical (<48, 48–52, ≥53 cm) or continuous variable], season of birth [modeled alternatively as a categorical variable (Dec–Feb, Mar–May, Jun–Aug, Sep–Nov) or a sinusoidal function as described below], birth order (1, 2, ≥3); multiple birth (singleton vs. twin or higher order), maternal and paternal age at birth (<25, 25–29, 30–34, ≥35 years; examined separately for mothers and fathers), parental country of birth (both parents born in Sweden vs. 1 or both foreign-born; note that other specific information on ethnicity was unavailable), maternal and paternal education level [compulsory high school or less (≤9 years), practical high school or some theoretical high school (10–11 years), theoretical high school and/or some college (12–14 years), college and/or post-graduate study (≥15 years), examined separately for mothers and fathers], and family history of brain tumor in a parent or sibling (yes or no; identified from the Swedish Cancer Registry from 1958 through 2010, not self-reported, thus enabling highly complete and unbiased ascertainment during this time period). Information on head circumference at birth was unavailable.

Poisson regression with robust standard errors was used to estimate incidence rate ratios (IRR) and 95% confidence intervals (CI) for associations between perinatal or familial variables and brain tumors overall and the most common subtypes (10). Two different adjusted models were used in the main analyses: The first was adjusted for birth year and sex, and the second was further adjusted for other variables that were found to be associated with brain tumors (fetal growth, parental country of birth, maternal education level, and family history of brain tumor in a parent or sibling). Poisson models were a better fit for these data than Cox proportional hazards models, which did not meet the proportional hazards assumption. Poisson goodness-of-fit was assessed using deviance and Pearson χ² tests, which showed a good fit in all models.

As noted, season of birth was examined alternatively as either a categorical variable or a sinusoidal function. In the sinusoidal analysis, date of birth (DOB, coded as an integer from 1 to 365) was modeled as a sinusoidal function in the Poisson regression model, using an iterative method to identify the peak date for brain tumor relative risk and to test for an overall seasonal association, as previously described (11). In the case of a leap year, February 29 was recoded as calendar day 59 so that the respective year had 365 days. Specifically, the trigonometric term entered into the Poisson model was

where t_max (the peak birthdate for brain tumor risk) was determined iteratively by finding the value from 1 to 365 that maximized the model coefficient (11).

First-order interactions between sex and other variables were explored by examining risk estimates after stratifying by sex and formally testing for interactions using likelihood ratio tests. Multinomial logistic regression was used to test for heterogeneity in the association between each risk factor and earlier-onset (age <15 years) versus later-onset (age ≥15 years) brain tumors or subtypes. All statistical tests were two-sided and used an α level of 0.05. All analyses were conducted using Stata version 13.0 (12).

Among the 3,571,574 persons in this cohort, 2,809 (0.08%) brain tumors were identified in 69.7 million person-years of follow-up. The most common subtypes included pilocytic astrocytomas (n = 1,023; 36.4%), other (nonpilocytic) astrocytomas [n = 363; 12.9%; which included anaplastic astrocytomas (n = 95), glioblastoma multiforme (n = 82), oligoastrocytomas (n = 21), and other or not further specified astrocytomas (n = 165)], medulloblastomas (n = 282; 10.0%), ependymomas (n = 222; 7.9%), and PNET (n = 121; 4.3%), whereas 813 (28.9%) brain tumors were other or unspecified subtypes. The median age at diagnosis was 11.1 years (mean 13.0, SD 9.5) for brain tumors overall, 11.1 years (mean 12.8, SD 9.0) for pilocytic astrocytomas, 16.0 years (mean 17.0, SD 10.0) for nonpilocytic astrocytomas, 5.9 years (mean 7.3, SD 5.6) for medulloblastomas, 5.5 years (mean 9.1, SD 8.9) for ependymomas, and 7.7 years (mean 9.2, SD 7.7) for PNET. Overall brain tumor incidence rates by age and sex are shown in Table 1. Incidence rates for brain tumor subtypes are reported by age and sex in Supplementary Table S1.

The strongest risk factor was a first-degree family history of brain tumor, which was associated with a >2-fold risk of brain tumor in the proband (fully adjusted IRR, 2.43; 95% CI, 1.86–3.18; P < 0.001; Table 2, adjusted model 2). There was no evidence that this association varied by whether the affected family member was male or female (P = 0.63), or by whether the affected family member was the same or opposite sex as the proband (P = 0.45). When examined separately, there was no significant difference in risk estimates comparing family history in a sibling (fully adjusted IRR, 3.10; 95% CI, 1.76–5.46; P < 0.001) to family history in a parent (fully adjusted IRR, 2.37; 95% CI, 1.76–3.19; P < 0.001; P_{heterogeneity} = 0.41; not shown in the table).

Other significant risk factors for brain tumors included high fetal growth (fully adjusted IRR per additional 1 SD, 1.04; 95% CI, 1.01–1.08; P = 0.02). Individuals whose parents were both Swedish-born had a modestly increased risk of brain tumors relative to those with at least 1 foreign-born parent (fully adjusted IRR, 1.21; 95% CI, 1.09–1.35; P < 0.001). Further stratification of foreign-born parents into Europe/US/Canada versus other countries yielded similar risk estimates, hence they were modeled as a single category. In addition, high maternal (but not paternal) education level was associated with a modestly increased risk of brain tumors (P_trend = 0.01; Table 2, adjusted model 2).

Birth weight and length were also examined as alternatives to the standardized fetal growth variable. High birth weight was associated with an increased risk of brain tumors (fully adjusted IRR for ≥4,000 vs. 2500–3999 g, 1.13; 95% CI, 1.03–1.25; P = 0.009), but the test for trend was nonsignificant (fully adjusted P_trend = 0.06). Birth length also had risk estimates in the same direction but the trend test was nonsignificant (fully adjusted P_trend = 0.13). Gestational age at birth, season of birth, birth order, multiple birth, parental age, and paternal education level were not associated with brain tumors overall (Table 2). When season of birth was examined alternatively as a sinusoidal function (rather than 4 categories), the peak birth date for overall brain tumor risk was April 29, and it remained nonsignificant (fully adjusted P = 0.19; not shown in the table).

There was no evidence of heterogeneity in the associations between the predictor variables and brain tumors by age at diagnosis. High fetal growth appeared to be at least as strongly associated with brain tumors with onset at age ≥15 years (fully adjusted OR, 1.06; 95% CI, 1.00–1.12; P = 0.06; based on 1,783 cases) compared with age <15 years (fully adjusted OR, 1.03; 95% CI, 0.99–1.08; P = 0.14; based on 1,026 cases; P_{heterogeneity} = 0.53; not shown in the table). When potential interactions by sex were examined, the association with family history appeared somewhat stronger among females (fully adjusted IRR, 3.24; 95% CI, 2.29–4.58; P < 0.0001) than males (fully adjusted IRR, 1.77; 95% CI, 1.15–2.73; P = 0.009) (P_interaction = 0.03). No other interactions were found between sex and any other variables, including fetal growth (P_interaction = 0.53), with respect to brain tumor risk.

Subtype analyses had limited statistical power to detect significant associations. The association between high fetal growth and pilocytic astrocytomas had a similar risk estimate as for brain tumors overall, but the trend test was nonsignificant (IRR per additional 1 SD, 1.05; 95% CI, 1.00–1.12, P = 0.07) (Table 3). Each 1,000 g increase in birth weight was associated with a 14% higher risk of pilocytic astrocytomas (IRR, 1.14; 95% CI, 1.02–1.27; P = 0.02). There were no clear associations between fetal growth or birth weight and non-pilocytic astrocytomas, medulloblastomas, or ependymomas (Table 3), nor with PNET (Supplementary Table S2).

Male sex was a significant risk factor only for medulloblastomas (IRR, 1.35; 95% CI, 1.07–1.72, P = 0.01) and PNET (IRR, 1.49; 95% CI, 1.03–2.15; P = 0.03). A first-degree family history of brain tumors was strongly associated (>2-fold risks) with each major subtype except nonpilocytic astrocytomas (Table 3). Individuals whose parents were both Swedish-born had a significantly increased risk of pilocytic astrocytomas (Table 3). Birth in winter appeared to be associated with an increased risk of ependymomas relative to fall (P = 0.009; Table 3). When modeled alternatively as a sinusoidal function, season of birth had a borderline-significant association with ependymomas, with peak risk corresponding to a birth date of March 3 (P = 0.05; not shown in the table). There were no significant seasonal associations between birth date and other subtypes. There was no evidence for heterogeneity in the association between any risk factor and brain tumor subtype by age at diagnosis, comparing <15 versus ≥15 years (P > 0.05 for each).

In this large national cohort study, we found that high fetal growth was associated with a modestly increased risk of brain tumors overall in childhood through young adulthood, independently of gestational age at birth, and irrespective of age at disease onset. Although subtype-specific analyses had limited power, the association with high fetal growth appeared to involve pilocytic astrocytomas but not other subtypes. We also identified several other risk factors, including a first-degree family history of brain tumor (associated with all major subtypes except non-pilocytic astrocytomas), male sex (medulloblastomas and PNET only), having Swedish-born parents (pilocytic astrocytomas), and high maternal education level (brain tumors overall). Gestational age at birth, birth order, multiple birth, and parental age were not associated with brain tumors.

Most previous studies have reported an association between high birth weight and brain tumors overall and the most common subtype (astrocytomas), whereas results have been more discrepant for less common subtypes. A meta-analysis of 8 studies (6 case–control and 2 smaller cohort studies with a total of 4,162 brain tumors) examined high birth weight (defined as >4,000 vs. ≤4,000 g) and reported odds ratios of 1.38 (95% CI, 1.07–1.79) for astrocytomas, 1.27 (95% CI, 1.02–1.60) for medulloblastomas, and no association with ependymomas (3). None of these studies examined the specific components of birth weight, gestational age and fetal growth, and none examined the risk of brain tumors beyond adolescence. The current study suggests that high fetal growth is associated with an increased risk of brain tumors overall, independently of gestational age, and not only in childhood but also later-onset disease into young adulthood. This finding appeared to involve pilocytic astrocytomas and not other major subtypes, although precision of subtype-specific risk estimates was limited. High birth weight or fetal growth has also been associated with various other cancers (13), including Hodgkin lymphoma (14) and non-Hodgkin lymphoma (15), in this cohort.

The underlying mechanisms are not yet established but are hypothesized to involve growth factors such as IGF-I and IGF-II, which are known to be overexpressed in various brain tumors, including gliomas (16). Circulating levels of these factors are highly correlated with fetal growth and have been shown to inhibit cell apoptosis and promote tumor growth (17, 18). Consistent with our findings, high circulating levels of IGF-I have been associated with increased risk of low-grade but not high-grade gliomas (19). Although genetic data are sparse, polymorphisms in the IGF-I gene (IGF1) and IGF-I receptor gene (IGF1R) have also been associated with low-grade but not high-grade gliomas (20), suggesting different genetic pathways for these tumors. More comprehensive assessments of genetic variants in the IGF family are needed to further elucidate their potential role in the etiology of brain tumors.

We also found that a first-degree family history of brain tumor was associated with a >2-fold risk of all major subtypes except nonpilocytic astrocytomas. This finding is overall consistent with previous risk estimates from Sweden (21) and the United States (22), and may reflect genetic factors as well as shared environmental exposures. Segregation analyses in families of brain tumor patients have suggested that inheritance is best explained by polygenic or autosomal recessive models (23, 24), but this accounts for only a small proportion of brain tumors. The heritability of brain tumors has been estimated to be only 2% in the United Kingdom (25) and 12% in Sweden (26), whereas a large proportion is likely related to unknown environmental exposures.

The modest associations we found between male sex and medulloblastomas or PNET, and between having Swedish-born parents and pilocytic astrocytomas, are broadly consistent with previous findings. A slight male preponderance for brain tumors and specifically medulloblastomas has been previously described (27–29), as well as higher reported brain tumor incidence in Sweden than in most other countries in Europe (30) or worldwide (31). These relationships are still not well understood, but the male preponderance may be partly explained by subtype-specific estrogen receptor expression in brain tumors and a protective effect of estradiol compared with testosterone on tumor proliferation (32), whereas ethnic differences may be partly explained by genetic variants that predispose to brain tumors (33). We also found a modest association between high maternal (but not paternal) education level and overall brain tumor risk. A link between high socioeconomic class and brain tumors has been previously reported (34, 35) but is inconsistent across studies, and may potentially be related to differences in diagnosis or unmeasured occupational-related exposures.

Previous studies have investigated potential infectious etiologies for brain tumors, including proxies for early infectious exposures such as season of birth, with discrepant results. Some (36–38) but not all (39, 40) have reported a link between maternal infections during pregnancy and an increased risk of brain tumor in the child. Reported findings for season of birth and brain tumors have also been inconsistent, including positive associations for brain tumors overall with peak risk occurring among persons born in winter (41–44), early summer (45), or no association (46, 47). We found a borderline-significant sinusoidal relationship between season of birth and ependymomas, with peak risk among persons born in spring, but did not confirm associations with other subtypes or brain tumors overall.

Important strengths of this study are its national population-based cohort design with a large sample size, enabling more robust and generalizable inferences, and sufficient follow-up to examine risks into young adulthood. Linkage of birth and cancer registries provided detailed information on perinatal factors and brain tumor incidence that was nearly 100% complete nationwide (48, 49). The cohort design prevented selection bias that may potentially occur in case–control studies, and the use of registry-based data prevented bias that may result from self-reporting. Subtype information enabled exploratory analyses of major brain tumor subtypes. Family history of brain tumors was also based on registry data with virtually complete ascertainment, thus improving the reliability of those risk estimates.

Study limitations included the unavailability of data on other potentially important environmental exposures such as radiation, chemicals, or smoking. Although we examined season of birth and birth order as proxies for early infectious exposures, we were unable to directly assess specific infections or infectious exposures later in life that could potentially influence disease risk. Additional cohort studies with perinatal and other environmental data are needed to examine more complex etiologic pathways and age windows of susceptibility. As in other previous studies, subtype-specific analyses were limited by the rarity of these tumors as well as changes in classification methods over time.

In summary, this large national cohort study identified several risk factors for brain tumors among persons born in Sweden during 1973–2008, including high fetal growth, family history of brain tumors, having Swedish-born parents, and high maternal education level. High fetal growth appeared to be related to pilocytic astrocytomas (but not other major subtypes), independently of gestational age at birth, and irrespective of age at disease onset. These findings suggest that growth factor pathways may play an important role in the etiology of certain brain tumor subtypes from childhood into young adulthood. Further elucidation of these pathways may potentially reveal new subtype-specific targets for preventive or therapeutic interventions.

No potential conflicts of interest were disclosed.

The funding agencies had no role in the design and conduct of the study, in the collection, analysis, and interpretation of the data, or in the preparation, review, or approval of the manuscript.

Conception and design: C. Crump, J. Sundquist, W. Sieh, M.A. Winkleby, K. Sundquist

Development of methodology: C. Crump, J. Sundquist, M.A. Winkleby, K. Sundquist

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Sundquist, K. Sundquist

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Crump, J. Sundquist, W. Sieh, K. Sundquist

Writing, review, and/or revision of the manuscript: C. Crump, J. Sundquist, W. Sieh, M.A. Winkleby, K. Sundquist

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Sundquist, K. Sundquist

Study supervision: C. Crump, J. Sundquist, K. Sundquist

This work was supported by the NCI at the NIH (grant number R03 CA171017), the Swedish Research Council, and ALF project grant, Region Skåne/Lund University (Lund, Sweden).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.