|Home | About | Journals | Submit | Contact Us | Français|
Children of different racial/ethnic backgrounds have varying risks of cancer. However, few studies have examined cancer occurrence in mixed ancestry children.
Population-based case-control study examining cancer among children age <15 years using linked cancer and birth-registry data from 5 U.S. states from 1978 to 2004. Data were available for 13,249 cancer cases and 36,996 controls selected from birth records. Parental race/ethnicity was determined from birth records. Logistic regression was used to examine the association of cancer with different racial/ethnic groups.
Relative to Whites, Blacks had a 28% decreased risk of cancer (odds ratio (OR) 0.72, 95% CI 0.65–0.80), while both Asians and Hispanics had an approximate 15% decrease. Children of mixed White/Black ancestry also were at decreased risk (OR 0.71, 95% CI 0.56–0.90), but estimates for mixed White/Asian and White/Hispanic children did not differ from those of Whites. Relative to Whites: 1.) Black and mixed White/Black children had decreased ORs for acute lymphoblastic leukemia, (0.39, 95% CI 0.31–0.49) and (0.58, 95% CI 0.37–0.91), respectively; 2.) Asian and mixed White/Asian children had decreased ORs for brain tumors, (0.51, 95% CI 0.39–0.68) and (0.79, 95% CI 0.54–1.16), respectively; and 3.) Hispanic and mixed White/Hispanic children had decreased ORs for neuroblastoma (0.51, 95% CI 0.42–0.61) and (0.67, 95% CI 0.50–0.90), respectively.
The tendency of mixed ancestry children to have risks more similar to racial/ethnic minority children than the White majority group may help formulate etiologic studies designed to more directly study possible genetic and environmental differences.
Children of different racial or ethnic backgrounds have varying risks of cancer (1). These observations may be attributable to differences in genetic predisposition as well as environmental exposures. Most United States (U.S.) studies have focused on comparisons between White and Black race children (reviewed in Ref(1)), although in recent years, studies also have examined cancer rates among Hispanic and Asian children as the sizes of those populations have grown in the U.S. (2–6). International studies also suggest that rates of different childhood cancers may differ across racial/ethnic groups (7;8). However, very few studies have specifically examined cancer occurrence in mixed ancestry children (9;10) as the proportion of mixed ancestry individuals is typically low in most countries (e.g. <3% in the U.S. (11)). Examination of mixed race/ethnicity children may provide added insight into possible genetic and environmental factors associated with oncogenesis.
Our analysis used cancer registry data from five states that comprise approximately 30% of the U.S. pediatric population (11). The large study size and the inclusion of several racially and ethnically diverse states allowed examination of common childhood cancers across racial/ethnic groups and, in particular, among mixed race/ethnicity children. In addition to providing race/ethnicity information about children’s parents (information unavailable in cancer registries), the use of birth records to obtain subject data also permitted adjustment of several birth factors which could potentially mediate differences in cancer risk.
Human subject protection committee approval was received by the participating institutions prior to study conduct. Five states (California, Minnesota, New York [excluding New York City], Texas, and Washington) have previously linked birth records to their state cancer registries for selected years ranging from 1970–2004. Cases were classified using the International Classification of Childhood Cancer (12) from the original International Classification of Disease for Oncology codes (version 2 or 3, depending on state). For this analysis, Burkitt lymphoma was analyzed as part of non-Hodgkin lymphomas. A separate “embryonal” tumor group also was created consisting of embryonal central nervous system (CNS) tumors, neuroblastomas, retinoblastomas, Wilms tumors, hepatoblastomas, and embryonal rhabdomyosarcomas.
Each state also selected controls from birth records matched to cases on birth year (all states) and sex (California and Texas only). Variable control:case matching ratios were used, ranging from 1:1 in Texas to 10:1 in Washington. Details of each state’s linkages have been described previously (13–17). For this analysis, these state datasets were pooled with the following modifications: cancer diagnoses were restricted to those aged 28 days through 14 years (through 4 years in California); cases who were selected as controls were excluded from the pooled control group (allowed in Minnesota and New York); and subjects with Down syndrome were excluded (data unavailable in Texas before 1984 and Washington before 1989). Last, analysis was restricted to years with parental race/ethnicity recorded in birth records and available for both parents: California (1983–1997), Minnesota (1989–2004), New York (1978–2001), Texas (1980–1998), and Washington (1988–2004). During this time period, 1,948 (12.8%) and 6,254 (14.3%) otherwise eligible cases and controls, respectively, had one or both parents’ race/ethnicity information missing. The final dataset included 13,249 cases of childhood cancer and 36,996 controls.
The primary variable of interest was parental race and ethnicity. Separate child’s race/ethnicity was not available from all states. Parental racial groups were recorded on birth records with varying categorizations by state and time period; for this analysis they were summarily categorized as: White, Black, Asian, Native American, and Other. Pacific Islanders were grouped with Asians in the pooled dataset. Ethnicity was classified as Hispanic versus non-Hispanic on birth records. For this analysis, a combined race/ethnicity categorization was created consisting of Hispanics (regardless of race) and all racial groups exclusive of Hispanics. Children were classified into discrete categories based on their parents’ ancestry, including mixed ancestry offspring (Table 1). Given the large number of possible mixed ancestry combinations and relatively small numbers, we decided a priori to analyze only selected groups of mixed White ancestry children.
Other variables used in this analysis were: maternal age (<25, 25–29, 30–34, ≥35 years), maternal education (<12, 12, 13–16, ≥17 years), birth year, offspring sex, gestational age (<37, 37–41, ≥42 weeks, preferentially based on last menstrual period), birth weight (<2500, 2500–3999, ≥4000 grams), plurality (singleton, multiple birth), and birth order (1st, 2nd, or higher). Gestation <20 weeks or ≥45 weeks, and birth weights <350 grams were considered implausible and treated as missing.
Cancer risk associated with each race/ethnicity category was estimated via odds ratios (OR) and 95% confidence intervals (CI) using unconditional logistic regression (SAS version 9.1, Cary, NC). Individual matching in the California dataset was broken for this reason; frequency matching was used by the other states. ORs were estimated only for those strata with at least three observations. All estimates were adjusted by state, maternal age, birth year, offspring sex, gestational age, birth weight, plurality, and birth order. As 26% of our study population did not have maternal education information recorded, maternal education was not included in the final model. However, sensitivity analyses that included maternal education showed no substantial differences from our main results (data not shown). Given the different control:case ratios for each state, a separate sensitivity analysis with only one randomly selected control per case was conducted, which also displayed no difference from our reported results (data not shown). Additional analyses examined differences between racial/ethnic associations stratified by age and possible statistical interactions between racial/ethnic groups and offspring sex.
Same race White children made up 62.9% of cases and 66.3% of controls (Table 1). Same race Hispanics formed the second largest racial/ethnic group, making up 20.3% of cases and 15.4% of controls, and mixed ancestry children accounted for 8.3% of cases and 9.3% of controls. Compared with the study population, the distribution of cancer diagnoses among children with incomplete parental ancestry was similar (data not shown). However, compared with the study population (Table 2), offspring with missing ancestry were more likely to have mothers who were younger (<25 years, 54.3 versus 32.3%) and less well-educated (<12 years, 33.6 versus 18.1%), and were themselves more likely to be first born (50.3 versus 39.6%), born premature (<37 weeks, 12.0 versus 8.0%), and with lower birth weight (<2500 grams, 8.8 versus 5.2%). The distribution of other demographic and perinatal characteristics among those with missing parental ancestry information did not differ between cases and controls (data not shown).
When the overall risk of cancer was examined across unmixed racial/ethnic groups, Black, Asian, and Hispanic children had significantly decreased ORs for cancer compared with those of White ancestry (Table 3; Figure 1A). Blacks had the largest decrease (OR 0.72, 95% CI 0.65–0.80) followed by Asian and Hispanic children, both with approximate 15% relative decreases. Compared with White children, those of mixed White/Black ancestry had similar risk as Black children (OR 0.71, 95% CI 0.56–0.90), but estimates for White/Asian and White/Hispanic children were no longer significantly decreased (OR=0.91 for both groups).
Black children also had the lowest OR for acute lymphoblastic leukemia (ALL) versus Whites (0.39, 95% CI 0.31–0.49; Table 3; Figure 1B) with mixed White/Black children at intermediate risk (OR 0.58, 95% CI 0.37–0.91 versus Whites; OR 1.49, 95% CI 0.91–2.44 versus Blacks). Native American children had the greatest risk of ALL compared with Whites (OR 1.63, 95% CI 0.84–3.15). Among lymphomas, only Hispanic and mixed White/Hispanic children had increased ORs (ranging from 1.84 to 1.94) for Hodgkin lymphoma compared with White children. In contrast, compared with White ancestry, being Hispanic or White/Hispanic was associated with borderline decreased risk of non-Hodgkin lymphoma (OR 0.73 and 0.59, respectively).
Compared with Whites, most non-White groups had decreased ORs for CNS tumors (Table 3; Figure 1C), with Asians having the lowest risk (OR 0.51, 95% CI 0.39–0.68), followed by Hispanics and Blacks. Children of mixed White/Asian ancestry were at intermediate risk (OR 0.79, 95% CI 0.54–1.16 versus Whites; OR 1.56, 95% CI 0.97–2.50 versus Asians). Risk estimates for mixed White/Black and White/Hispanic children were similar to same race Blacks and Hispanics, respectively. This pattern of decreased ORs among non-White groups also was seen across most individual CNS tumor histologies, although estimates often lacked precision.
Among embryonal tumors, there were wide differences in the ORs for neuroblastoma, with Blacks, Asians, and Hispanics at progressively decreased risk versus the White referent group (Hispanics, OR 0.51, 95% CI 0.42–0.61; Table 3; Figure 1D). Neuroblastoma risk estimates for mixed race/ethnicity groups compared with Whites were imprecise except for mixed White/Hispanics (OR 0.67, 95% CI 0.50–0.90). Estimates for retinoblastoma were similarly imprecise for most groups, with Blacks and mixed White/Hispanics associated with borderline increased ORs (ranging from 1.34 to 1.39) versus Whites. Hispanic and particularly Asian ancestry (OR 0.39, 95% CI 0.24–0.63) were associated with a decreased risk of Wilms tumor, while Blacks were at borderline increased risk. Mixed ancestry groups generally had imprecise estimates. Only Black children were at significantly different risk of hepatoblastoma compared with Whites (OR 0.29, 95% CI 0.11–0.80; Black versus mixed White/Black children, OR 0.17, 95% CI 0.04–0.71).
Only Hispanics had a significantly decreased OR for soft tissue sarcomas versus Whites (0.74, 95% CI 0.59–0.92; Table 3). In contrast, Hispanics had an increased risk of osteosarcoma (OR 1.65, 95% CI 1.05–2.69). Only two Black and one mixed White/Black individuals were diagnosed with Ewing sarcoma in our population (combined OR 0.25, 95% CI 0.08–0.79). Most sarcoma histologies had insufficient sample size to allow estimation for mixed ancestry groups.
Both Hispanics and Asians had significantly increased ORs for germ cell tumors compared with Whites (1.33, 95% CI 1.01–1.75, and 2.63, 95% CI 1.78–3.90, respectively; Table 3). This risk appeared to be primarily due to increased risk of gonadal tumors rather than other extracranial lesions. Hispanic ancestry was associated with a decreased risk of extracranial non-gonadal germ cell tumor versus Whites (OR 0.53, 95% CI 0.29–0.94). In general, estimates for mixed White/Hispanic and White/Asian individuals appeared similar to same race Hispanic and Asian estimates, respectively.
Most racial/ethnic groups showed little difference in overall cancer risk when those diagnosed at age <5 years were compared with those diagnosed ≥5 years. Risk of cancer among Hispanics (but not White/Hispanics) relative to Whites was lower among the younger group (<5 years, OR 0.82, 95% CI 0.76–0.88; ≥5 years, 1.00, 95% CI 0.89–1.13). Among Native Americans, the opposite pattern was seen although estimates were imprecise (<5 years, OR 1.05, 95% CI 0.63–1.78; ≥5 years, 0.56, 95% CI 0.22–1.39). When ALL was examined separately, Hispanic ancestry showed the greatest discrepancy relative to Whites (<5 years, OR 0.99, 95% CI 0.89–1.11; ≥5 years, 1.38, 95% CI 1.12–1.69). Among children with lymphoma, Asians had the greatest discrepancy relative to Whites (<5 years, OR 1.61, 95% CI 1.03–2.51; ≥5 years, 0.73, 95% CI 0.48–1.10). Racial/ethnic groups had similar risk estimates for CNS tumors across age groups (data not shown). ORs for neuroblastoma and Wilms tumor were examined among those diagnosed age <2 years versus ≥2 years. Although no differences were observed for Wilms tumor, Asian and mixed White/Asian children had lower risks of neuroblastoma relative to Whites, particularly among those diagnosed <2 years (OR 0.34, 95% CI 0.20–0.58, and 0.66, 95% CI 0.34–1.30, respectively) versus ≥2 years (OR 0.85, 95% CI 0.53–1.37, and 1.12, 95% CI 0.55–2.30, respectively).
When statistical interaction between race/ethnicity and offspring sex was explored among the more common cancer subtypes, only neuroblastoma showed possible interaction (overall p-value=0.04). The decreased risk of neuroblastoma associated with Black ancestry relative to Whites was primarily among females (OR 0.41, 95% CI 0.25–0.67; males, OR 1.03, 95% CI 0.75–1.42). A similar pattern of relatively decreased risk also was observed among female mixed White/Black children, but estimates were imprecise. Risk estimates for other ancestry groups were similar across sex.
In our pooled analysis, we observed significant variability in childhood cancer risk across different racial/ethnic groups. The size and diversity of our population catchment area is similar to that surveyed by the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) program, except we had fewer Native Americans but a larger number of Hispanics (11;18). Similar to our findings, SEER data suggest that compared with Whites, Black children have a decreased risk of cancer overall, and particularly of ALL, Ewing sarcoma, and germ cell tumors (1). More modest decreased risks of CNS tumors and neuroblastoma also were observed. We did not find Black children to be at decreased risk for Hodgkin lymphoma or at increased risk of soft tissue sarcomas, as reported by SEER. However, this discrepancy may be due to our study population being limited to those diagnosed <15 years of age; racial differences for these tumor types were greatest amongst older teens (1). Differences between our results and those using cancer registry data alone also may be due to our ability to adjust for several birth factors potentially related to both race/ethnicity and risk of childhood cancer.
Few studies have examined the cancer risk among mixed ancestry children. South African studies suggest that children of European ancestry had similar incidence rates of ALL as White children in the U.S., but rates among South African Black children were more than 50% lower (9;10). Mixed race children in South Africa had intermediate rates of ALL (9;10). In our study, mixed White/Black children had a similar decreased overall risk of cancer as Blacks, although estimates for ALL may be intermediate between those of same race Black and White children.
Data from our study and others suggest that Hispanic children have a decreased risk of cancer overall versus non-Hispanic Whites (1;3). In our study compared with non-Hispanic Whites, Hispanics had an increased risk of Hodgkin lymphoma, osteosarcoma, and gonadal germ cell tumor, but a decreased risk of non-Hodgkin lymphoma, extracranial germ cell tumor, CNS tumor, neuroblastoma, Wilms tumor, and soft tissue sarcoma. Data from Florida suggest that Hispanics there, who are primarily of Cuban and Central American ancestry as opposed to predominantly Mexican ancestry Hispanics living in California and Texas, had a significantly higher rate of non-Hodgkin lymphoma (4). In contrast to the Florida study (4), we also did not find Hispanics to be at increased risk of ALL overall. Similar to results from a prior Californian study (3) with which we shared overlapping cases, we did not find Hispanic children <5 years of age to be at significantly increased risk of ALL. However, although our study did not include Californian children diagnosed ≥5 years, similar to that study (3) our data from other states suggest that risk of ALL may be increased among older Hispanic children versus Non-Hispanic Whites. Our use of birth records also restricts cases to those born in the U.S., and it is possible that relevant pre- and post-natal exposures may differ between foreign and native-born Hispanics. Last, our risk estimates for mixed White/Hispanic children either were similar to Hispanics or intermediate between Hispanic and White same race/ethnicity groups. We are unaware of prior studies examining this mixed ancestry group specifically.
Compared with Whites, we found that Asians had a decreased risk of cancer overall, and specifically of CNS tumors, neuroblastoma, and Wilms tumor, but an increased risk of germ cell tumor. A California study also reported decreased incidence rates of CNS and Wilms tumor and an increased rate of germ cell tumor among Asians (5). Although the number of Asians diagnosed with lymphoma was small in our study, Asians may have lower incidence rates of Hodgkin lymphoma but higher rates of non-Hodgkin lymphoma compared with Whites (5). An increased risk of both Hodgkin and non-Hodgkin lymphomas also was observed among South Asians ascertained in Great Britain (8). In our study, risk estimates for mixed White/Asian children mostly were intermediate between White and Asian same race estimates.
The number of Native Americans in our study was limited. Data from SEER, New Mexico, and Alaska suggest that Native Americans have lower rates of cancer versus Whites (1;19). However, those of Eskimo/Aleutian ancestry may have cancer rates more similar to Whites than other Native Americans (20).
Overall, although our study included nearly 5000 racial/ethnic minority children, our power to detect significant differences among less common cancer types and racial/ethnic groups was limited. It is possible some associations we reported may be due to chance. Furthermore, because of limited sample size and differences in how detailed racial/ethnic data from individual states were, certain groups such as Asians and Pacific Islanders were combined, masking potential heterogeneity. Nevertheless, for overall cancer risk and the more common cancer types, the statistical power associated with this study was robust, as demonstrated by analysis of confidence limits (21).
Race/ethnicity is a surrogate measure of underlying genetic as well as environmental exposures, neither of which we could measure directly except maternal and perinatal characteristics as recorded on birth certificates. We adjusted for maternal age and birth weight, both of which have been associated with differential risk of some childhood cancers (22;23). Studies with access to genetic material have shown that differences may exist in both tumor as well as germline DNA across racial/ethnic groups. For example, Hispanic and White children may have different ALL cytogenetic profiles (24) which may partially account for racial/ethnic differences in survival (25). The distribution of germline polymorphisms also differ across racial/ethnic groups, potentially making certain groups more susceptible to certain cancers (26). However, with the exception of rare inherited tumor predisposition syndromes, the magnitude of most genetic associations identified to date have been modest (1;27). Similarly, few consistent environmental risk factors have been identified for childhood cancers, and most also have been associated with only modestly increased risk (1;28). However, it remains plausible that different racial/ethnic groups may have variable levels of environmental exposures, and that there may be important interactions between selected exposures and one’s underlying genetic susceptibility.
Our study also is limited by the methods used to identify race/ethnicity, which may vary between and within states (29;30). Race/ethnicity may be self-reported by parents, abstracted from the medical chart, or in some instances, recorded by hospital staff based on their own observations (29). However, in validation studies where birth certificate data were compared with structured post-partum interviews, the sensitivity of birth records to correctly identify most racial/ethnic groups was >94% with the exception of Native Americans (29). Therefore, race information collected on birth records remains widely used in health research (31). Although there may be some misclassification, there is no reason to suspect that within any racial/ethnic group, parents of children who go on to develop cancer would have their race/ethnicity recorded differently than controls. Our use of same state controls and adjustment by state and birth year should minimize possible biases introduced by differences in racial/ethnicity classification across states and over time.
While cancer and birth registries typically record a child’s race/ethnicity, our use of parental race/ethnicity to define the child’s race/ethnicity has some advantages. Compared with single race/ethnicity individuals, the racial/ethnic identification of multiracial individuals tends to be less consistent over time, even in studies based on individual self-report (32). More importantly, since 1989, race and ethnicity assigned to infants on birth certificates has often automatically been that of the mother (29), which would preclude using child’s race/ethnicity to identify mixed ancestry individuals.
In summary, we found significant variability in childhood cancer risk across racial/ethnic groups. Lacking a better understanding of the genetic and environmental factors that predispose towards cancer, identification of race/ethnicity, however imprecise, remains an important characteristic in describing cancer risk. The tendency of mixed ancestry children to have risks more similar to racial/ethnic minority children than the White majority group in this study may help formulate etiologic studies designed to more directly study those genetic and environmental differences.
Sources of support: Children’s Cancer Research Fund, Minneapolis, MN; National Cancer Institute (T32 CA099936 to MN, N01-CN-05230 to WA, R01CA717450 to CA, R01CA92670 to TX); Fred Hutchinson Cancer Research Center; Centers for Disease Control and Prevention’s National Program of Cancer Registries by cooperative agreement (U58DP000783-01 to NY); Leukemia and Lymphoma Society Special Fellow in Clinical Research (4447-09 to EJC).
Financial disclosures: none