Data were combined from five previous studies in which the incident cases of childhood cancer were identified from the population-based cancer registries of five states: California, Minnesota, New York (excluding New York City), Texas, and Washington. Cases were diagnosed between 1980 and 2004. The details of each state's selection and inclusion criteria have been previously reported 14 (link). Children up to age 14 years at diagnosis were included except in California where only cases less than 5 years of age were included (the CA study was originally designed to study early childhood cancers only). Cases were matched to birth certificates using probabilistic or sequential deterministic record linkage. Controls were randomly selected from each state's birth registry, in ratios to cases varying from 1:1 to 1:10 (differed by state). They were matched on date of birth in all states and also matched on sex in California and Texas. Any subject reported to have Down syndrome was excluded (n=100). Because subjects diagnosed before age 28 days were excluded in some of the states, this criterion was applied to all cases for consistency.
We classified the cancers according to the International Classification of Childhood Cancer (ICCC-3) and examined all groups with at least 200 cases 15 (link). We made one exception to this rule in order to examine the 109 cases of chronic myeloproliferative diseases (CMD) because of our interest in leukemia sub-types. Wilms tumors and retinoblastoma were further examined by unilateral and bilateral occurrence. Additionally we examined the CNS tumors by type to reflect clinically relevant biological differences using categories previously developed 16 . We classified pilocytic astrocytomas, astrocytomas not otherwise specified, and other grade I and II gliomas into the low grade glioma category. Malignant gliomas, anaplastic astrocytomas, and other grade III and IV gliomas were grouped into the high grade glioma category. Other separate categories included medulloblastomas, primitive neuroectodermal tumors (PNET), ependymomas, and intracranial/intraspinal germ cell tumors.
Odds ratios (OR) and 95% confidence intervals (CI) were calculated using unconditional logistic regression (SAS version 9.1). The individual matching of the California cases and controls was broken to allow the use of this procedure. The other states used frequency matching. We examined birth order in four categories: first, second, third, and fourth or more. In the multivariable analyses we adjusted for matching and pooling variables (state, sex, year of birth), maternal race, maternal age, singleton vs. multiple birth, gestational age, and birth weight (all categorized as shown in Table 2). We also stratified the analyses for the leukemia sub-types by age at diagnosis (0-4 years, 5-9 years, 10-14 years).