Childhood ALL and Indicators of Early Immune Stimulation
Childhood ALL and Indicators of Early Immune Stimulation
Data from 11 Childhood Leukemia International Consortium case-control studies conducted in 8 countries from 1980 to 2010 were used (Table 1): Australian Study of Causes of Acute Lymphoblastic Leukemia in Children (AUS_ALL); Quebec Childhood Leukemia Study, Canada (CA_QCLS); Adele Study, France (FR_ADELE); Electre Study, France (FR_ELECTRE); Epidemiologic Study on Childhood Cancer and Leukemia, France (FR_ESCALE); Nationwide Registration for Childhood Hematological Malignancies, Greece (GR_NARECHEM); Study on the Etiology of Childhood Lymphohematopoietic Malignancies, Italy (IT_SETIL); New Zealand Childhood Cancer Study (NZ_NZCCS); United Kingdom Childhood Cancer Study (UK_UKCCS); Children's Oncology Group Study, United States (US_COG15); and Northern California Childhood Leukemia Study (United States) (US_NCCLS). The study design and participant characteristics for each study have been summarized previously.
All the data were collected with questionnaires that were administered to the parents face-to-face (FR_ADELE, GR_NARECHEM, IT_SETIL, NZ_NZCCS, UK_UKCCS, US_NCCLS) or by telephone (CA_QCLS, FR_ESCALE, US_COG15) or were self-administered (AUS_ALL, FR_ELECTRE). The questionnaires included information on demographic and socioeconomic characteristics and factors potentially associated with childhood leukemia. For the purpose of the present analysis, birth order, number and age of siblings, breastfeeding, history of common infections in the first year of life, and day-care attendance of the index child were provided by the investigators, as were sex, age at diagnosis or recruitment, and any other variables used for matching, as well as parental age at child's birth, parental education, and other indicators of socioeconomic status. All the studies included both B-cell and T-cell ALL.
Parental Education and Socioeconomic Status. Maternal and paternal levels of education were classified by using the same categories in all the studies: none or primary education, secondary education, and tertiary education (university). A heterogeneous 3-class (low, medium, high) indicator of socioeconomic status was also derived from the deprivation index based on address at diagnosis or interview (UK_UKCCS), household income (AUS_ALL, CA_QCLS, US_COG15, US_NCCLS), parental professional status (FR_ADELE, FR_ELECTRE, FR_ESCALE, GR_NARECHEM, NZ_NZCCS), or maternal education (IT_SETIL), depending on the data available (Web Table 1 http://aje.oxfordjournals.org/content/181/8/549/suppl/DC1 available at http://aje.oxfordjournals.org/).
Breastfeeding. Breastfeeding was classified by using the ever/never variable provided by the investigators. Children breastfed 1 month or less were also classified in the never breastfed group in some sensitivity analyses. The duration of breastfeeding was available for all the studies.
Day Care. Information on day-care center attendance was available for all the studies, and age at start of attendance was available for 10 studies (Web Table 1 http://aje.oxfordjournals.org/content/181/8/549/suppl/DC1). Attendance was considered full-time when attendance was at least 6 half-days per week, except in the French and Italian studies, which reported the frequency of attendance as full-time or part-time with no further details. Care by a child minder was also available in 5 studies.
Early Common Infections. History of common infections in the first year of life, as reported by mothers, was available in 8 studies (FR_ADELE, FR_ELECTRE, FR_ESCALE, GR_NARECHEM, IT_SETIL, NZ_NZCCS, UK_UKCCS, US_NCCLS). Web Table 1 http://aje.oxfordjournals.org/content/181/8/549/suppl/DC1 shows the sites of infections that were collected in the different studies. Four studies provided the total number of episodes for each site and, in the 3 French studies, a 3-class variable was available: no infection for the given site, between 1 and 3 episodes, and 4 or more episodes. The history of ear, nose, and throat surgery before age 3 years was available in 5 studies.
The analyses were restricted to children aged at least 2 years, first, to ensure that all the cases and controls had had the opportunity of having been breastfed for a prolonged period, of day-care attendance, and of contracting infections in their first year of life; and second, because common infections occurring before 1 year of age may have been related to a prediagnostic phase of the disease in the ALL cases aged less than 2 years at diagnosis.
Meta-analysis. Meta-analyses based on study-specific odds ratios were conducted for the main exposures of interest. The odds ratios were estimated by using either unconditional or conditional logistic regression, depending on the design of each study, and including study-specific matching variables in the models. The sociodemographic characteristics significantly associated with both case-control status and the exposure were also included in the study-specific models. Between-study heterogeneity was assessed by using Cochran's Q and I statistics. Summary odds ratios and 95% confidence intervals were implemented by using the inverse variance method, with random effects in the event of heterogeneity.
Pooled Analysis. Pooled odds ratios were estimated from individual data by unconditional logistic regression systematically adjusted for age, sex, and study. Maternal education and maternal age at the child's birth were also included in the models, as they were significantly associated with both case-control status and exposures.
Trend for breastfeeding duration, age at start of day care, and birth order were investigated. In line with validation studies, which showed that mothers tend to round reported breastfeeding durations, the dose-response relationships with breastfeeding duration were estimated with cutoffs centered on digits 3, 6, 9, and 12 months. The analysis of age at the start of day care was also undertaken with cutoffs centered on these rounded values. The tests for trend were computed from categorical variables. The subjects of each class of the categorical variables were assigned the median value of that class. Deviation from linearity was tested by a likelihood ratio test, comparing the model having the quantitative variable with that having the categorical variable. If linearity was not rejected, the P value of the trend was obtained by testing the slope of the quantitative variable.
Stratified analyses were also conducted to investigate the association between ALL and each of the exposures of interest by the strata of the other exposures. P values for the interaction between each pair of variables were estimated in the logistic models by using the Wald χ statistic.
Subgroup analysis The analyses were performed for B-cell and T-cell ALL subtypes by using polytomous logistic models and by age (2–5 years corresponding to the peak of incidence and 6–14 years).
Sensitivity analysis The robustness of the results was tested by excluding each study in turn and then 2 studies in turn. The analyses were also repeated after adjustment for socioeconomic status instead of maternal education, as well as for age at start of day care and breastfeeding duration, after consideration of alternative categorizations. For each exposure of interest, the potential for participation bias was investigated by estimating the difference in participation between exposed controls and unexposed controls that would have generated an odds ratio of the magnitude observed, under the assumption of no true effect, and assuming no difference in participation between exposed and unexposed cases.
All of the studies were approved by institutional ethics committees, and informed consent was provided by all participants.
Methods
Data from 11 Childhood Leukemia International Consortium case-control studies conducted in 8 countries from 1980 to 2010 were used (Table 1): Australian Study of Causes of Acute Lymphoblastic Leukemia in Children (AUS_ALL); Quebec Childhood Leukemia Study, Canada (CA_QCLS); Adele Study, France (FR_ADELE); Electre Study, France (FR_ELECTRE); Epidemiologic Study on Childhood Cancer and Leukemia, France (FR_ESCALE); Nationwide Registration for Childhood Hematological Malignancies, Greece (GR_NARECHEM); Study on the Etiology of Childhood Lymphohematopoietic Malignancies, Italy (IT_SETIL); New Zealand Childhood Cancer Study (NZ_NZCCS); United Kingdom Childhood Cancer Study (UK_UKCCS); Children's Oncology Group Study, United States (US_COG15); and Northern California Childhood Leukemia Study (United States) (US_NCCLS). The study design and participant characteristics for each study have been summarized previously.
Data Collection
All the data were collected with questionnaires that were administered to the parents face-to-face (FR_ADELE, GR_NARECHEM, IT_SETIL, NZ_NZCCS, UK_UKCCS, US_NCCLS) or by telephone (CA_QCLS, FR_ESCALE, US_COG15) or were self-administered (AUS_ALL, FR_ELECTRE). The questionnaires included information on demographic and socioeconomic characteristics and factors potentially associated with childhood leukemia. For the purpose of the present analysis, birth order, number and age of siblings, breastfeeding, history of common infections in the first year of life, and day-care attendance of the index child were provided by the investigators, as were sex, age at diagnosis or recruitment, and any other variables used for matching, as well as parental age at child's birth, parental education, and other indicators of socioeconomic status. All the studies included both B-cell and T-cell ALL.
Data Harmonization
Parental Education and Socioeconomic Status. Maternal and paternal levels of education were classified by using the same categories in all the studies: none or primary education, secondary education, and tertiary education (university). A heterogeneous 3-class (low, medium, high) indicator of socioeconomic status was also derived from the deprivation index based on address at diagnosis or interview (UK_UKCCS), household income (AUS_ALL, CA_QCLS, US_COG15, US_NCCLS), parental professional status (FR_ADELE, FR_ELECTRE, FR_ESCALE, GR_NARECHEM, NZ_NZCCS), or maternal education (IT_SETIL), depending on the data available (Web Table 1 http://aje.oxfordjournals.org/content/181/8/549/suppl/DC1 available at http://aje.oxfordjournals.org/).
Breastfeeding. Breastfeeding was classified by using the ever/never variable provided by the investigators. Children breastfed 1 month or less were also classified in the never breastfed group in some sensitivity analyses. The duration of breastfeeding was available for all the studies.
Day Care. Information on day-care center attendance was available for all the studies, and age at start of attendance was available for 10 studies (Web Table 1 http://aje.oxfordjournals.org/content/181/8/549/suppl/DC1). Attendance was considered full-time when attendance was at least 6 half-days per week, except in the French and Italian studies, which reported the frequency of attendance as full-time or part-time with no further details. Care by a child minder was also available in 5 studies.
Early Common Infections. History of common infections in the first year of life, as reported by mothers, was available in 8 studies (FR_ADELE, FR_ELECTRE, FR_ESCALE, GR_NARECHEM, IT_SETIL, NZ_NZCCS, UK_UKCCS, US_NCCLS). Web Table 1 http://aje.oxfordjournals.org/content/181/8/549/suppl/DC1 shows the sites of infections that were collected in the different studies. Four studies provided the total number of episodes for each site and, in the 3 French studies, a 3-class variable was available: no infection for the given site, between 1 and 3 episodes, and 4 or more episodes. The history of ear, nose, and throat surgery before age 3 years was available in 5 studies.
Statistical Analysis
The analyses were restricted to children aged at least 2 years, first, to ensure that all the cases and controls had had the opportunity of having been breastfed for a prolonged period, of day-care attendance, and of contracting infections in their first year of life; and second, because common infections occurring before 1 year of age may have been related to a prediagnostic phase of the disease in the ALL cases aged less than 2 years at diagnosis.
Meta-analysis. Meta-analyses based on study-specific odds ratios were conducted for the main exposures of interest. The odds ratios were estimated by using either unconditional or conditional logistic regression, depending on the design of each study, and including study-specific matching variables in the models. The sociodemographic characteristics significantly associated with both case-control status and the exposure were also included in the study-specific models. Between-study heterogeneity was assessed by using Cochran's Q and I statistics. Summary odds ratios and 95% confidence intervals were implemented by using the inverse variance method, with random effects in the event of heterogeneity.
Pooled Analysis. Pooled odds ratios were estimated from individual data by unconditional logistic regression systematically adjusted for age, sex, and study. Maternal education and maternal age at the child's birth were also included in the models, as they were significantly associated with both case-control status and exposures.
Trend for breastfeeding duration, age at start of day care, and birth order were investigated. In line with validation studies, which showed that mothers tend to round reported breastfeeding durations, the dose-response relationships with breastfeeding duration were estimated with cutoffs centered on digits 3, 6, 9, and 12 months. The analysis of age at the start of day care was also undertaken with cutoffs centered on these rounded values. The tests for trend were computed from categorical variables. The subjects of each class of the categorical variables were assigned the median value of that class. Deviation from linearity was tested by a likelihood ratio test, comparing the model having the quantitative variable with that having the categorical variable. If linearity was not rejected, the P value of the trend was obtained by testing the slope of the quantitative variable.
Stratified analyses were also conducted to investigate the association between ALL and each of the exposures of interest by the strata of the other exposures. P values for the interaction between each pair of variables were estimated in the logistic models by using the Wald χ statistic.
Subgroup analysis The analyses were performed for B-cell and T-cell ALL subtypes by using polytomous logistic models and by age (2–5 years corresponding to the peak of incidence and 6–14 years).
Sensitivity analysis The robustness of the results was tested by excluding each study in turn and then 2 studies in turn. The analyses were also repeated after adjustment for socioeconomic status instead of maternal education, as well as for age at start of day care and breastfeeding duration, after consideration of alternative categorizations. For each exposure of interest, the potential for participation bias was investigated by estimating the difference in participation between exposed controls and unexposed controls that would have generated an odds ratio of the magnitude observed, under the assumption of no true effect, and assuming no difference in participation between exposed and unexposed cases.
Ethics
All of the studies were approved by institutional ethics committees, and informed consent was provided by all participants.
