No butts about it! Parental smoking is not a major risk factor for childhood cancer

Numerous studies have included investigations of the potential role of maternal and paternal smoking as a risk factor for childhood cancer. While some have focused on childhood cancer overall, many of the more recent studies have explored specific associations with leukemia, lymphoma, or central nervous system tumors. Boffetta et al [Env Health Perspectives, 108:73-82, 2000] have reviewed over 30 published reports that examined parental smoking and childhood cancer. In addition, they conducted meta-analyses among studies when feasible. Four cohort studies were identified that explored an association between maternal smoking and childhood cancer. Cohort studies (where a participant=s exposure is characterized prior to the development of disease) are often considered superior to case-control studies since they are free of recall bias (e.g., parents under- or over-reporting their smoking habits because their child was diagnosed with cancer). Of these four studies, one found a statistically significant positive association (OR=2.5, 95% CI=1.2-5.1); one study found a weakly positive association (OR=1.3, 95%CI=0.8-2.2), one a weakly negative association (OR=0.7, 95% CI=0.4-1.2), and the other no association (OR=1.0, 95%CI=0.8-1.3). Of eight case-control studies that examined maternal smoking and overall risk of childhood cancer; two studies reported a statistically significant positive association with an observed dose-response (one of which was limited by selection of diabetic children as controls), another study found a non-significant positive association, while the remaining five studies reported no clear relationships. Overall, the meta-analysis of these 12 studies combined provided an estimated relative risk (RR) of 1.10 (95% CI 1.03-1.19).

Specific malignancies were also examined. For childhood leukemia or lymphoma, with the exception of two studies, there was no strong association with maternal cigarette smoking. The results of the meta-analysis produced an overall OR of 1.13 (95% CI=0.85-1.49) for NHL or all lymphoma (6 studies), and 1.05 (95% CI=0.82-1.34) for leukemia (eight studies). Moreover, of the 12 case-control studies that examined maternal smoking and CNS tumors, no statistically significant effect was observed (summary OR=1.04, 95% CI=0.92-1.18). When the authors conducted a meta-analysis for paternal smoking and risk of CNS tumors, the data suggested a modest positive association (OR=1.22; 95% CI=1.05-1.40; 10 studies). The data concerning paternal smoking and the development of NHL (OR=2.08; 95% CI=1.08-3.98, 4 studies) and ALL (OR=1.17, 95% CI=0.96-1.42; 4 studies) also suggested a positive relationship, although these risk estimates are based on fewer studies. No conclusive relationships could be determined for other malignancies given the small number of studies. Moreover, the authors were not able to adequately address such issues as timing of the exposure (e.g., preconceptional, prenatal, postnatal) or modification by age due to the limited number of studies that specifically explored these issues. The authors conclude that the data suggest a weak association between parental smoking (in particular, the father) and childhood cancer. 

COMMENT: Some of the criteria epidemiologists use to determine whether a causal association exists is a) biological plausibility (does it make biological sense?); b) dose-response (does more exposure causes more disease?); c) strength of the association (are the odds ratios or relative risks notably high?); d) temporality (does the exposure come before the disease); and e) consistency (is there some consistency in the relationships observed across studies conducted in similar populations?). This latter criterion is often addressed by conducting a comprehensive review of all data and, when feasible, conducting a meta-analysis --- where data across different studies are combined to produce an overall estimate of the relative risk. It should be noted that prior to conducting the meta-analyses, Bofetta et al tested for one aspect of publication bias by performing a statistical test that measures the precision of each study in estimating a true effect by taking into account its sample size [Egger et al, BMJ 315:629-34, 1997]. Overall, they reported >no evidence of publication bias=. However, it is still unknown how many studies that found no association were never published; therefore, this cannot be interpreted as no evidence of publication bias.

Nonetheless, this was a very useful paper by Boffetta and colleagues, given the controversy surrounding the potential carcinogenic effects of cigarette smoke on fetuses and young children. However, given the preponderance of the evidence presented here, it is unlikely that parental (especially maternal) cigarette smoking increases the risk of childhood cancer. It is difficult to understand the underlying biology that would explain a more powerful association with paternal smoking than maternal smoking, although it is possible that germ cell mutations which would develop before conception might explain this association. However, with the exception of identified genetic syndromes, no studies to date have suggested that childhood cancer is the result of a parental germ cell mutation. Finally, one caveat is that we cannot rule out the possibility that a genetically susceptible subgroup of children or mothers exists (e.g., children who may not have the capacity to detoxify harmful components of cigarette smoke) who may be at an increased risk.  Julie A. Ross
 

Genetic susceptibility in childhood acute leukemia

N-acetyltransferases (NAT1) and (NAT2) metabolize aryl- and heterocyclic amines found in tobacco smoke, medications, pesticides, dyes, antioxidants, and certain medications, some of which have been associated with childhood leukemia. NAT1 and NAT2 genotypes correlate with observed phenotypes in individuals, and range from slow to fast acetylation. Importantly, susceptibility to certain malignancies has been associated with both phenotypes, which could reflect differences in the activation of certain carcinogens in target organs. For example, fast acetylation genotypes have been associated with lung and colorectal cancer, whereas slow acetylation genotypes have been associated with bladder cancer. In this study, Krajinovic et al [Cancer Epi Biom Prev, 9:557-562, 2000] explore DNA variants of NAT1 and NAT2 in 176 French-Canadian children who have acute lymphoblastic leukemia and a randomly selected control group (n=306) from an institutional DNA bank. They examined five polymorphisms present in NAT1 and five polymorphisms present in NAT2 using allele-specific oligonucleotide hybridization and compared the distribution of these genotypes between cases and controls. Two of the NAT2 genotypes (NAT2*5C and NAT2*7B) were over-represented in the case group, while the NAT2 genotype, NAT2*4, was under-represented in the case group. When all of the possible NAT2 genotype combinations were stratified into either slow or fast activity, the authors found a statistically significant risk of childhood ALL associated with slow acetylation (OR=1.5, 95% CI=1.0-2.2). There was no independent effect of NAT1 acetylation genotype on ALL risk. Further, the authors combined these results with their previous findings that reported other risk-elevating genotypes (including glutathione-s transferase m-1 null and Cyp1A1*2A  wildtype or heterozygote) in this same case group [Krajinovic et al, Blood 93:1496-1501, 1999]. They found that the combination of these >unfavorable= genotypes (i.e., slow acetylator, null GSTM1, CYP1A1*2A wildtype or heterozygote compared with rapid acetylator, GSTM1, CYP1A1*2A absence) was more predictive of risk than each taken independently. 

COMMENT: Explorations of  gene-environment and gene-gene interactions characterize the field of molecular epidemiology. These types of studies may be very important, as they attempt to address questions of susceptibility (i.e., given the same level of exposure, why are only certain individuals at risk?)  One of the difficulties in conducting these studies (particularly for childhood cancer) is the potential lack of statistical power due to diminishing numbers of cases in stratified cells.  For example, in the instance above, the authors explored the combination of several different types of genes and their association with ALL, which is commendable.  However, the statistical ability to tease out a specific genotype combination that might reflect risk among several possible combinations in 176 cases is marginal (at best).  Nevertheless, these studies need to be done and are an appropriate step in the right direction. In particular, with the power of meta-analyses (statistically combining results of studies on the same topic), characterization of genotypes that may be important in the development of childhood cancer is possible.  Julie A. Ross
 

Increased incidence of precursor B-cell ALL?

The childhood age peak in ALL (primarily accounted for by precursor B (pre-B) cell ALL) that occurs between the ages of 2 and 5 has been noted in developed countries since the early part of the last century.  It is unknown why this age peak in incidence occurs, although theories suggest potential roles for infectious agents. In a brief report to the Lancet [356:485-486, 2000], McNally and colleagues in the United Kingdom note a temporal increase in the incidence of childhood ALL (n=435 cases) during the period 1980-1998, of which nearly 80% of were pre B-cell ALL. They also analyzed four separate time periods (1980-84, 1985-89, 1990-94, and 1995-98) and stratified the data by leukemia type (pre-B cell versus other), age and sex.  They found that the annual percent increase (API) during the overall time period was significant for children diagnosed with pre-B cell ALL between the ages of 1 and 4 (API=3.0%, p=0.024). Moreover, the API was higher for girls (API=3.6%, p=0.08) in this age group than for boys (API=2.6%, p=0.15). The authors speculate that these temporal increases in pre-B cell ALL may be a) due to the result of the same underlying cause(s) that initially produced the age peak, or b) a causal factor that is distinct from the factor producing the age peak. The authors also suggest that the gender differences may be due to differences in genetic susceptibility (e.g., HLA haplotypes).

COMMENT: These types of studies are entirely descriptive and are hypothesis generating in nature. It is possible that the factors the authors raise could explain some of the observations reported here. However, it must be recognized that analyzing small numbers in stratified categories, such as age at diagnosis and sex, reduces precision and can contribute to instability. For example, the number of females diagnosed with pre-B cell leukemia actually remained relatively constant during three of the four time periods, only increasing in the last time period of 1995-98. Moreover, for boys, the increase was also only apparent in the 1990s. Therefore, this may be a newer phenomenon, or it may reflect variability related to analysis of small numbers. It will be important for other registries to examine data in a similar manner to see if these trends can be confirmed. Julie A. Ross
 

C3 Quarterly Newsletter
Children's Cancer Research Fund
Epidemiology Research Unit
Division of Pediatric Epidemiology
Clinical Research
University of Minnesota
420 Delaware St. SE, Box 422
Minneapolis, MN 55455
pedsepi@umn.edu

Editors: 
Stella M. Davies, MD, PhD, and Julie A. Ross, PhD