Childhood Cancer Epidemiology [M.Tevfik DORAK]

In the US, the incidence of childhood cancer overall is approximately 125 per million persons (childhood is usually defined as 0-14 years of age). Between 1 in 600 and 1 in 500 children in Europe develop a malignant disease before the age of 15 years (Parkin et al, 1998). The total incidence of childhood cancer varies rather little between different regions of the world, with cumulative risk to age 15 nearly always in the range 1.0 -2.5 per thousand (Stiller & Parkin, 1996). The most comprehensive source of systematic information concerning childhood cancer incidence is the 8th Volume of IARC’s Cancer Incidence in Five Continents series (Parkin et al, 2002). Data from USA (Linabery & Ross, 2008, CDC-MMWR, 2007; SEER 1975-2000 / PDF; SEER 1975-2012 in Siegel, 2016; Ward, 2014), Europe overall (Steliarova-Foucher, 2004; Coebergh, 2006), UK (Shah & Coleman, 2007), Germany (Spix, 2007), France (Lacour, 2010), Greece (Petridou, 2008), Italy (Magnani, 2003; Dalmasso, 2005), Spain (Peris-Bonet, 2010), Scandinavia (Svendsen, 2007), Sweden (Lannering, 2009), Brazil (de Camargo, 2010), Shanghai, China (Bao, 2009), Japan (Baba, 2010), Australia (Milne, 2008) and New Zealand (Dockerty, 1996; Dockerty, 2009) have been presented separately. Although cancer has an annual incidence of only about 150 new cases per 1 million U.S. children, it is the second leading cause of childhood deaths (following accidents) (Jemal, 2008). Notable ethnic differences in incidence are also documented (Goodman, 1989; Zahm, 1995; McNeil, 2002; Ross & Swensen, 2000; Parkin et al, 1998; Pratt, 1998; Stiller & Parkin, 1996; Dockerty, 1996; Reynolds, 2002; Kaiser, 2002; Lanier, 2003; Oksuzyan, 2015). The incidence of childhood cancers also varies by age group (Statbite, JNCI 2004). The analysis of SEER data shows an annual increase of 0.5% in childhood cancer overall from 2004 to 2008 as the continuation of a trend since 1975 (Siegel, 2012; see also Mangano, 1999, Linet, 1999; Linabery & Ross, 2008).

In children, about half of all cancer cases diagnosed before 15 years of age actually occur below the age of 5 years (Stiller & Draper, 2005; Page and Ben-Eliyahu, 1999). Whereas adult cancers are overwhelmingly carcinomas, in children, hematological malignancies and tumors of the central nervous system account for the majority of all cancers. Leukemias make up most (approximately 25%) childhood cancers. Leukemias (of which nearly 80% are ALL) are followed in frequency by tumors of the CNS (20%; unlike adult brain tumors, most childhood brain tumors are in the posterior fossa), neuroblastoma (7%), non-Hodgkin lymphoma (NHL; 6%), Wilms tumor (6%; virtually all kidney tumors), Hodgkin lymphoma (formerly Hodgkin’s disease; 5%), rhabdomyosarcoma (3%), retinoblastoma (3%), osteosarcoma (3%), and Ewing sarcoma (2%). Like NHL, Hodgkin lymphoma is associated with immunodeficiency and Epstein-Barr virus (EBV) as well as cytomegalovirus (CMV) and human herpesvirus 6 (HHV6).

We know much less about risk factors for childhood cancer than adult cancers because of their rarity and extreme diversity. Only 1 in 100 new cancers is a childhood cancer. Most data are on childhood leukemia because it is the most common one.

Childhood Cancer Classification

Childhood neoplasms are classified, as is customary in childhood cancer studies, according to a special scheme, based mainly on morphology and tissue of origin rather than on anatomical site, as in the International Classification of Diseases (ICD) (ICD-O-3 Manual; ICD -10 Neoplasms Section Online; International Classification of Childhood Cancer - ICCC). Data on cancers in adults are normally presented by ICD site code.

Most childhood cancers are classified in the following broad categories:

- Leukemias (SEER)

ALL, AML, CML and others

- Lymphomas and reticuloendothelial neoplasms (SEER)

Hodgkin lymphoma and non-Hodgkin lymphoma (NHL)

- Central nervous system (intracranial and intraspinal) neoplasms (SEER)

Astrocytoma and other gliomas (especially brain stem), primitive neuroectodermal tumors

(PNET), Medulloblastoma (cerebellar PNET) and others

- Sympathetic nervous system tumors (SEER)

Neuroblastoma, ganglioneuroblastoma and others

- Retinoblastoma (SEER)

- Renal tumors (SEER)

Wilms tumor (nephroblastoma), renal cell carcinoma and others

- Hepatic tumors (SEER)

Hepatoblastoma, hepatocellular carcinoma

- Malignant bone tumors (SEER)

Osteosarcoma, Ewing sarcoma

- Soft tissue sarcomas (SEER)

Rhabdomyosarcoma, fibrosarcoma and others

- Germ cell, trophoblastic and other gonadal neoplasms (SEER)

Gonadal (testicular/ovary) germ cell tumors and others

- Malignant epithelial neoplasms (carcinomas) (SEER)

Thyroid carcinoma, malignant melanoma and others

- Other unspecified malignant neoplasms

Note that Langerhans cell histiocytosis (histiocytosis X) is no longer considered as cancer. See also OncoLink University: Pediatric Cancers.

General Features of Childhood Cancer

The gender effect (or more appropriately, sex effect) in incidence of childhood cancer is well-established and consistent worldwide (Ashley, 1969; Greenberg & Shuster, 1985; Linet & Devesa, 1991; Little J, 1999; Pearce & Parker, 2001; Desandes, 2004; Forsythe, 2010; Linet, 2016). Among newly diagnosed childhood cancers, the incidence rates for all cancers yields a boys to girls ratio of 1.2. The male predominance is a feature of cancer incidence in all ages (Cartwright, 2002; Boyle & Ferlay, 2005 & 2007; Cook, 2009; Edgren, 2012). Infant leukemia is more frequent in females; carcinomas (thyroid carcinoma and malignant melanoma) and alveolar soft part sarcoma (Bu, 2005) tend to occur more frequently in young females. Interestingly, the striking male-to-female ratio in Hodgkin lymphoma in younger ages reverses for adolescents (Spitz, 1986). Not just primary susceptibility but late-effects also show sex effect (associations between female sex and cognitive dysfunction after cranial irradiation, cardiovascular outcomes, obesity, radiation-associated differences in pubertal timing, development of primary hypothyroidism, breast cancer as a second malignant neoplasm and suggests an increased prevalence for the development of osteonecrosis among females) (Armstrong, 2007; Jain, 2009).

Just a fraction of all cancers can be attributed to a single defective gene usually causing familial clusters of disease. Common allelic variants of susceptibility genes with relatively low penetrance account for a higher percentage of cancers (and other chronic diseases). Cancers arising as a result of highly penetrant mutations associated with hereditary cancer syndromes (e.g. retinoblastoma, Li-Fraumeni syndrome and certain congenital overgrowth syndromes such as Beckwith-Wiedemann syndrome; Simpson-Golabi-Behmel syndrome) are unlikely to account for more than 15% of all cases (Anderson, 2000). A more likely situation is that cancer in a child develops as a result of exposure to a risk factor in a genetically susceptible individual (Birch, 1999). However, environmental agents may be the major etiological factor in some childhood cancers with stronger effects in genetically susceptible individuals (gene-environment interaction). From an epidemiologic point of view, we cannot change a person's genetic architecture for prevention but genetic studies help to identify modifiable environmental factors more easily. Whenever ethically possible, identification of carriers of such genetic markers can then be monitored more closely for environmental exposures.

Children are not simply small adults and there are special considerations in risk assessments for childhood cancers (Kim, 2006). The major difference between children and adults, which is relevant to their susceptibility to cancer, is the fact that children are developing and growing. Exposure to environmental agents, which result in mutations in genes involved in control of cell division, apoptosis, growth and differentiation, can result in more profound consequences in a child, than a similar exposure in an adult. Infants and children differ from adults in their exposure both qualitatively and quantitatively, in part because they eat more food, drink more water, and breathe more air per unit of body weight than adults do (Roberts RJ, 1992), and the activity patterns of children further increase their exposure to environmental agents. For example, children are engaged in more physical activity, play close to the ground, and engage in characteristic hand-to-mouth behavior, resulting in high opportunity of exposure to toxicants such as pesticides, radon, and particulate matter (Carroquino, 1998). For ionizing radiation where this has been studied to some extent, it was concluded that in general children appear to be at somewhat greater risk for the same dose than adults. In addition, studies of transplacental carcinogenesis in experimental animals using DNA alkylating agents have consistently demonstrated a much greater susceptibility of the fetus than adult animals administered the same dose. This greater susceptibility could be predicted for physiological reasons since the fetus and young child are undergoing multiple processes of growth and differentiation. The number of cell divisions per unit of time is considerably greater in the fetus and young child than in adults. In summary, increased susceptibility in children has both biological and behavioral reasons (Wild & Kleinjans, 2003).

Fetuses and children may be more sensitive to carcinogenic agents than adults (Kim, 2006). Although direct evidence is lacking, there are indirect arguments pointing towards a possible higher sensitivity of the fetus, the infant and the child to carcinogens (Tomatis, 1989; Anderson, 2000; Perera, 2002; Birnbaum & Fenton 2003; Landrigan, 2003; Bockskay, 2005) due to higher exposure and immature biological response. Heightened fetal susceptibility could result from higher rate of cell proliferation and differentiation, greater absorption or retention of xenobiotics, and/or less efficient detoxification, DNA repair, or apoptotic mechanisms (Perera, 2006). For example, cancer cells have been shown to grow and metastasize more easily in very young animals (Page and Ben-Eliyahu, 1999). During intra-uterine development and childhood the number of cell divisions per unit of time is much greater than in adulthood, and the mutant frequency rises rapidly. In mice, about one third of mutations arise before birth, about one third during growth to adulthood and the remaining third during the rest of the animal's life (Paashuis & Heddle, 1998). Reduced detoxification and repair capabilities result in a greater susceptibility of the fetus to DNA damage (Whyatt, 2001). Anderson, 2000 & 2004 and Birnbaum, 2003 have reviewed the literature on carcinogenesis in offspring with environmental exposures to known carcinogens, including endocrine disruptors, during gestation. In some cases, the number of tumors (incidence or multiplicity) in exposed offspring was greatly increased after early life-stage exposure; in others, the time to spontaneous tumors (latency) was decreased by exposure. Because direct carcinogens would consistently induce tumors in exposed offspring, the timing of developmental exposure (critical window of exposure) is highly significant (Kim, 2006). There is evidence fromanimal and human epidemiologic studies of causal relationshipsfor preconception, in utero, perinatal, infancy, and postinfancyexposures and cancer occurrence in children (Selevan, 2000; Anderson, 2000; Olshan, 2000; Birnbaum, 2003). One example isthe statistical association between prenatal exposure to diagnosticradiographs, particularly during the last trimester of pregnancy,and subsequent small increase in childhood leukemia risk (Doll & Wakeford, 1997; Linet, 2003). Findings suggest that developmental exposures to endocrine disruptors play an important role in childhood cancer causation (Birnbaum, 2003; Landrigan, 2003).

With improving survival rates in childhood cancer, one important point is the late-effects of therapy (Stevens, 1998; Schwartz CL 1999 & 2003; Mertens, 2001). Currently, in some populations one in every one thousand young adult is a childhood cancer survivor (Hawkins, 1996). Their health status and quality of life will be affected by the cumulative late-effects of chemotherapy and radiotherapy they have received. Survivors of childhood cancer also live with 8% - 10% risk of having a second primary cancer within twenty years (Meadows, 1985; Hawkins, 1987, 2004, 2005; Neglia, 2001; Jenkinson, 2004; Robison, 2005; Maule, 2007; Hijiya, 2007 & Second Neoplasms in CCSS and Health Maintenance for Childhood Cancer Survivors). The latest reports from USA (Olsen, 2009) and UK (Reulen, 2011) suggest standard incidence ratios of 3.3 to 4.0 for secondary cancers in childhood cancer survivors. For colorectal cancer, the magnitude of risk observed in survivors treated with direct abdominopelvic irradiation is equivalent to the risk of those who have two first-degree relatives with colorectal cancer (Reulen, 2011).

Environmental Risk Factors in Childhood Cancer

There is no doubt that environment plays a major role in childhood cancer (CHEC Report; Zahm & Devesa, 1995; Wild, 2003; Linet, 2003; Miller, 2004; Bunin, 2004; Eden, 2010; Kaatsch, 2010). There are only few well-established environmental causes of childhood cancer. The earliest known exposures were radiation (Giles, 1956; MacMahon, 1962; Graham, 1966; Bithell, 1975; Harvey, 1985; McKinney, 1987; Doll & Wakeford, 1997), diethylstilbestrol (Herbst, 1971 & 1977 & 1981 & 1999; Gill, 1976 & 1977; Depue, 1983; Niculescu, 1985; Anderson, 2000) and chemotherapeutic agents (Table in Stolley, 1995). The studies of Japanese children who were exposed to atomic bombradiation have found that these children, as they grew, hadhigher rates of childhood leukemia than unexposed children (Miller, 1987). Exposure to atomic bomb radiation in childhood also was associated with higher rates of adult cancers, including breast cancer (Wakeford, 1995; Miller, 1995). Leukemia development following preconceptional radiation exposure is not a universal feature but involves host susceptibility. It is estimated that for about 1% of the exposed persons who are affected by radiation, there is a 50-fold increase in the risk of leukemia in the offspring (Bross & Natarajan 1972 & 1977). Many other agents such as electromagnetic fields, pesticides, and some parental occupational exposures are suspected of playing roles, but the evidence is not conclusive at this time. Environmental factors other than the few major ones with strong effects increase the risk only slightly. It is very difficult to unravel those in population studies. One way to augment their effect and to detect them more easily is to run gene x environment interaction studies, where multiplicative interactions will be easily identified even in small case-control studies. Most genetic associations may also be a product of interactions with unknown environmental factors and may get stronger if the data could be stratified for exposure levels.

Non-ionizing radiation is the result of electromagnetic fields with frequencies too low for their radiation energy to cause breaks in chemical bonds but it can still cause molecular damage. It encompasses a wide variety of electromagnetic fields, including static fields, power frequency fields, radio frequency fields, and UV radiation (for a review, see a CureToday article by L Weber). These fields are ubiquitous in modern society: examples of sources of exposure are production and use of electric power, electronic surveillance systems, wireless communications, tanning machines, and solar exposure. The most visible source to extremely low frequency (ELF) magnetic fields is arguably power lines and in particular high voltage transmission lines. There may be an increased risk of leukemia in the group of children at highest exposure (See PowerLine Health Facts Website). This was first pointed out in 1979 (Wertheimer & Leeper, 1979; Fulton, 1980). Initial studies found that children with leukemia were more than twice as likely as their healthy counterparts to have lived in homes near high voltage power lines. For many reasons it has been difficult to establish the association with leukemia in a single study. Combining results in a ‘pooled analysis' has overcome some of these difficulties and showed that that long-term exposure in excess of 0.4µT is associated with a doubling (95% confidence limit: 1.27 to 3.13) of leukemia risk in children (Ahlbom, 2000, Greenland, 2000). Further meta-analyses have concluded that the data supporting residential electromagnetic fields as the principal risk factor are suggestive with small to moderate effect sizes but inconsistent (Wartenberg D, 1998 & 2001). The UK Childhood Cancer Study carried out in the 1990s showed a trend of increasing incidence with increasing exposure but the study was not powerful enough to provide a statistically significant result (UKCCS, 1999). (See also Bristol University Human Radiation Effects Group Website for more details). However, Draper et al. have published a very large case-control study and concluded that a child's risk of leukemiaincreased steadily with proximity to high voltage power linesof the home they lived in at birth and the distances up to 600m may increase the risk (Draper, 2005). See also a book published by the National Academy of Sciences (USA): Possible Health Effects of Exposure to Residential Electric and Magnetic Fields (1997).

On the strength of the epidemiological data to date, the UK National Radiological Protection Board (NRPB), the International Agency for Research on Cancer (IARC) and the US National Institute of Environmental Health Sciences (NIEHS) have independently concluded that electric and magnetic fields are possible human carcinogens (group 2B). NRPB guidelines now caution against prolonged exposure at 0.4µT. Exposure to fields of 0.4µT is relatively uncommon, affecting perhaps 0.5 per cent of children. One centre that is involved heavily in this research is at the University of Bristol (EMF Research at Bristol University, UK). See also a presentation on Childhood Leukemia & EMF & California EMF Program - Risk Evaluation Report.

Dietary exposure to N-nitroso compounds, heterocyclic amines (HCA) and polycyclic aromatic hydrocarbons (PAH) are risk factors for certain adult cancers (Sugimura, 2002; Key, 2004; see also AICR Brochures: Facts About Grilling). Studies in experimental animal systems have shown that N-nitroso compounds are powerful transplacental neurocarcinogens (Rice et al, 1989). Nitrosamines and nitrosamides can be synthesized endogenously in the stomach from nitrate or nitrite and secondary or tertiary amines and amides that are found as common dietary constituents. The reaction is inhibited by vitamin C (Mirvish, 1981). These experimental observations lead to the hypothesis that nitrosamines and nitrosamides synthesized from relevant food stuffs ingested by mothers during pregnancy might act as human transplacental carcinogens and that a high intake of fruit and vegetables or vitamin C supplements may be protective.

A number of epidemiological studies, focused on childhood brain tumors, have been carried out to test this hypothesis. The main focus of these studies has been dietary intake of cured meats that have high nitrite content. These studies also included estimates of dietary intake of fruit, vegetables and vitamin supplements (Kuijten, 1990; McCredie, 1994; Preston-Martin, 1996; Bunin GR, 2000; Pogoda & Preston-Martin, 2001). Blot et al (1999) critically reviewed 14 such epidemiological studies including 13 case-control studies. The majority of these showed no significant association between total cured meat intake and childhood brain tumor or other childhood cancer risk but more studies found a positive association than a negative. A meta-analysis provided further support for the suspected causal association between ingestion of N-nitroso compounds from cured meats during pregnancy and subsequent childhood brain tumor in offspring (Huncharek, 2004). Besides brain tumors, N-nitroso compounds (excessive hot dog intake) are also implicated in childhood leukemia (Peters, 1994).

Atmospheric pollution may also contribute to childhood cancer development. A study from the UK reported associations with proximity of birth places to sites of industrial combustion, volatile organic compound (VOC) uses, and associated engineexhausts, 1,3-butadiene, dioxins and benz(a)pyrene (Knox, 2005). In reports from the ESCALE study, the Clavel group found associations of childhood acute leukemia (both ALL and AML) with with the indicators of proximity and density of heavy-traffic roads such as traffic NO₂ concentration in the residence (Brosselin, 2009; Amigou, 2011). Recent studies by Heck et al in California have reported correlations of exposure to traffic-related air pollution (Heck, 2013; Ghosh, 2013) and ambient air toxics (Heck, 2013) during pregnancy and early childhood with childhood cancer risk.

Initiated by a hoax e-mail and spread as an urban legend, aspartame has been blamed on many adverse health effects including (brain) cancer without any scientific evidence (see Green Facts on Aspartame). Both NCI and ACS have issued statements on aspartame and cancer connection (i.e., lack of it). A small case-control study found no association between aspartame consumption and childhood brain cancer (Gurney, 1997). No other formal epidemiologic studies have evaluated whether aspartame is associated with brain cancer or any other cancers in humans (see also Weihrauch & Diehl, 2004; Gallus, 2007; Bosetti, 2009; Marinovich, 2013).

Epidemiological studies have reported associations between childhood cancer, and either parental or child exposure to pesticides (Ma, 2002; reviewed in Daniels, 1997 and Zahm & Ward, 1998). Collectively, these studies suggest an increase in risk of brain cancer, leukemia, non-Hodgkin lymphoma, Wilms tumor, Ewing sarcoma, and germ cell tumors associated with parental occupational and non-occupational exposure to pesticides. Certain pesticides are classified as endocrine disruptors and linked to childhood cancer development (Landrigan, 2003). Infante-Rivard (1999) found indoor use of some insecticides by the owners, and pesticide use in the garden and on interior plants, in particular frequent prenatal use, associated with increased risks for ALL in the offspring. Further study of a sub-sample of cases revealed a gene x environment interaction with an additional risk increase among cases with the CYP1A1m1 and with the CYP1A1m2 mutation. The role of genetic polymorphisms in environmental health issues has been reviewed and this issue will gain increasing importance in childhood cancer studies (Kelada, 2003). See also an essay at the Environmental Health Policy Institute by Joanne L Perron.

Overall, the conclusions of the reviews by Daniels et al (1997) and Zahm & Ward (1998) remain valid. Although an increase in risk of several childhood cancers has been reported for parental occupational and parental and childhood non-occupational exposure to biocides, the available evidence does not allow to conclude that a causal association has been demonstrated between any agent or group of agent and any neoplasm. The conclusion can be summarized as 'another 40 epidemiological studies (on the role of agrochemicals and biocides) similar to the majority of those conducted thus far will not provide clarity' (Olshan & Daniels, 2000).

Parental occupation and childhood cancer link associations may indicate associations with social class or parental occupational exposure. The earliest studies found only weak and inconclusive associations with childhood cancers (Kwa & Fine, 1980; Sanders, 1981; Arundel, 1986). Most positive associations concern hydrocarbon-related parental jobs and CNS tumors (Kwa & Fine, 1980; Johnson, 1987; Savitz & Chen, 1990). Other studies also reported several associations in childhood brain tumors (Wilkins & Koutras, 1988; Johnson & Spitz, 1989; Cordier, 1997 & 2001; Colt & Blair, 1998). A relatively consistent association has been found between paint, benzene, wood dust and radiation exposures of parents and leukemias (Savitz & Chen, 1990; McKinney, 1991; Colt & Blair, 1998) (the most common sources of benzene exposure for humans are gasoline filling stations, tobacco smoke, and vehicle exhaust fumes). The latest report from UKCCSG agrees with the inconsistencies and lack of a strong parental occupation effect on childhood cancer risk (McKinney, 2003).

Tobacco, Alcohol and Recreational Drugs

Tobacco smoking and alcohol drinking are classified as human carcinogens (IARC, 2002; IARC, 1988). Recreational use of drugs, such as marijuana and cocaine, is suspected to entail a cancer risk (Hall & MacPhee, 2002; Tashkin, 2001). Animal studies provided evidence that some genotoxic and carcinogenic compounds can pass the placenta to the developing fetus (Sasco & Vainio, 1999). Parental exposures to these agents were therefore hypothesized as risk factors for childhood cancer.

The conclusions of the comprehensive review by Little (1999) were that (i) there is no consistent association between maternal tobacco smoking during pregnancy and any type of the childhood cancers; (ii) there is a possible association between paternal smoking and ALL (see also Milne, 2012); (iii) there is a positive association between maternal alcohol consumption during pregnancy with AML; and (iv) there is a possible association between recreational use of drugs of parents and childhood leukemia and lymphoma. An analysis of Swedish nationwide cancer registry data suggested a protective association in ALL and risk associations in AML and NHL (Mucci, 2004).

That environmental exposure is indeed important in the causation of childhood cancers is implicated by observations on associations of enzyme polymorphisms with risk of childhood cancers (Infante-Rivard, 1999; Krajinovic, 1999; Alves, 2000; Davies, 2000; Infante-Rivard, 2000; Sinnett, 2000; Saadat & Saadat, 2000; Krajinovic, 2000; Wiemels, 2001; Ezer, 2002; Davies, 2002; Krajinovic, 2002a; Krajinovic, 2002b; Robien & Ulrich, 2003; Krajinovic, 2004; Canelli, 2004). Exogenous chemicals are excreted from the body after metabolic conversion by enzymes mediating oxidation activation (Phase I) and conjugation detoxification (Phase II). For several childhood cancers risk is modified by genetic characteristics affecting these enzymes and the activation or inactivation of exogenous chemicals. This holds also for acute lymphoblastic leukemia (ALL), the most common pediatric cancer. The risk for ALL increased to an odds ratio of 3.3 for children who presented three such genetic traits (Sinnett, 2000). This implicates that environmental exposure is important in the causation of childhood cancers.

Topoisomerase II Inhibitors

Dietary bioflavonoids, some of which are known DNA topoisomerase II (DNAt2) inhibitors, have been shown to cause site-specific DNA cleavage in the MLL gene breakpoint cluster region on chromosome 11q23 (Ross, 1994). It has been hypothesized that de novo infant leukemias (of which 80% have the MLL abnormality) may occur as a result of maternal exposure to agents in diet and medications that inhibit DNAt2. Maternal bioflavonoid intake during pregnancy has been associated with increased risk of infant leukemias (Ross, 1996). A COG study found that for AML(MLL+) cases, maternal consumption of specific DNAt2 inhibitors may increase risk (Spector, 2005). Another study of 83 children with ALL and 166 age- and sex-matched controls found an inverse association between maternal folate supplementation during pregnancy (folate supplementation may prevent some of the chromosome breakage leading to translocation formation) and risk of childhood ALL, but did not evaluate MTHFR polymorphisms (Thompson, 2001; but see meta-analyses by Dockerty, 2007 & Milne, 2010) . For a review of topoisomerase inhibitors and leukemia risk, see Lightfoot, 2004 & Ross, 2008.

Anemia

A UKCCS report showed an increased risk for AML if anemia (Hb < 10 gr) has been detected during pregnancy (OR = 2.6; 95% CI = 1.7 to 4.1) (Roman, 2005). This finding is in agreement with previous associations reported in leukemia (Roman, 1997; Petridou, 1997). In the interpretation of this association, difficulties in the diagnosis of anemia in pregnancy, its relationship with birth weight and possible interaction with folate intake may have to be considered (Steer, 2000).

Atopic Diseases and Childhood Cancer

Negative associations with asthma and other allergic diseases have been consistently reported in childhood leukemia (Nishi & Miyake, 1989; Magnani, 1990; Petridou, 1997; Schuz,1999 and 2003; Wen, 2000; Jourdan-Da Silva N, 2004; Rosenbaum, 2005), including the latest UKCCS study (Hughes, 2007) and in neuroblastoma (Greenberg RS, 1983). Only one relatively small study found a significant positive association between allergies and childhood leukemia and lymphoma (Manning & Carroll, 1957). There have been other studies which did not find an association in various types of childhood cancer (Little J. 1999). Although there is no study in childhood brain tumors, this negative association has been observed for adult gliomas but not in meningiomas (Wiemels, 2002; Brenner, 2002; Wrensch, 2002; Schwartzbaum, 2003). A general inverse relationship with cancer in adults has also been reported (Vena, 1985) (and in NHL (Grulich, 2005)). Hygiene hypothesis has been put forward as an etiologic possibility for both childhood leukemia (Smith, 1998) and allergies (Strachan, 1989; Romagnani, 2004). It is important to note that childhood allergies are more common in boys (until adolescence) (Almqvist, 2008) and since childhood cancers are also more common in boys, analysis of effect modification by sex by stratification is particularly important in the analysis of association studies between allergies and cancer in childhood. A meta-analysis showed protective associations of atopy/allergies, eczema, asthma and hay fever in childhood leukemia with substantial heterogeneity (Linabery, 2010).

The inverse association of allergic disorders with childhood leukemia is generally interpreted as the protective effect of a hyperactive immune system, thus this association lends support to the immune surveillance concept (Dorak, 2007; see also Cancer Genetics). The UKCCS study showed the same association (Hughes, 2007) but the observed reciprocal relationship between allergy and childhood ALL is interpreted as “a dysregulated immune response is a critical determinant of childhood ALL”. Particularly because the same UKCCS study has already shown a positive association with the number of infections and subsequent ALL risk (Roman, 2007; Roman, 2009), it is more likely that the results further strengthen the concept of immune surveillance in the development of childhood ALL, which is equally applicable to most other cancers (as suggested by the general inverse associations between allergy and cancer).

Autoimmune Diseases and Childhood Cancer

The original risk association of autoimmune disorders in the family and childhood leukemia (Till, 1979) has been confirmed by later studies (Woods, 1987; Buckley, 1989; Mellemkjaer, 2000; Perillat-Menegaux, 2003) and more specifically for childhood ALL (including a meta-analysis (Linabery, 2010)) and lymphoma. No such effect has been shown in infantile leukemia (Wen, 1998). Parallels have been shown between diabetes and childhood leukemia epidemiology (Feltbower, 2004). A Scandinavian study only noted an association between childhood cancer and maternal diabetes but no other autoimmune disorder (Westbom, 2002). Earlier studies, however, reported a link between multiple sclerosis and leukemia / lymphoma (Bernard, 1987). Specifically, an increased risk of ALL in children of women with MS has been observed (RR = 4.0; 95% CI = 1.3 to 9.3) (Buckley, 1989). Intriguingly, the genetic risk markers for multiple sclerosis in HLA-DRA, HLA-C and IFNG also modify the risk of childhood ALL with sex effect (Morrison, 2010).

Childhood Leukemia and Infections

Kinlen suggested that in rural areas, which experienced an influx of residents from other areas, a situation referred to as population mixing, there would be an increased level of contacts between susceptible and infected individuals (impaired herd immunity hypothesis). Non-exposed individuals living where the relevant infection is not endemic, subsequently become at risk when exposed by mixing with an incoming population of infected carriers. Kinlen tested these ideas in a number of areas and population groups in the UK and in each situation an increased relative risk (about 2-fold) of childhood acute leukemia, was observed in the period after the mixing had occurred (Kinlen, 1995). Subsequently similar patterns of incidence for childhood ALL were observed in independent studies outside of the UK; in Hong Kong (Alexander, 1997) and France (Boutou, 2002). Alternative measures of population mixing which analyzed the diversity and volume of migrants within small geographical areas in some instances found increased risks of leukemia with high levels of population mixing (Dickinson & Parker,1999; Stiller & Boyle, 1996) but in others high levels of population mixing were associated with lower risks (Parslow, 2002 & 2005; Law, 2003; see also Kinlen, 2004).

The second hypothesis was put forward by Greaves who proposed that the common subtype of ALL (precursor B-cell ALL; formerly c-ALL) arises as a result of two independent mutations (note that a similar two-stage model for other cancers has been known for a while (Moolgavkar & Knudson, 1981; Slaga, 1983; Armitage, 1985)). The first occurring in utero or soon after birth creates a pre-leukemic clone of cells. The second mutation arises after an average period of three years and precipitates the onset of disease. Greaves suggested that common childhood infections serve as a promoter of the second mutation and that delay in the normal pattern of exposure of the child’s immune system to infection increases the pool of susceptible cells and hence, increases the risk of the second critical mutation. Lifestyle changes associated with socio-economic development would favor delays in exposure of infants and young children to the common relevant infections (delayed infection hypothesis) (Greaves & Alexander, 1993).

The third hypothesis, due to Smith (Smith, 1997a; 1997b; Smith, 1998), attributes ALL to an in utero exposure to infection that occurs as a result of improved hygiene conditions, arising in parallel with increased socio-economic status (SES). Smith has suggested JC virus (JCV) a member of the polyoma family of viruses, as a candidate agent (Smith, 1998). However, polyoma virus sequences including JCV were not detected in children with precursor B-cell ALL but this does not rule out their involvement in the pathogenesis of ALL (Smith, 1999; MacKenzie, 1999; Priftakis, 2003).

All three hypotheses put forward that childhood ALL arises as a result of a rare outcome of a common infection but whereas Kinlen and Smith suggest the involvement of one or more specific agents, the Greaves model involves infections in general. The hypotheses lead to certain predictions about patterns of incidence and conditions favoring exposure to and transmission of the relevant infectious agents. Childhood ALL demonstrates space-time clustering, which is presumed to be characteristic of infectious diseases. Seasonality of onset has been shown for childhood ALL (Badrinath, 1997; Westerbeek, 1998; Ross, 1999; Feltbower, 2001; Karimi & Yarmohammadi, 2003) which may be consistent with an infectious origin (the same has also been shown with adult ALL (Badrinath, 1997). A review of earlier studies concluded that there is limited support for a seasonal pattern of onset of childhood leukemia with cases tending to occur during the first half of the calendar year (Greenberg & Shuster, 1985). Seasonality with a peak of birthdates in July or August has been reported in childhood HD with male-specificity (Fraumeni & Li, 1969) and a peak of diagnosis dates in March (Langagergaard, 2003). In animals, immune functions show changes with seasons which are attributed to day length changes (correlation with melatonin which has immunomodulatory effects) and adrenal steroid hormone fluctuations in response to winter stressors (Nelson, 1996). If such changes also occur in humans, they may contribute to the seasonality observed in childhood cancers. The better established seasonality in infectious diseases has been attributed to dark-light photoperiods and associated variation in melatonin levels too (Dowell, 2001). Other possible explanations for seasonality in infectious diseases may also help to better understand the same phenomenon in childhood cancers, if explored (Dowell, 2001; Grassly & Fraser, 2006; Altizer, 2006). Another confounder of the seasonality studies may be the correlation between the seasons and the sex ratio in animals if the same applies to humans (Clutton-Brock, 1986) since sex is a risk marker for childhood ALL (see Gender Effect in Cancer).

Viruses seem to be involved in a proportion (around 15%) of all cancers (zur Hausen, 1991; Klein, 2002). Burkitt lymphoma, Kaposi sarcoma, cervical cancer and hepatocellular carcinoma have all been linked to viruses. Infections early in life are thought to be related to childhood leukemia; the etiology, however, may be infections per se rather than specific agents. Only few papers have directly addressed the role of early common infections in acute leukemia. Most studies found negative associations (Van Steensel-Moll, 1986; Neglia, 2000; Perrillat, 2002; McKinney, 2002; Jourdan-Da Silva, 2004) except two studies that reported a risk association between early viral infections (in the first 6 and 12 months, respectively) and subsequent leukemia development (McKinney, 1987; Dockerty, 1999). The latter study was conducted in New Zealand and found an increased risk of leukemia in relation to reported influenza infection of the child during the first 12 months of life (adjusted OR = 6.8, 95% CI = 1.8 to 25.7). A study from Northern England found an association between exposure to measles around birth and Hodgkin lymphoma (Nyari, 2003). It has been hypothesized that CMV may be involved in Ewing sarcoma (Cope, 2000) and adenovirus in childhood ALL (Dorak, 1996) development. No direct data have been generated to support these postulates with the exception of limited data on the adenoviral connection (Fernandez-Soria, 2002; Kosulin, 2007). Most recently, adenovirus DNA was detected in Guthrie cards of childhood ALL cases more frequently than healthy children (Gustafsson, 2007; Honkaniemi, 2010).. Another study from Northern California, however, does not support this finding (Vasconcelos, 2008). Other studies examined parvovirus B19 (Kerr, 2003; Isa, 2004), polyomavirus-JCV (Smith, 1999; MacKenzie, 1999; Priftakis, 2003), polyomavirus-BKV (Smith, 1999; MacKenzie, 1999), polyomavirus-SV40 (Smith, 1999), herpesvirus (MacKenzie, 2001) with no positive finding. One study found a suggestive link between exposure to HHV-6 and childhood leukemias based on the presence of IgM and IgG antibodies in patients (Salonen, 2002). HHV-6 is a T-lymphotropic herpesvirus that is a common cause of acute febrile illnesses in young children, including exanthema subitum (roseola). It infects almost all children by the age of two years and persists lifelong (Dewhurst, 1997). HHV-6 is closely related to HHV-7 (Dewhurst, 1997). For a review of infections and childhood leukemia connection, see O’Connor, 2007.

Various associations have been described between acute leukemia and flu, chicken pox, measles and mumps, happening either to the mother during the index pregnancy or to the index child. The only common feature of these studies is the lack of consistency (Little J. 1999). Case-control studies providing data on the prevalence of antibodies in childhood and adult cancers are still very sparse (McBeath & Harnden, 1968; Gahrton, 1971; Sprecher-Goldberger, 1971; Geser, 1979; Miller, 1972; Sutton, 1974; Gotlieb-Stematsky, 1975; Clark, 1990; Gentile, 1990; Levine, 1992; Schlehofer, 1996; Lehtinen & Lehtinen, 1998; Salonen, 2002; see also Dorak, 1996 for a discussion of some of these reports). A study showed a risk for childhood ALL in the offspring of mothers with high IgM antibody levels for EBV (Lehtinen, 2003). There is no consistent evidence of an association between vaccinations of any type and the risk of leukemia (Dockerty, 1999; Groves, 2002; Mallol-Mesnard, 2007). The latest UKCCS paper unambiguously reports a correlation between increased infections in infancy and childhood ALL (earlier onset) (Roman, 2007, see also Dorak, 2007; Roman, 2009). The authors acknowledge that these findings do not support the original (Greaves) hypothesis that a deficit of exposure to infectious agents is associated with an increased risk of ALL development (i.e., delayed infection hypothesis) but interpret the data as supporting the hypothesis that a dysregulated immune response to infection in the first few months of life promotes transition to overt ALL later in childhood (Roman, 2007). The original hypothesis was that delayed common infections would cause an abnormal or dysregulated immune response and that would trigger ALL development (Greaves 1988; 1993; 2006). Another recent study that examined the “actual” infection records of children in the UK found no evidence for reduced risk of leukemia following one or more recorded infection in the first year of life (Cardwell, 2008). This study used the UK General Practice Research Database and included 162 leukemia cases and 2215 controls. Another negative result was reported in a population-based study in New York (Rosenbaum, 2005). Medically diagnosed infections before age 1 show a correlation with future childhood leukemia risk in Taiwan too (Chang, 2012). These studies basically confirmed earlier studies. One such study was the inter-regional epidemiological study of childhood cancer (IRESCC) (McKinney, 1987). In the IRESCC study (234 children with leukemia or lymphoma and 468 age- and sex-matched controls), it was found that viral illnesses in the first six months of life increased the risk for childhood leukemia and lymphoma (RR = 4.1, 95% CI = 1.5 to 11.3, P < 0.01). The overall results of these studies that used infection events rather than proxies for infectious exposure do not support the Greaves hypothesis which predicts lower leukemia risk with increasing number of early infections. In fact, increasing number of viral illnesses increases the risk for adult AML (Cooper, 1996) and hospital admissions for infections during infancy increase the risk for adult non-Hodgkin lymphoma (Paltiel, 2006). These findings may help better interpretation of studies of childhood infections and cancer connection and appear to support the simple immune surveillance concept (Dorak, 2007; see also Cancer Genetics), which was proposed specifically for childhood leukemia by Till et al in 1975 in the immunologic study of parents and grandparents of leukemic children (Till, 1975). However, see also an alternative explanation by Greaves & Buffler (2009) and also the latest studies from California Cancer Registry (Marcotte, 2014) and the Northern California Childhood Leukemia Study (Urayama, 2011) which report results that appear to support the protective role of "early" infections in infancy from childhood leukemia.

A study by Kroll et al (2006) analyzed the Oxford-based UK National Registry of Childhood Tumors (NRCT) for annual percentage change in rate of childhood cancers in relation to influenza epidemics. The results showed small peaks in occurrence of childhood c-ALL coincided with influenza epidemics. The findings were exclusive to childhood c-ALL and at least this aspect of the study confirmed the general belief that c-ALL differs in etiology from other childhood leukemias and some contributing factor for c-ALL has changed over time. Another study reported a mildly increased risk for childhood ALL in the offspring of others who had influenza/pneumonia during pregnancy with no subtype specificity (Kwan, 2006). An earlier study, however, found no association (Randolf & Heath, 1974). This issue is by no means closed as other studies found positive or negative correlations between maternal infections during pregnancy and subsequent malignancy their in children (Leck, 1972; Fedrick & Alberman, 1972; Bithell, 1973; Hakulinen, 1973). Thus, there seems to be a lack of justification for the proposed vaccination of pregnant women for influenza with the perceived benefit of protection the children from ALL (BBC News, 24 Oct 2006).

Preschool day care of the index child with leukemia and of their siblings represents a proxy measure of exposure to infections (Alexander, 1993). Some studies found a protective effect of early day care attendance, which supports a role for infections, in particular, the Greaves model of delayed exposure to common infections (Ma, 2002; Perrillat, 2002; Gilham, 2005; Urayama, 2011). There are also negative result studies on the association of day care attendance and childhood leukemia (Dockerty, 1999; Neglia, 2000). Nevertheless, a meta-analysis confirmed a strong protective association of day care attendance with childhood ALL (Urayama, 2010).

If the patterns of exposure to infection and immune stimulation in early childhood is important than a protective effect of breast-feeding could be predicted. This has been investigated in numerous studies. The largest comprehensive study to date using data from the UK Childhood Cancer Study (UKCCS) found weak evidence that having been breast-fed was associated with a small reduction in the odds ratio for leukemia and for all cancers combined. No sub-group of leukemia showed a statistically significant association with having been breast-fed and there were no significant trends with duration of breast-feeding (Beral, 2001). The same report included a meta-analysis of data from the UKCCS with results from other published studies. This showed a small reduction in odds ratios for not only leukemia, but also for HD, non-hematological cancers and all childhood cancers combined, associated with ever having been breast-fed. However, since the hypothesis was that any association with breast-feeding should apply to leukemia and possibly lymphoma, but not to other childhood cancers, the results from the UKCCS alone and from the meta-analysis are difficult to interpret. It was concluded that it is unclear whether there is a generalized small protective effect of breast-feeding across most childhood cancers or whether the results reflect some systematic bias in the majority of studies conducted thus far. Subsequently an analysis of more than 3,000 mothers of children who had died of cancer in the UK up to 1981, and mothers of healthy control children found no evidence of protection from breast-feeding for ALL, or any other diagnostic group (Lancashire & Sorahan, 2003). Other studies reported either no association with breast-feeding (Schuz,1999) or weak protective association (Shu, 1999; Smulevich, 1999; Infante-Rivard, 2000; Bener, 2001 & 2008; Perrillat, 2002; MacArthur, 2008; Ortega-Garcia, 2008) in childhood leukemia. An earlier meta-analysis concluded that there was a small but significant protective effect of breast-feeding for ALL, Hodgkin lymphoma and neuroblastoma (Martin, 2005). However, the latest population-based and large UKCCS study did not find any statistically significant difference in breasty-freeding frequencies between control children and children with ALL, precursor B-cell ALL or AML (Table 1 in Hughes, 2007). If there is any protective effect of breast-feeding, it is more likely to be due to the immune system promoting effect of human milk (Hosea Blewett, 2008).

One interpretation of HLA associations with disease is the prediction of an infectious model of etiology in that disease because of the key role played by HLA in adaptive immune responses to infection. A number of studies have investigated this possibility (Taylor, 1998; Dorak, 1999; for a review, see also MHC and Leukemia and HLA Associations in Leukemia (PPT). A study based on the UK Childhood Cancer Case-Control Study included data on almost 1,000 children with leukemia and demonstrated a statistically significant association between precursor B-cell ALL (formerly c-ALL) and T-cell ALL (T-ALL) and a specific HLA-DPB1 allele, DPB1*0201, compared with children with solid tumors and with normal infants. This has been interpreted as evidence for susceptibility to the leukemogenic effects of infectious agents as the result of their ability to mount an immune response (Taylor, 2002). While HLA associations may suggest an infectious etiology, a number of alternative explanations are possible. For example, another gene within the HLA complex that is in linkage disequilibrium may be the reason for observed association (see for example, Rajsbaum 2001 & 2002). This is not a remote possibility given the fact that HLA complex is the most gene-dense regions in the genome (MHC; Dorak, 2006).

In conclusion, there might be some indirect epidemiological evidence for an infectious etiology for childhood acute leukemia but even if so, the agents involved and the precise mechanisms are unclear (for reviews, see Bunin, 2004; McNally & Eden, 2004; O’Connor, 2007; Ma, 2009). Despite that they are commonly presented as evidence for infectious etiology, day care attendance, clustering, population mixing and HLA associations lack consistency and specificity, and should not be used to promote certain hypotheses unless all alternative explanations are explored and ruled out.

Other Cancers and Infection

Evidence for the involvement of infections in childhood tumors is weak. However, space-time clustering among childhood CNS tumors in the UK has been reported (McNally, 2002). This could support an infectious etiology for at least some CNS tumors. Polyoma viruses have been shown to induce brain tumors in experimental animals (Kryn, 1999) and viral sequences including JC, BK, SV40, CMV and expression of viral oncogenic protein have been detected in human pediatric and adult brain tumors (Cobbs, 2002; Boldorini, 2003; Khalili, 2003). These viruses represent candidate agents in the etiology of childhood brain tumors (Krynska, 1999; Weggen, 2000). Polyoma virus sequences have also been detected in osteosarcomas (Yamamoto, 2000; Butel et al, 1998) and in neuroblastomas (Flaegstad, 1999). A quantitative real-time PCR-based study found SV40 sequences in a variety of childhood ALL and osteosarcomas (Heinsohn, 2005).

Clustering

Epidemiologists have been concerned for many years by the existence of clusters of cases of childhood cancer, particularly of leukemia: if clusters were reflecting a tendency of cases to aggregate they might point to an infectious origin or to localized environmental exposures (reviewed in Greenberg & Shuster, 1985; Schmidt, 1998; Alexander, 1999). The strongest clusters have been observed for children with leukemia diagnosed before six years of age. However, to know whether the occurrence of clusters is due to chance, arising from Poisson variation, or whether attributable to geographically varying underlying factors, raises difficult methodological questions. The development of theoretical and empirical statistical methods for the a priori investigation of general space and space-time clustering engendered extensive publications, debates and exchanges (Draper, 1991; Alexander & Boyle, 1996; Alexander, 1998; Elliott, 2000), particularly in the last 15 years in the UK. Briefly, a first category of methods, exemplified by the Knox test, is based on the geographical and temporal distances between cases and tests where an excess number of cases are observed within various small geographical locations, but only at limited points in time. A second category of methods tests for the non-heterogeneity of risk estimates in geographical space. For this purpose, cases have to be counted in specified space partitions. One of the standard tests of this kind is the Potthoff-Whittinghill test, which allows consideration of areas of various sizes. However, this approach tests for general clustering but does not locate clusters. Some other tests such as SCAN or Besag and Newell statistics have been developed to permit the localization of spatial clusters in addition to the test of overdispersion. Dilutions in time or space, or inappropriate boundaries, are liable to lower the power of the studies to detect spatial or space-time clustering. Besides, clustering reports are usually retrospective studies, which limit the value (O’Connor, 2007).

Although any evidence for a cluster is usually presented as evidence for infectious etiology, no known cause for clustering of childhood cancers exists (Alexander, 1999). The speculations include aberrant responses to common infectious agents following population mixing and environmental pollutants. It has to be remembered that these cancer clusters do not represent cases diagnosed simultaneously (as in an infectious disease outbreak). Possible explanations for observed clusters of childhood cancers include infectious origin, environmental pollutants (Mueller, 2001; Costas, 2002; Seiler, 2004), proximity of birth place to EMF (Draper, 2005) or atmospheric pollutant sources (Knox, 2005). Alternative explanations may even include presence of certain plants chemicals that may be carcinogenic for humans in the areas of clustering (Recouso, 2003; see also Falk, 2000). Cancer clusters are not restricted to childhood leukemias and there are many other cancers appearing in clusters with possible alternative explanations for them (Thun & Sinks, 2004; Kingsley, 2007) (See also CDC Cancer Clusters FAQs; CDC Cancer Clusters Resources; NCI Cancer Clusters; NCI Publication: Cancer & the Environment).

Social Class and Childhood Cancer

In Britain (Hewitt, 1960; Sanders, 1981; Alexander, 1991), Australia (McWhirter, 1982), USA (Browning & Gross, 1968) and in general (Greenberg & Shuster, 1985; Stiller & Parkin, 1996), the incidence of ALL is higher in areas of higher SES or social class, particularly in early childhood. The lower incidence of ALL among Black children in the United States of America (USA) (Koller & MacMahon, 1957; Young & Miller, 1975; Reynolds, 2002; Johnson, 2008) and Africa may be at least partly a social class effect. A higher incidence of Hodgkin lymphoma in older children and young adults has been associated with higher socio-economic status in several studies (Grufferman & Delzell, 1984; Alexander, 1991). In many developing countries, Hodgkin lymphoma has a higher incidence among young children and this pattern seems to be linked to poor socio-economic conditions (Stiller & Parkin, 1990). Neuroblastoma may be slightly more common in children of lower socio-economic status (Davis, 1987). Results from the Czech Republic indicate that the incidence of ALL in children aged 1 to 4 years in Eastern Europe has been increasing with improved socio-economic conditions, resulting in a more marked early childhood peak similar to that found in Western industrialized countries (Hrusak, 2002). However, studies are more positive when based on ecological rather than individual measures of social class. This suggests that differences in leukemia risk might reflect characteristics of the areas and the populations living on them, rather than individual differences (Little J. 1999).

Parental Age

Most studies recognized an increased risk for childhood leukemia with advanced maternal age (Manning & Carroll, 1957; Stewart, 1958; Macmahon & Newill, 1962; Stark & Mantel, 1966; Fasal, 1971; Kaye, 1991; Schuz, 1999; Dockerty, 2001; Reynolds, 2002; Johnson, 2009; Ezzat, 2016). A risk association in ALL with young maternal age has been reported (Shu, 2002); for pre-B-cell ALL both young and advanced maternal age were risk markers in the same study (Shu, 2002). Older fathers are also a risk marker for childhood ALL (Kaye, 1991; Dockerty, 2001; Shu, 2002). Some studies did not find any association in childhood leukemia (Shaw, 1985; Murray, 2002). Advanced age is also associated with risk for childhood rhabdomyosarcoma (Gruffermann, 1982) and astrocytoma (Emerson, 1991). Advanced maternal age and risk association with leukemia is limited to boys in one study (Fasal, 1971). Neuroblastoma shows an association with younger maternal age (Greenberg & Shuster, 1985).

Prenatal Environment and Childhood Cancer

The evidence for exposures occurring during the preconceptional period that have an association with childhood or adulthood cancers is equivocal. Agents definitely related to cancer in children, and adulthood if exposure occurs in utero, include: maternal exposure to ionizing radiation during pregnancy and childhood leukemia and certain other cancers (Macmahon, 1962; Sternglass, 1963; Bithell & Stewart, 1975; MacMahon, 1985), and maternal use of diethylstilbestrol during pregnancy and clear-cell adenocarcinoma of the vagina in their daughters (Herbst, 1971 & 1977 & 1981 & 1999; Anderson, 2000). Male progeny also suffer the effects of in utero exposure to DES. These effects include epididymal cysts, small penile size, cryptorchidism and subsequent testicular cancer (Gill, 1976; Gill, 1977; Depue, 1983; Niculescu, 1985). The demonstration of cancers induced by transplacental exposure to diethylstilbestrol has confirmed the speculation that the prenatal environment may influence subsequent carcinogenesis.

Prenatal Risk Factors for Childhood Cancer

Associations with birth weight of the child and childhood cancer -especially for leukemias- (Macmahon & Newill, 1962; Fasal, 1971; Wertelecki, 1973; Daling, 1984; Robison, 1987; Kuijten, 1990; Cnattingius, 1995; Ross, 1997; Yeazel, 1997; Westergaard, 1997; Schuz, 1999; Smulevich, 1999; Shu, 2002; Murray, 2002; Okcu, 2002; Hjalgrim, 2004; Paltiel, 2004 (female-specific); Roman, 2005; Podvin, 2006; McLaughlin, 2006; Schuz & Forman, 2007; Dorak, 2007 (male-specific); Milne, 2008; Johnson, 2008; Smith, 2009 (female-specific); Tower & Spector, 2007 (review); Samuelsen, 2009 (all cancers in all age groups except CNS tumors); Bjorge, 2013 (all cancers); Oksuzyan, 2012 (ALL and AML); Hjalgrim, 2003 / Caughey & Michels, 2009 / Roman, 2013 / Milne 2013 / O’Neill, 2015 (meta-analyses) have been reported. In the latest and largest (n=40,000) meta-analysis of UK and USA datasets, known positive associations with childhood leukemia, CNS, renal and soft tissue tumors as well as the negative association with hepatic tumors are confirmed with consistency between US and UK data. No associations with retinoblastoma or bone tumors were observed (O’Neill, 2015). Birth weight association also applies to childhood AML (Roman, 1997; Yeazel, 1997), brain tumors (Gold, 1979; Savitz & Ananth, 1994; Heuch, 1998; Harder, 2008 (meta-analysis), see also Milne, 2008 for a negative result), non-CNS solid tumors (Laurvick, 2008 (fetal growth rate)), and Wilms tumor (see Bunin, 2004), but not carcinomas (Johnson, 2011). In neuroblastoma, a tendency towards high birth weight (Greenberg & Shuster, 1985), associations with high birth weight (Yeazel, 1997; O'Neill, 2015) and low birth weight (Johnson, 1985; Schuz, 2001), and no association (Neglia, 1988) have been reported. In childhood lymphoma (McKinney, 1987; Schuz, 1999), hepatoblastoma (see Bunin, 2004), hepatic tumors (O'Neill, 2015) and AML (Bjorge, 2013), lower birth weight has been found to be a risk marker. A meta-analysis in lymphomas did not reveal any association (Papadopoulou, 2012). In leukemia, both low and high birth weight are associated with risk in one study (Schuz, 1999), but a large meta-analysis found a deficit of very low-birth-weight among cases with leukemia (Roman, 2013). In osteosarcoma, an association has been described for children taller than average (Fraumeni, 1967; Mirabello, 2011) as well as with high birth weight (Mirabello, 2011). This observation has similarities to higher incidence of osteosarcoma in large dogs (Tjalma, 1966). In Western Australia (Davis, 2011) and California (Huang & Ducore, 2014), children with ALL have been found to be taller than their healthy counterparts of the same age and sex. A US study has found that both birth weight and birth length are both associated with childhood ALL risk with birth length being the more dominant one (Kennedy, 2015). Similarly, breast cancer is also associated with both, but birth length is the independent association with risk (Silva Idos, 2008). Height shows an association with adult cancers and has been considered as a contributor to the gender effect (Walter, 2013). Birth weight and adult height are correlated and likely to be markers of some aspect of growth that is associated with cancer risk in adulthood. Birth weight, however, adds little additional information to adult height as a predictor of cancer incidence in women (Yang, 2014). Height between 7 and 13 years of age is reported to show correlations with adult glioma risk, but only in males with no association in females (Kitahara, 2014). In males, childhood height (especially at age 13) is positively associated with prostate cancer-specific mortality, especially if diagnosed earlier than age 60 (Aarestrup, 2015).

Not all studies, however, have noted an association with birth weight. Most notable of those is the population-based Californian study (Reynolds, 2002) although another large registry-based study in California reported associations in ALL and AML (Oksuzyan, 2012). One study on birth weight and childhood cancer association found that heavier babies have higher risk for childhood ALL but in males only in the North of England (Dorak, 2007; but see also Milne, 2007a & McNally, 2007), and in females in the UKCCS study (Smith, 2009). A meta-analysis of studies performed by CLIC participants did not observe a sex difference in birth weight association (Milne, 2013). However, very large studies in adults also reported male-specificity of birth weight association with adult cancer risk (Kajantie, 2005) and cancer mortality (Risnes, 2011). Likewise, the protective association of low birth weight with adult cancer mortality has been observed only for men (Syddall, 2005). A Swedish study reported that men had an 8% increased risk for overall cancer at all ages per SD increase in birth weight for gestational age, while women only had an increased risk under age 50 years and it was mainly driven by the association with breast cancer (McCormack, 2005). Birth weight has been reported to correlate linearly with the majority of adult cancer risk in an analysis of more than 200,000 Danish men and women without a sex effect (relative risk = 1.07 per 1000g increase in birth weight) (Ahlgren, 2007). In the Danish study, pancreas and bladder cancers showed a V-shaped correlation with birth weight and testicular cancer had an inverse association with birth weight.

In a cohort study of all children born in the Western Australia between 1980 and 2004, Milne et al (2007b) have reported that both birth weight and fetal growth rate (proportion of optimal birth rate) are associated with childhood ALL but the association with fetal growth rate is also present among children who would not meet the definition of high birth weight (usually defined as > 4,000g). They concluded that it is the accelerated growth rather than high birth weight that is associated with leukemia risk (the same conclusion was reached by a meta-analysis by CLIC (Milne, 2013)). The same group (2008) has reached the same conclusion for Hodgkin and Burkitt lymphoma (in males) and in non-Hodgkin lymphoma (in males). A similar observation was made by Schuz & Forman (2007) in first-borns with ALL or CNS tumors. Fetal growth rate has been found to be a better indicator of risk for CNS tumors than birth weight alone too (Sprehe, 2010). Hispanics in Texas show a positive association between large for gestational age (LGA) and childhood ALL (Barahmani, 2015). The analysis of Swedish Medical Birth Register revealed an association between LGA and non-Hodgkin lymphoma, but only in females (Petridou, 2015). The Swedish study did not observe any association in Hodgkin lymphoma. Findings in another study (McLaughlin, 2006) that high birth weight is associated with ALL only when the mother is not overweight while heavier maternal weight is associated with ALL only when the infant is not high birth weight are interpreted as childhood ALL may be related to factors influencing abnormal fetal growth patterns. It is important to note that the fetal growth rate is inherently higher in males (Mittwoch, 1993; Kochhar, 2001) as evident by the higher birth weight in males. This may be one contributor to the generally increased male-to-female ratio in childhood cancer.

Several plausible mechanisms have been proposed to explain the birth weight and leukemia association. High birth weight results in a higher rate of cell proliferation and/or a larger number of precursor cells being at risk of malignant transformation (Westergaard, 1997). The correlation between birth weight and bone marrow volume is long known (Hudson, 1965) and bone marrow volume also correlates with fetal weight (Wilpshaar, 2002). Insulin-like growth factor-1 (IGF-1) hypothesis suggests that high birth weight is an epiphenomenon of increased growth factor stimulation (Ross, 1996). Birth weight is correlated positively with circulating levels of IGF-1 (Petridou, 2000; Yang & Yu, 2000). IGF-1 is important in blood cell formation and regulation and has been shown to stimulate the growth of both myeloid and lymphoid cells in culture. Thus, it has been proposed that high levels of IGF-1 may produce a larger baby and contribute to leukemogenesis (higher IGF-1 levels are associated with most adult cancers; Renehan, 2004.) The IGF-1 connection in childhood ALL has been reviewed (Tower & Spector, 2007). Although the IGF-1 hypothesis has a strong biological background, SNPs that correlate with serum IGF-1 levels do not show any association with birth weight, childhood growth or glucose metabolism (Vella, 2008). Another hypothesis concerns in utero exposure to high endogenous estrogen levels (Shu, 2002). The basis of this hypothesis is the correlation between maternal estrogen levels and birth weight (Petridou, 1990). Very convincing results on this connection, and their relevance in breast cancer and birth weight association have been presented (Bukowski, 2012). Among 23,824 pregnancies, a strong positive relationship was observed between the infant birth weight (or percentile of individual growth potential), and the mother’s serum estriol 3 (E3)/ anti-estrogen alpha-fetoprotein (AFP) ratio (the E3/AFP ratio reflects net estrogenic activity) or serum pregnancy-associated plasma protein-A (PAPP-A) concentration (PAPP-A has proteolytic activity targeting IGFBPs, which increases concentrations of free IGFs; see Figure 2); and the same study confirmed the birth weight association with breast cancer risk (Figure 1 in Bukowski, 2012). Whether a similar connection exists for childhood cancer -in which estrogens have not been implicated as firmly as in breast cancer- remains to be tested. Finally, iron levels in developing fetus have been proposed to be a determinant of high birth weight (Dorak, 2007), which is associated with childhood leukemia. In this hypothesis, birth weight is just an epiphenomenon, and leukemia risk is increased directly due to high iron levels during fetal development. Experimental support has recently been reported in support of this hypothesis (Dorak, 2009). In a New Zealand study, iron supplementation during pregnancy or childhood did not show any significant association with risk (Dockerty, 2007) but odds ratios were in the risk direction for iron supplementation with no folate during pregnancy (OR = 1.3; 95% CI = 0.8 to 2.3) and childhood (OR = 1.6; 95% CI = 0.1 to 19.3) (Dockerty, 2007; Table 1). A US study, however, reported a decreased risk of ALL in relation to use of iron supplements by mothers in the period 3 months before pregnancy, during pregnancy or while breastfeeding, with an overall odds ratio of 0.7 (95% CI = 0.5 to 0.9) (Kwan, 2007). The risk for childhood cancer with vitamin supplementation in early childhood has shown no change in New Zealand (Dockerty, 2007), but an increase in Canada (OR = 1.7; 95% CI = 1.2 to 2.3) (MacArthur, 2007). Since both iron deficiency and excess are associated with increased risk for cancer, inconsistent results may have resulted from the differences in variable of iron status of populations where the studies were performed (in populations where iron deficiency is prevalent, iron supplementation would shos a protective association, but in iron replete populations, it may increase the risk).

In light of the 'light pollution' theory of cancer (Pauley, 2004; Kantermann & Roenneberg, 2009; Stevens, 2009), body weight shows an intriguing negative correlation with serum melatonin levels in children and adolescents (Waldhauser, 1988). (See also Pineal Gland and Cancer.)

Being the first-born child has been associated with increased risk in some studies and generally for those under 5 years of age (van Steensel-Moll, 1986; Westergaard, 1997; Petridou, 1997; Schuz, 1999; Dockerty, 2001) but the opposite has also been reported (Shaw, 1985; Shu, 1988; Buckley, 1994; Savitz & Ananth, 1994; Infante-Rivard, 2000; Bener, 2001; Shu, 2002; Jourdan-Da Silva, 2004). The largest of all studies has recently reported a generally reduced childhood cancer risk with higher birth order (von Behren, 2010). Although the first-born association in childhood ALL has been promoted as supporting evidence for the Greaves hypothesis (Greaves, 2001), the UKCCS study did not find any association with being first-born (P = 0.86 (total leukemias), 0.91 (ALL), 0.82 (B-cell precursor ALL), 0.74 (AML) in Table 1 in Hughes, 2007). Likewise, a recent Australian cohort study did not find an increased risk for first-borns (in fact, found a statistically non-significant decreased risk) (Milne 2007). One study in Japan even found that being first or fourth born conferred equal risk (Wakabayashi, 1994). A much stronger association was found in childhood AML with having siblings as opposed to being the only child or with higher birth order (Shu, 1988; Ross, 1997; Westergaard, 1997), and AML development is not linked to delayed exposure to common childhood infections. One study identified being the only child as a risk factor (van Steensel-Moll, 1986) although not as strong as being the first-born. There are also negative results studies (Kaye, 1991; Roman, 1997; Shu, 1999; McKinney, 1999; Schuz, 1999; Dockerty, 1999; Neglia, 2000; Ma, 2005; Podvin, 2006; McLaughlin, 2006). A study on the number of siblings and childhood ALL risk compared having older siblings (high birth order) with younger siblings (low birth order or being first-born). In childhood ALL, acute monocytic leukemia and Hodgkin lymphoma, younger siblings were strongly protected compared with older siblings confirming the protective effect of higher birth order (Altieri, 2006), but not just in childhood ALL. Likewise, another large US study reported stronger inverse associations with birth order in CNS tumors, neuroblastoma, bilateral retinoblastoma, Wilms tumor and rhabdomyosarcoma than in ALL (von Behren, 2010). Arguably the most unbiased and largest study of all, a study in the Danish birth cohort of 1973 - 2010 failed to find any association between birth order and risk of any childhood cancer (Schuz, 2015). Given the strong and consistent association of miscarriages with childhood ALL risk (see below), an alternative explanation for any observation regarding birth order or sibling numbers may be the fertility connection in childhood ALL. Miscarriage history increases the risk in subsequent children and having no older sibling may be equivalent to miscarriage history (see below). The birth order studies should ideally examine maternal reproductive history, presence of older and younger siblings, being first-born or the only child and the parental status of all siblings (miscarriage may be partner dependent) in the same study. In germ-cell testicular cancer, there is an inverse association with birth order, but when analyzed in detail, the sibship size also has an inverse association. These findings have been interpreted as “a higher prevalence of parental subfertility among patients with testicular cancer” (Richiardi, 2004). Another study reported a large-for-gestational-age (LGA) association in ALL and CNS tumors in first-borns (Schuz & Forman, 2007).

Some observations on animals suggest that there may be an environmental cause for the first-born effect. Along the west coast of Florida, nearly all first-born bottlenose dolphin calves die before they separate from their mothers. This is thought to be due to high levels of environmental toxins in the fat of marine mammals. Research suggests mother dolphins unload as much as 80 percent of their accumulation of pollutants into each of their calves, probably through nursing. In general, excess risk of reproductive failure measured in terms of stillbirth or neonatal mortality for primiparous females was estimated as 60% (Beaufort), 79% (Sarasota), and 78% (Matagorda Bay). Females of higher parity, who have previously off-loaded a majority of their polychlorinated biphenyls (PCB) burden, exhibit a much lower risk (Schwacke, 2002). The first-born gets the highest dose by far, as the mother has been accumulating toxins for many years while subsequent siblings receive the toxins accumulated over shorter periods of time (Chicago Zoological Society Magazine; NWF Children at Risk Report). Also in Orcas (Orca Network Newsletter) and Northern fur seals (Beckmen, PhD Dissertation, 1997; see Table 8 in p.30 of Persistent Organic Pollutants in Alaska Report) firstborn babies are exposed to milk with higher concentrations of pollutants than the calves/pups of mothers who have previously given birth, and that firstborns have significantly higher pollutant blood concentrations. Following the learning from wildlife experience which led to the banning of DDT (reviewed in Landrigan, 2003), these observations in natural world may contribute to the understanding of the first-born effect in humans. In humans, breastfeeding increases organochlorine transfer to infants (Yakushiji, 1984; Koopman-Esseboom, 1994; Patandin, 1999; Boersma, 2000; Nickerson, 2006). Maternal age and length of previous lactation show positive and inverse correlations, respectively, with the level of contaminants in human milk (Albers, 1996). In the mother, breast feeding decreases the risk for breast cancer among BRCA1/BRCA2 carriers (Jernstrom, 2004) but the mechanism of this is not known. It is more likely to be due to some local effect as another study shows an increased risk for glioma in breast-fed women (Huang, 2004).

Miscarriages

Maternal history of previous miscarriages is a frequently reported risk factor for development of childhood ALL in a subsequent child (Stewart, 1958; Graham, 1966; Gibson, 1968; van Steensel-Moll, 1985; Fajardo-Gutierrez, 1993; Yeazel, 1995; Cnattingius, 1995; Kaatsch, 1996; Smulevich, 1999; Perrillat, 2002; Podvin, 2006; Johnson, 2008) except for infant leukemia which occurs more frequently in females (Ross, 1997). A long birth interval between the index case and preceding birth is a risk marker for childhood ALL in one study (Kaye, 1991) but not in other (Neglia, 2000); and fetal loss in preceding pregnancy for those diagnosed before 2 yr of age (Kaye, 1991). Also in childhood AML, there is an association with maternal history of fetal loss (Ma, 2005), which has been observed in one study only in AML diagnosed before age 2 (Yeazel, 1995). The association extends to threatened abortions/threatened preterm labor (Stewart, 1958; Hewitt, 1966; van Steensel-Moll, 1985; Smulevich, 1999; Milne 2007). Rarely, a protective effect is associated with previous fetal loss in leukemia (Shu, 1988; Murray, 2002) and in astrocytoma (Kuijten, 1990). Few studies did not find any link between miscarriages and childhood ALL and/or childhood cancer (Zack, 1991; Savitz & Ananth, 1994; Ross, 1997; McKinney, 1999; Schuz, 1999; Ma, 2005). Generally, no association between maternal reproductive history and childhood brain tumors has been found (McCredie, 1994; McCredie, 1999; Linet, 1996) but one study reported a risk association with astrocytoma (Emerson, 1991). An increased incidence of cancer, leukemia and lymphoma in the families of women experiencing spontaneous recurrent abortions is also reported (Ho, 1991). The association of miscarriages with childhood leukemia may have some similarities with decreased cancer risk in male twins (Hewitt, 1966, Inskip, 1991, Rodvall, 1992) (probably because of their preferential loss prenatally). The first study that has investigated the sex effect in miscarriage association has found that the association is stronger and statistically significant in boys (Dorak, 2007). A strong and significant protective effect of increasing parity on risk of childhood ALL (Dockerty, 2001) lends support to the risk association with miscarriages. In one study, (adult) females with HD tended to have a lower parity than did controls (Abramson, 1978). Prior treatment for ovulation induction for subfertility also increases the risk for childhood ALL (Ezzat, 2016).

Birth Defects

The prevalence of 55 well-defined mild errors of morphogenesis (MEMs) was determined in 100 children with acute lymphoblastic leukemia (ALL), their 80 sibs, 91 mothers, and 76 fathers. A significantly increased prevalence of MEMs was found in the ALL patients and their sibs of both sexes with no specificity for MEMs or combinations was recorded (Mehes, 1998). Increased incidence of birth defects has also been noted in childhood cancers especially in leukemia in other studies (Savitz & Ananth, 1994; Nishi, 2000). In a study of 90,400 children with and without congenital abnormalities followed-up till 18 years of age, significant increases were found in the incidence of certain childhood cancer (Agha, 2005). The risk was nearly six-times higher ion the first year of life for a child with a birth defect. These results agree with those of earlier registry-based studies (Altmann, 1998; Narod, 1997).

Childhood Cancer in Twins

Generally, there is a high degree of concordance among twins with cancer (MacMahon, 1964, Clarkson, 1971, Chaganti, 1979, Hartley, 1981, Ford, 1997) [although not in all studies (Buckley, 1996)]. There is also evidence for intrauterine single cell origin, with twin-to-twin transmission, of concordant leukemia in twins (Chaganti, 1979, Hartley, 1981, Mahmoud, 1995, Maia, 2003). Intrauterine transmission is a more likely cause for concordance than genetic factors. Therefore, increased concordance for leukemia in twins cannot be used to infer a genetic basis for the disease as classical twin studies do. {It has been shown that leukemia may indeed arise in utero (Gale, 1997, Ford, 1998, Wiemels, 1999, Wiemels, 2002; McHale, 2003). Several molecular studies found the same clonotypical (leukemia-specific) chromosomal changes (including MLL or TEL rearrangements) such as t(12;21), t(4;11), and t(8;21) in patients' blood samples taken at birth. This was shown for patients with infant leukemia and for those with c-ALL indicating prenatal initiation of acute leukemia in most patients. Thus, transmission to the co-twin is possible.}

Classical twin studies and as their extension adoption studies try to estimate the proportional contributions of genetic and environmental factors to diseases or traits (Boomsma, 2002; Kyvik KO, 1997). This is usually attempted by comparing the concordance rate among monozygotic (MZ; genetically identical and shared the same intrauterine environment) and dizygotic (DZ; sharing one-half of their genes and shared the same intrauterine environment) twins. If the concordance rate is 100% in MZ twins but 25-50% in DZ twins, the conclusion would be that the disease is strictly genetic and the causative gene is recessive (if the concordance rate in DZ twins is 25%) or dominant (if the concordance rate is 50%). In multifactorial diseases in which genetics play a more predominant role than environment (such as insulin-dependent diabetes), the concordance rate in MZ twins will not be anywhere close to 100% but will be greater than for DZ twins. In a disease more or less totally environmental (such as infectious diseases), the concordance rate will be close to 100% in both MZ and DZ twins. In a disease with a greater environmental role, the concordance rate will be very similar between MZ and DZ twins. Although in general, these conclusions will be accurate, there are significant exceptions to the assumptions of classical twin studies. Most importantly, in childhood cancer, classical twin studies would be most misleading. For example, high concordance rate in childhood leukemia in MZ twins does not point to a strong genetic basis but it is a result of shared circulation. One general problem with classical twin studies is the ascertainment bias which results from better recognition of concordant pairs and greater willingness of MZ twin pairs and female twin pairs to volunteer compared to DZ twins and male twins. This problem can be best overcome by using population-based twin registries. One assumption of MZ twin studies that they are genetically identical is not always true; MZ twins can differ in terms of somatic rearrangement/mutation of immunoglobulin and T cell receptor genes, X chromosome inactivation in females, genomic imprinting and in utero immune relationship (Ollier & MacGregor, 1995). The assumption of intrauterine similarity has also been questioned. Studies have demonstrated an impact of the intrauterine environment (i.e. low birth weight) for the development of the components in the metabolic syndrome. The validity of conclusions drawn from classical twin studies has therefore been questioned due to the different prenatal circumstances characterizing monozygotic (MZ) and dizygotic (DZ) pregnancies. Due to a potentially more adverse intrauterine environment among MZ compared to DZ twins, MZ twins may be more prone to develop various metabolic abnormalities (Poulsen & Vaag, 2003).

Twin studies on cancer address two general questions, one about the possible carcinogenic effects of twinning and the second about heritable effects of cancer. There is no substantial evidence for carcinogenic effects of twinning. One study reported an increased frequency of kidney tumors in twins (Windham, 1985) and another one found increased frequency of twins among cases in a case-control study (OR = 2.6; 95% CI, 0.8-8.2) (Savitz & Ananth, 1994). In vitro fertilization, which increases multiple births, does not have an effect on childhood cancer risk (Bergh, 1999; Klip, 2001). Another group of studies have compared cancer incidence in twin-born children with incidence in singletons. Earliest twin studies in US and UK found slightly lower cancer incidence in twins compared to singleton births (MacMahon & Newill, 1962; Hewitt, 1966; Miller, 1968; Jackson, 1969; Norris & Jackson, 1970, and the US NorthEast Leukemia Series quoted in Jackson, 1969). This was also one of the earliest observations in the Oxford Study of Childhood Cancers: low proportion of twins from like-sexed pairs (presumably monozygotic) among cancer cases (Stewart & Barber, 1962). A Norwegian study also reported a slightly lower cancer incidence in twins (Windham, 1985) except renal cancer perhaps due to its genetic basis. A study of childhood cancer incidence in 30,925 twins born in Connecticut between 1930 and 1969 revealed 30% less than expected leukemia and cancer risk in twins (Inskip, 1991). Males accounted for most of the deficit which was most profound among boys younger than five years of age. In a study of 35,582 Swedish twins (Rodvall, 1992), the childhood cancer incidence was similar to that in the general population of primarily single-born children except for males under age five. In this group, cancer incidence was reduced (O/E = 0.3, CI = 0.1-0.7). The childhood cancer risk of twins appeared similar to that of singletons except for males aged 0-4 years. The nationwide Swedish Family-Cancer Database contains three twin pairs with childhood leukemia and all three are female pairs (Hemminki, 2001). The deficit of leukemia in male twins was also noted in the Oxford Study of Childhood Cancers (Stewart & Hewitt, 1965; Hewitt, 1966; Hewitt, 1970) but not in the early Californian studies (Jackson, 1969; Norris & Jackson, 1970). A study that analyzed 1063 twins in England and Wales (Swerdlow, 1996) concluded that the results for leukemia accord with previous suggestions that leukemia may be of prenatal origin and may sometimes lead to intrauterine death (Stewart, 1973; Knox, 1984). A more comprehensive analysis of the twin data from England & Wales confirmed the deficit of cancer among twins (Murphy, 2002).

These results have three possible explanations: (1) given the association of high birth weight with childhood cancer risk, and twins are usually low birth weight, the findings may be a reflection of the correlation between birth weight and cancer risk but this would not explain the sex effect (birth weight association is female-specific in more than one study (Fasal, 1971; Paltiel, 2004); if this was the explanation, the deficit in twins should have been in girls not in boys which may be the case for California studies (Jackson, 1969; Norris & Jackson, 1970)); (2) overall perinatal mortality rates for twins is about four-times that for single born due to mainly due to low birth weight and complications of second-born (Gittelsohn & Milham, 1965). This early mortality may cause a lower than expected cancer incidence in later ages; (3) selective early in utero mortality of (male) twins destined to develop cancer (Osborne & Degeorge, 1964; Stewart & Hewitt, 1965; Hewitt, 1966; Hewitt, 1970, Inskip, 1991). This possibility is supported by generally increased vulnerability of males to prenatal selection events reflected by the decrease of male-to-female ratio from 160:100 at the time of fertilization to 106:100 at birth and male-specificity of MHC-mediated prenatal selection (Dorak, 2002 and references therein). Male vulnerability to prenatal challenges is also evident in male children of untreated diabetic or prediabetic mothers who have a high risk of being stillborn (Gellis & Hsia, 1959). In singleton births, it is not possible to show the link between prenatal selective loss of males who would otherwise have developed cancer and postnatal cancer but twin data may be pointing such a link. The epidemiologic association between recognized prenatal loss (recurrent miscarriages) and increased risk for childhood cancer in subsequent live births may be another sign of this link.

Genetics of Childhood Cancer

Cancer in general is a genetic disease but this concerns somatic cells (acquired genetic changes in cancer cells). Genetic changes are usually the target, not the origin, of the cancer process. On the other hand, there are constitutional genetic factors that modify cancer risk. In all common cancers, two classes of inherited genes may be considered in the etiology:

(1) High-penetrance predisposition genes: These genes confer a high degree of cancer risk, usually associated with hereditary major predisposition syndromes, but disease-associated germline mutations are relatively rare in the general population. Cancer due to germ-line DNA mutations (inherited/hereditary/familial cancer) is much less than 10% of all cancers (Pritchard-Jones, 1996; Birch, 1999), the most common one being retinoblastoma (Narod, 1991; Buckley, 1996). These high-penetrance mutations are generally sufficient to initiate carcinogenetic process and include TP53, HNPCC/MLH1, HNPCC/MSH2 and BRCA1/BRCA2. Involvement of these major genes in cancer does not rule the existence of other minor susceptibility genes.

As in other cancers, primary genetic predisposition to childhood cancer is due to mutations in three principal categories of high-penetrance predisposition genes: proto-oncogenes (transcription factor gene activations generally due to translocations), tumor suppressor genes (TP53 gene, cyclin-dependent kinase (CDK) inhibitors; retinoblastoma tumor suppressor gene RB1), and DNA repair genes (Pritchard-Jones, 1996; Shannon, 1998) (see also Oncogenes; Tumor Suppressor Gene Defects in Human Cancer; and Genetic Basis of Cancer Syndromes in Cancer Medicine; Cancer Genetics). Cancers arising as a result of highly penetrant mutations associated with cancer predisposition syndromes (e.g. retinoblastoma, Li-Fraumeni syndrome and certain congenital overgrowth syndromes) account for a minority of all cases (PubMed search for hereditary childhood cancer).

In a US study of 556 twins with childhood cancer, the overall rate of inherited cancers was 5% but just 2% when retinoblastoma is excluded (based on MZ case-wise concordance rates) (Buckley 1996). The Swedish Family-Cancer Database data showed parental cancer as a risk factor for childhood cancer for nervous system cancers, lymphomas, endocrine tumors and retinoblastomas but no excess risk for leukemia and Wilms tumor (Hemminki & Mutanen, 2001). From a UK study of 16,564 children with cancer, it was estimated that 4% of childhood cancers had a known genetic-hereditary basis and were due to inherited highly penetrant genetic mutation (Narod, 1991). The most frequently recorded diagnoses with a hereditary basis were: bilateral retinoblastoma (37% of retinoblastoma cases), hereditary Wilms tumor (7% of cases), leukemia (3% of cases) - associated with Down syndrome, and brain tumors (2% of cases) - associated with neurofibromatosis (type 1 and type 2; Shannon, 1998). Children with Down syndrome are estimated to have an approximate 10- to 15-fold increased risk for the development of acute leukemia (ALL or AML) (Krivit & Good, 1957; Robison & Neglia, 1987; Reynolds, 2002) [Down syndrome is the most common cause of M7 (megakaryoblastic) variant of AML but AML-M7 is not the most common leukemia type caused by it.] Besides Down syndrome and neurofibromatosis, a number of inherited genetic syndromes are associated with childhood cancer (Bondy, 1991; Ross, 1994; Pritchard-Jones, 1996; Greaves, 1997; Birch, 1999; Anderson, 2000; Ganjavi, 2002; Pakakasama, 2002; Rahman, 2005). These include Shwachman-Diamond syndrome, Bloom syndrome, Fanconi anemia, Von Hipple-Lindau syndrome, ataxia-telangiectasia, multiple endocrine neoplasia type I-type II-type IIB, tuberous sclerosis; overgrowth syndromes (Sotos syndrome, Beckwith-Wiedemann syndrome) and several immunodeficiency syndromes (SCID, WAS, XLP; CHS) are other inherited syndromes that predispose children to cancer.

A genetic component in familial cancers is suggested by the following observations:

* Parental consanguinity is seen more often in childhood cancer families (Li, 1976)

* A tendency to cluster in families that experience an excess incidence of leukemia or cancer (Li, 1976, Horwitz, 1997, Draper, 1977, Gunz, 1978, Kato, 1983, Shpilberg, 1994, Farwell, 1984). Excess numbers of cancers in parents of children with cancer are mainly due to retinoblastoma and gonadal germ cell tumors in children (Pang, 2003)

* Having a close relative with cancer increases the risk for childhood cancer (Kuijten, 1990; Smulevich, 1999) some of which are known to be due to germline TP53 mutations (Chompret, 2000)

* Increased cancer tendency in sibships extends to their parents and other relatives in cancer-prone families (Li, 1976)

* Aggregations within sibships are more frequent than would be expected by chance. There is a slightly increased risk that sibs of children with malignant disease will also be affected by such diseases. The magnitude of the risk for such sibs is about 1 in 300, in other words, double the general population risk (Draper, 1977; Draper, 1996). In childhood leukemia, the increased risk for siblings has been shown independently (Miller, 1963; Miller, 1967; Miller, 1971)

* There is a high degree of concordance among twins with cancer although this may be due to intrauterine initiation and transmission in leukemia.

(2) Low-penetrance susceptibility genes: Variant alleles at these genes may confer a smaller degree of cancer risk, but they are carried by a larger proportion of the general population. As a result, the proportion of cancer that could be explained by these more subtle genetic factors may be relatively large. Even in the presence of a major cancer predisposition gene, low-penetrance genes may modify the risk. An example of this is the BRCA1/2 mutations and adult breast cancer (Rebbeck, 2002).

The low-penetrance susceptibility genes include genes encoding enzymes involved in carcinogen metabolism and detoxication 'xenobiotic enzymes', cytokine and chemokine genes (see Polymorphisms Implicated in Cancer Susceptibility). The Mendelian heritability of chromosomal radiosensitivity also appears to be a marker for low penetrance susceptibility to cancer (Scott, 2004). This category of genes is neither necessary nor sufficient for cancer development but they contribute to the risk in a multifactorial setting.

The direct evidence for the relevance of risk modification by susceptibility genes is the genetic association studies at the population level. The following genetic markers have been identified as risk markers in childhood cancer in the context of a multifactorial etiology:

* CYP genes, N-acetyltransferase (NAT), glutathione S-transferase (GST), methylenetetrahydrofolate reductase (MTHFR), 5-methyltetrahydrofolate-homocysteine methyltransferase reductase (MTRR), 5-methyltetrahydrofolate-homocysteine methyltransferase (MTR) and serine hydroxymethyltransferase 1 (SHMT1) gene polymorphisms in childhood ALL (Infante-Rivard, 1999; Krajinovic, 1999; Alves, 2000; Infante-Rivard, 2000; Sinnett, 2000; Saadat & Saadat, 2000; Krajinovic, 2000; Sinnett, 2000 [review], Wiemels, 2001; Ezer, 2002; Davies, 2002; Krajinovic, 2002a; Krajinovic, 2002b; Robien & Ulrich, 2003; Krajinovic, 2004; Canelli, 2004; Gast, 2007; Petra, 2007; Lightfoot, 2010)

* GST polymorphisms in childhood AML (Davies, 2000)

* ABCB1 (MDR1) polymorphism in childhood ALL (Jamroziak, 2004)

* NAD(P)H:quinone oxidoreductase (NQO1) polymorphisms and benzene/chemotherapy-induced bone marrow toxicity/leukemia (Nebert, 2002; Infante-Rivard, 2007)

* DNA repair genetic polymorphisms in childhood ALL (Infante-Rivard, 2000 & 2003; Mathonnet G, 2003; Joseph, 2005)

* NBN (NBS1) 'Nijmegen breakage syndrome gene' association (I171V variant) (Varon, 2001; Mosor, 2006)

* HLA gene polymorphisms in childhood ALL (Dorak, 1995; Taylor, 1995; Dearden, 1996; Taylor, 1998; Dorak, 1999; Taylor, 2002; Dorak, 2002; MHC and Leukemia [review] [PPT])

* HFE-C282Y (rs1800562) polymorphisms in childhood ALL (Dorak, 1999; Dorak, 2005 [review])

- See also a recent systematic review of childhood ALL associations by (Vijayakrishnan & Houlston, 2010)

Childhood cancer as a whole, therefore, is not an inherited disease (Bondy, 1991; Hawkins, 1995, Horwitz, 1997; Birch, 1999). Rates of cancer among offspring of childhood cancer survivors are reported to be similar to those for the general population confirming the lack of a primary genetic cause for most childhood cancers.

For reviews of childhood cancer and genetics, see Shannon, 1998; Stiller, 2004; Canalle, 2004; Smith, 2005; Kim, 2006; Belson, 2007; Chokkalingam & Buffler, 2008.

Sex Differential in Childhood Cancer

The sex effect in incidence of childhood cancer is well-established and consistent worldwide (Ashley, 1969; Greenberg & Shuster, 1985; Linet & Devesa, 1991; Little J, 1999; Pearce & Parker, 2001; Desandes, 2004; Johnson, 2008; Forsythe, 2010). Tower and Spector provide graphs or leukemia rates worldwide for each sex separately which show the increased risk for boys clearly (Tower & Spector, 2007). Among newly diagnosed childhood cancers, the standardized (with European reference) incidence rates for all participating registries in Europe yields a boys to girls ratio for adjusted rates is on average 1.22. The incidence of ALL among children younger than 15 years of age is consistently higher among males (approximately 20%) relative to females. For the 15-19 year olds, however, the male preponderance was greater, with males having a 2-fold higher ALL incidence than females (SEER Report, see also Average Annual Age-Specific Incidence Rates per Million, SEER, 1993-1997). The male predominance is a feature of cancer incidence in all ages (Cartwright, 2002; Boyle & Ferlay, 2005; Cook, 2009; Edgren, 2012).

Although the male-to-female (M:F) age-adjusted incidence is>1.0 for all types of leukemias and lymphomas, the ratiois highest (M:F: 3.0) for non-Hodgkin lymphoma, similar forALL and HD (both M:F: 1.3), and lowest for acutemyeloid leukemia (M:F: 1.1; Table 1 in Linet, 2003). Burkitt lymphoma is one of the childhood (and adult) tumors with the highest M:F ratio (Boerma, 2004). The M:F ratio alsovaries among the subtypes of central nervous system tumors,with the highest ratio apparent for ependymomas (M:F: 2.0) andprimitive neuroectodermal tumors (M:F: 1.7), but there is littledifference between male and female age-adjusted incidences forastrocytomas and other gliomas (Table 2 in Linet, 2003). Boys and girls have a similar incidence of retinoblastoma and Wilms tumor. Only for extragonadal, non-intracranial germ cell tumors, malignant melanoma and some carcinomas, notably those of the adrenal cortex and thyroid (Inskip, 2001), including radioactive iodine-induced form (Cardis, 2005), and alveolar soft part sarcoma (Bu, 2005), there is an excess among girls (UK National Childhood Cancer Statistics, 2004). For M-to-F ratio in each childhood cancer, see Table 13.1 in UK National Childhood Cancer Statistics (see also Table 4 in Linet, 2003). Reasons are unknown for the male predominance in incidence ofnon-Hodgkin lymphoma and ependymomas; the higher incidencesamong young females for thyroid cancer and malignant melanoma;and the lack of sex-related differences in incidences ofacute myeloid leukemia, astrocytomas, and other gliomas, butetiologic leads to consider include exposures that differ bysex, effects of hormonal influences, and sex-related geneticdifferences (Linet, 2003). The sex effect is not only seen in incidence of childhood ALL but also in prognosis; males having more relapse, worse prognosis and secondary cancer (Sather, 1981; Gustafsson & Kreuger, 1983; Woodcock, 1984; Lanning, 1992; Chessells, 1995; Shuster, 1998; Pui, 1999; Eden, 2000; Devarahally, 2003). In fact, boys present with higher risk features at the diagnosis (Forsythe, 2010).

The susceptibility by sex at different ages is a phenomenon rarely addressed in the analyses of epidemiological studies, yet the risks for males of certain ages can be between two- and fivefold greater than females, which is in need of further investigation (Cartwright, 2002). As one possible mechanism of the male-female differential in childhood cancers, in particular Hodgkin lymphoma, greater frequency of an asymptomatic carrier state in this sex has been suggested but not investigated (Vianna & Polan, 1978).

Following observations have been made in relation to sex effect in childhood leukemia / cancers and may be relevant in the explanation of this phenomenon:

* The male excess in childhood ALL is consistent worldwide and the populations with a lower M:F ratio tend to have low total leukemia and ALL incidence (Linet & Devesa, 1991)

* Males present with high risk features for B-precursor ALL (Forsythe, 2010)

* In leukemia, prognosis is worse in boys compared to boys (Sather, 1981; Gustafsson & Kreuger, 1983; Lanning, 1992; Chessells, 1995; Shuster, 1998; Pui, 1999; Eden, 2000; Hussein & Xie, 2015)

* The risk for second primary malignancies is higher in males following childhood CNS tumors (Devarahally, 2003)

* In twin studies, there is a deficit of twin boys with cancer (Hewitt, 1966; Hewitt, 1970, Inskip, 1991; Rodvall, 1992)

* Male survivors of childhood cancer have a lower proportion of live birth and a reversed male-to-female ratio in their offspring suggesting a male deficit among their children (Green, 2003)

* Advanced maternal age and risk association is seen only in boys in two studies (Fasal, 1971; Reynolds, 2002)

* Paternal exposure to chemicals (dibromochloropropane and dioxin) (Potashnik, 1984; Mocarelli, 2000; Jonbloet, 2002) decreases the sex (M/F) ratio in the offspring although the opposite effect has also been reported (Karmaus, 2002). Parental smoking during the periconceptional period also decreases male-to-female ration at birth (see a commentary at a CCC newsletter)

* In the original Oxford Study of Childhood Cancer (Hewitt, 1966), out of 14 survivors of threatened abortions who developed a malignancy in the first six months, only one was a male

* In the original Oxford Study of Childhood Cancer (Hewitt, 1966), unaffected sibs of familial cases of childhood leukemia have a low male-to-female ratio (0.71)

* Male children of untreated diabetic or prediabetic mothers have a higher risk of being stillborn (Gellis & Hsia, 1950)

* Seasonality in childhood HD is restricted to males only in one study (Fraumeni & Li, 1969).

* If infections have anything to do with childhood cancers, boys are more vulnerable to childhood infections than girls (Washburn, 1965; Schlegel, 1969; Purtilo, 1979; Schmitz, 1983; Rechavi, 1992; Green, 1992; Read, 1997). The most striking example is of course EBV infections in X-linked lymphoproliferative disease (Seemayer, 1993)

* The association of childhood leukemia with cleft lip and palate is based on three male cases (Zack, 1991)

* Association of childhood leukemia with high birth weight is more pronounced in a subgroup of female children of older mothers with a high socioeconomic status (Fasal, 1971; Paltiel, 2004). This has been shown in twin females too (Jackson, 1969)

* A more recent population-based study showed that in childhood ALL, the birth weight association is male-specific (Dorak, 2007)

* Miscarriage association in childhood ALL is stronger and statistically significant in boys only (Dorak, 2007)

* Females have a greater risk of developing thyroid cancer than males following postnatal irradiation (Hempelmann, 1975; Inskip, 2001; Cardis, 2005)

* Familial aggregation of NHL is male-specific (Chatterjee, 2004)

* Genetic susceptibility studies have shown sex-specific associations:

- Blood groups ABO frequencies differ between male and female patients in leukemia (Jackson, 1999)

- DNA repair gene XRCC1 (Joseph, 2005); MSH3 (Infante-Rivard, 2000); APEX1 (Infante-Rivard, 2003)

- Xenobiotic enzyme polymorphisms (Krajinovic, 1999; Sinnett, 2000)

- HLA-DRB4 and HFE associations (Dorak, 1995; Dorak, 1999a & 1999b; 2005), HLA-B67 association (Ng, 2006), HLA-DQA association (Taylor, 1998)

- Genetic risk markers for multiple sclerosis in HLA-DRA (rs3135388), HLA-C (rs9264942 as a proxy for HLA-C*05 association in MS) and IFNG (rs2069727) also modify the risk of childhood ALL with the same sex effect (Morrison, 2010)

- Interferon regulatory factor-4 (IRF4) SNP rs12203592 is associated with childhood ALL risk in males and its mechanism has also been examined in vitro (Do, 2010)

* The growth rate of the embryo is higher for males than females in different species including humans (Mittwoch, 1993; Kochhar, 2001). Because accelerated rates of cell division and proliferation may increase the predisposed to the development of cancer (Preston-Martin, 1990), this inherent feature of males may explain some of the sex effect in (childhood) cancers.

* The primary sex ratio at fertilization may be as high as 165:100 (see for example: Tricomi, 1960; Shettles, 1964; Serr & Ismajovich, 1963; Lee & Takano, 1970; McMillen, 1979; Kellokumpu-Lehtinen & Pelliniemi, 1984; Vatten, 2004; C3 Newsletter 13/2) but it falls down to 106:100 at birth in humans (and similarly in most mammals). A continuation of this process (elimination of excess males) is the increased morbidity and mortality of male infants and children (well-known male disadvantage (Stevenson, 2000) or fragile male (Kraemer, 2000), which has evolutionary explanations (Trivers & Willard, 1973; Wells, 2000; Dorak, 2002)). It can be speculated that the excess risk in males for childhood cancers and infections may be due to the continuing elimination of excess males.

* Homozygosity for HLA-DR haplotypes (one of which associated with risk for childhood ALL in males) shows a deficit in newborn males (Dorak, 2002)

* A finding that may be relevant in sex effect is that newborn boys have a higher homozygote TT frequency for MTHFR 677C>T SNP (Rozen, 1999). However, the 677T allele is protective for childhood ALL (Wiemels, 2001; Robien & Ulrich, 2003)

* One of the major groups of oxidative enzymes involved in drug metabolism, the CYP450 enzymes, have differential activity between males and females (Harris, 1995; Anderson, 2002). CYP3A4 activity, for example, is higher in women than in men (Harris, 1995). Likewise, GST activity also shows sex-specific differences (Singhal, 1992)

* In adults, MDR activity is higher in males with chronic lymphoid leukemia (Steiner, 1998). It has been suggested that this may be one reason for the less aggressive clinical course in women.

* Penetrance of mutations in DNA mismatch repair genes MLH1/MSH2 is significantly higher in males (approximately 80%) than in females (40%) (Mitchell, 2002). DNA mismatch repair gene mutations usually cause adult colon cancer in heterozygous form but a variety of childhood cancer in homozygous forms (Lucci-Cordisco, 2003)

* In animal studies, males are more susceptible to oxidative damage (Ma, 1998). In humans, sex effect in oxidative damage has also been suggested (Loft, 1992; Proteggente, 2002)

* In animal studies, sensitivity to mutagenic carcinogens and the risk of radiation carcinogenesis are greater in males (Hattis, 2004)

* An in vitro study showed a higher radiosensitivity of lymphocytes from males regardless of age and ethnicity (Wang, 2000)

* Maternal serum ferritin levels are at 36 weeks of gestation correlate with umbilical cord serum ferritin of male but not female infants (Tamura, 1999). This may be relevant in the male-specificity of HFE-C282Y (rs1800562) association in childhood ALL (Dorak, 1999)