Breastfeeding and childhood cancer

The relationship between childhood cancer and having been breastfed in infancy was investigated in the UK Childhood Cancer Study (UKCCS), a national, population-based case-control study. Analyses included 3500 children with cancer (cases) of whom 1637 were diagnosed with leukaemia, 114 with Hodgkin's disease, 228 with non-Hodgkin's lymphoma and 1521 with other cancer and 6964 controls. 62% cases and 64% controls were reported to have ever been breastfed. There was weak evidence, of borderline statistical significance, that having been breastfed was associated with a small reduction in the odds ratios for leukaemia (odds ratio = 0.89, 95% Cl 0.80–1.00, P = 0.06), and for all cancers combined (odds ratio = 0.92, 95% Cl 0.84–1.00, P = 0.05). Combining data from the UKCCS with results from other published studies showed a small reduction in the odds ratios for leukaemia, Hodgkin's disease, non-haematological cancers, and all childhood cancers combined, associated with ever having been breastfed. It is unclear whether the apparent small reduction in the odds ratio for these various types of childhood cancer is a generalized effect of breastfeeding or whether it reflects some systematic bias in the majority of studies that have investigated the question. © 2001 Cancer Research Campaign http://www.bjcancer.com

The identification of leukaemia and lymphoma viruses in various animal species (Gross, 1951;Vogt and Ishizaki, 1965) prompted the search for viral causes of human leukaemia and lymphoma. The Epstein Barr virus has been associated with Burkitt's lymphoma in children (Epstein et al, 1964) and the human T-cell leukaemia virus with a rare form of adult leukaemia/lymphoma (Poiesz et al, 1980) but, as yet, no specific infectious agent has been associated with the majority of human leukaemias or lymphomas. Recently, attention has focused on the association between surrogate markers of exposure to infectious agents and the risk of childhood leukaemia (Greaves and Alexander, 1993;Kinlen, 1995).
Breastfed infants are at a reduced risk of gastrointestinal and respiratory infections (WHO Collaborative Study Team on the Role of Breastfeeding on the Prevention of Infant Mortality, 2000), but are also at an increased risk of mother-to-child transmission of certain infectious agents (Ruff, 1994;Nduati et al, 2000). Several case-control studies of children with leukaemia have reported a significantly reduced risk associated with having been breastfed as infants (Shu et al, 1999;Smulevich et al, 1999;Infante-Rivard et al, 2000;Bener et al, 2001), although not all studies have noted such a relationship (van Steensel-Moll et al, 1986;McKinney et al, 1987;Davis et al, 1988;Magnani et al, 1988;Shu et al, 1995;Petridou et al, 1997;Dockerty et al, 1999;Schüz et al, 1999;Rosenbaum et al, 2000;Hardell and Dreifaldt, 2001). Others have reported a significantly reduced risk of Hodgkin's disease (Grufferman et al, 1998) and non-Hodgkin's lymphoma (Bener et al, 2001) in breastfed children.
We report here the results of an analysis of breastfeeding patterns in infancy among children with leukaemia and other cancers and among control children in the UK Childhood Cancer Study (UK Childhood Cancer Study Investigators, 2000). An a priori hypothesis was that any association with breastfeeding should be seen for childhood leukaemia, and possibly for childhood lymphoma, but not for other childhood cancers.

Subjects
The UKCCS is a national population-based case-control study. Details of the study design are given elsewhere (UK Childhood Cancer Study Investigators, 2000). Briefly, children aged 0-14 with cancer diagnosed in England and Wales (1992-96) and Scotland (1991-94) were eligible for inclusion in the study. Detailed diagnostic review information was obtained from multiple sources, including histopathology reviews and clinical trial data. Subtyping of leukaemia was made possible by collecting together and reviewing cytogenetic and molecular details (UK Childhood Cancer Study Investigators, 2000). For each case, two controls were randomly selected from Family Health Services Authorities (in England and Wales) and Health Boards (in Scotland) matched to the case by month and year of birth, sex and region of residence at diagnosis. Eligible controls who did not participate in the study were replaced by another matched control until, where possible, two controls had been enrolled for each participating case. Participation rates were 87% for cases and 72% for first-choice controls.
The mothers of 3838 cases and 7629 controls were interviewed, using a standard questionnaire, that included questions about social, medical and behavioural factors. Included among the questions to the mothers were 'Did you ever breastfeed (the child)? How old was (the child) when you gave your last breast feed? Did you ever use formula milk? If so, how old was (the child) when he/she had his/her first formula feed? Was (the child) ever given expressed milk from a milk bank?' Responses to these questions are examined in this report.

Analysis
A total of 943 children aged under 12 months at diagnosis/pseudodiagnosis were excluded from the analyses to prevent the introduction of possible biases if breastfeeding were to cease prematurely in children aged under 12 months with cancer. A further 60 children were excluded because information about breastfeeding had not been provided by their biological mother. Overall 338 cases and 665 controls were excluded leaving 3500 cases and 6964 controls -91% of all interviewed and potentially eligible subjects.
Analyses were carried out for all cancers combined (3500 cases), and separately for leukaemia (1637 cases), Hodgkin's disease (114 cases), non-Hodgkin's lymphoma (228 cases) and all other cancers (1521 cases). For certain analyses children with leukaemia were subdivided into those with acute lymphoblastic leukaemia (ALL, 1401 cases) and acute myeloid leukaemia (AML, 214 cases). Those with ALL were subclassified further, as having common ALL (cALL, 879 cases), T cell ALL (134 cases), and those with B-lineage ALL were further subclassified as TEL-AML 1 (95 cases) or hyperdiploidy (358 cases).
Since the hypotheses concerning the role of infectious agents in childhood cancer are chiefly concerned with leukaemia and lymphoma, and because of the possible differential participation of the parents of children with and without cancer and differential recall of events in childhood, some analyses were restricted to children with cancer. In these restricted analyses, the breastfeeding histories of children with leukaemia or lymphoma were compared with those of children with other cancers, i.e. taking the children with other cancers as 'controls'.
Odds ratios (ORs) and their 95% confidence intervals (CIs) were estimated using unconditional logistic regression (StataCorp, 1999) and adjusted for region (10 in total), sex, age at diagnosis (in single years), birth order (1, 2, 3, 4+) and deprivation index (7 levels). For certain analyses, additional adjustments were performed for the mother's age at leaving school and at the time of the child's birth, child's birthweight, whether or not the child had been admitted to a special care baby unit, and had attended day care in the first year of life, and maternal smoking when the child was 1 year old. Trends were assessed using likelihood ratio tests.
Results showing data from more than one study are presented in the form of plots of odds ratios. Each odds ratio is plotted as a black square, the area of which is proportional to the amount of statistical information for that particular estimate, i.e. the reciprocal of the variance of the log odds ratio, which reflects the reliability with which the odds ratio is estimated. The corresponding 95% CIs are shown as horizontal black lines. Where CIs are so narrow as to be contained entirely in the black box, they are plotted as white lines. Summary odds ratios are plotted in the form of a diamond, the width of which indicates the corresponding 95% CIs, and the dotted vertical line indicates the value of the summary odds ratio. Table 1 shows the distribution of children with leukaemia, Hodgkin's disease, non-Hodgkin's lymphoma, other cancers, all cancers and controls, according to whether or not they were reported to have ever been breastfed and the reported duration that they had been breastfed; it also shows the corresponding adjusted odds ratios, with 95% confidence intervals, and the results of tests for trend by duration of breastfeeding. About two thirds of the children were reported to have been breastfed, with almost one fifth for longer than 6 months. The adjusted odds ratios associated with ever having been breastfed were below 1.0 and of borderline statistical significance for leukaemia (OR = 0.89, 95%CI 0.80-1.00, P = 0.06) and for all cancers combined (OR = 0.92, 95% CI 0.84-1.00, P = 0.05). For each cancer type shown in Table 1, none of the trends with increasing duration of breastfeeding was statistically significant. The odds ratios in Table 1 were not materially altered after additional adjustment for maternal education, maternal age, birthweight, admission to a special care baby unit, attendance at day care, or maternal smoking.   2 shows results, for various subtypes of leukaemia, for children who had ever versus never been breastfed and according to the reported duration of breastfeeding. The numbers in each category are comparatively small and none of the odds ratios or tests for trend with duration of breastfeeding was statistically significant.

RESULTS
The results for leukaemia were also examined separately for first born children and the odds ratio associated with ever having been breastfed in this subgroup was 0.98 (95% CI 0.82-1.18). Table 3 shows age-specific results according to whether or not the children were breastfed, separately for children with Table 2 Distribution of children with leukaemia, according to the type of leukaemia and breastfeeding history, and the associated odds ratios* (ORs) and 95% confidence intervals (CIs)

Acute myeloid
Acute lymphoblastic Sub-categories of acute lymphoblastic leukaemia leukaemia (AML) leukaemia (ALL) n = 214 n = 1401 T cell ALL Common ALL Sub-categories of B-lineage ALL n = 134 n = 879  leukaemia, Hodgkin's disease, non-Hodgkin's lymphoma, other cancers, and all cancers combined. The divisions by age were determined a priori, with the specific hypothesis that the effects of any infectious agent would be greatest for cALL at ages 2-5 (Greaves and Alexander, 1993). For leukaemia the odds ratios did not differ significantly from 1.0 at any age and there is no statistical evidence that the results varied according to age at diagnosis. Most children with leukaemia were diagnosed at ages 2-5 years and in this age group the odds ratios for the various leukaemia subtypes for ever having been breastfed versus never having been breastfed were: 0. . Both these tumour types contributed to the comparatively high odds ratio for children aged 1 and the comparatively low odds ratio at ages 2-5, and these may well be chance findings. Table 4 shows results according to whether or not the child was ever given formula feed and, if so, the age the child was when such feeds were introduced. No consistent pattern is evident, and no results are statistically significant. Nor did the risk of childhood cancer appear to be influenced by having been given expressed milk from a milk bank. The results are, however, based on small numbers (about 2% of children ever received milk from a milk bank), the respective odds ratios were: 0.   When children with other cancers are taken as the comparison group, the odds ratios associated with ever having been breastfed Test for heterogeneity ≤ 6 months: χ 10 2 = 13.01; P > 0.1; NS Test for heterogeneity > 6 months: χ 10 2 = 17.92; P < 0.1 Test for heterogeneity between short-term and long-term breastfeeding: χ 1 2 = 6.90; P < 0.01 a Data provided by McKinney (personal communication) b Numbers of cases and controls derived from data presented by the authors c "Never" includes breastfed for <2 days (Dockerty et al 1999); < 1 month (Magnani et al 1988, Shu et al 1995, Schüz et al 1999, Shu et al 1999, Hardell et al 2001; < 2 months (Smulevich et al 1999); are 1.00, 1.24 and 1.26, respectively for leukaemia, Hodgkin's disease and non-Hodgkin's lymphoma (Table 5).
The results from the UKCCS study with respect to breastfeeding were compared with relevant results from other epidemiological studies of childhood cancer (van Steensel-Moll et al, 1986;McKinney et al, 1987;Davis et al, 1988;Hartley et al, 1988;Magnani et al, 1988;Golding et al, 1990;Shu et al, 1995;Petridou et al, 1997;Grufferman et al, 1998;Dockerty et al, 1999;Schüz et al, 1999;Shu et al, 1999;Smulevich et al, 1999;Infante-Rivard et al, 2000;Rosenbaum et al, 2000;Bener et al, 2001;Hardell and Dreifaldt, 2001). The UKCCS results did not differ significantly from those of previous studies, they have been combined with these and the overall results are shown for leukaemia (Figure 1), Hodgkin's disease (Figure 2), non-Hodgkin's lymphoma (Figure 3), other childhood cancers ( Figure 4) and all childhood cancers combined ( Figure 5). Not all studies published separate results for each cancer type and so the totals for each type do not add up to the 'all cancer' values. For childhood leukaemia, there is evidence of a statistically significant reduction in the odds ratio associated with ever having been breastfed (OR = 0.86, 95% CI 0.81-0.92) and having been breastfed for more than 6 months (OR = 0.78, 95% CI 0.71-0.85). However, there is also evidence of a reduction in the odds ratio of similar magnitude associated with ever having been breastfed for Hodgkin's disease (OR = 0.74, 95% CI 0.58-0.93), all non-haematological cancers (OR = 0.85, 95% CI 0.76-0.95) and for all childhood cancers combined (OR = 0.90, 95% CI 0.84-0.97).

DISCUSSION
Breastfeeding provides a range of immunological benefits to the infant (Jelliffe and Jelliffe, 1978;Goldman, 1993;Cuthbertson, 1999; WHO Collaborative Team on the Role of Breastfeeding on the Prevention of Infant Mortality, 2000). In the first hours after birth, breast milk contains anti-microbial immunoglobulins; subsequently it also contains lymphocytes, macrophages and soluble factors, such as cytokines, lactoferrin, lysozyme, complement components, that may alter response to infection (Goldman, 1993;Hanson, 2000). However, breastfeeding also predisposes to the transmission of various infectious agents from mother to child (Ruff, 1994;Nduati et al, 2000).
Greaves' delayed exposure hypothesis predicts a protective effect for protracted breastfeeding specifically for childhood acute lymphoblastic leukaemia as a consequence of priming of the developing immune system by both vertical transmission of infective agents and immunological properties of breast milk (Greaves, 1997). The results from the UKCCS provide, however, equivocal Test for heterogeneity ≤ 6 months: χ 4 2 = 0.80; P > 0.1; NS Test for heterogeneity > 6 months: χ 4 2 = 17.55; P < 0.01 Test for heterogeneity between short-term and long-term breastfeeding: χ 1 2 = 2.44; P > 0.1; NS a Numbers of cases and controls derived from data presented by the authors b "Never" includes breastfed for < 1 month (Shu et al 1995, Hardell et al 2001; < 2 months (Smulevich et al 1999); ≤6 months ( Although there is weak evidence, of borderline statistical significance, of a small reduction in the odds ratio for leukaemia in children who were breastfed, a reduction in the odds ratio of a similar magnitude was also found for all childhood cancers. Furthermore, the combined results from all relevant published studies suggest that, although having been breastfed is associated with an apparent small reduction in the odds ratio for childhood leukaemia (Figure 1), it is also associated with a small reduction in the odds ratio for Hodgkin's disease (Figure 2), non-haematological childhood cancers ( Figure 4) and for all childhood cancers combined ( Figure 5). Information on breastfeeding practices were reported by mothers at the time of interview. Although there are known inaccuracies in such reporting (DHS Comparative Studies, 1999), it is unknown whether this is biased between the mothers of cases and controls. Studies comparing the duration of breastfeeding reported by mothers with the duration that was recorded at the time when their children were young indicate that women tend to round the reported durations to the nearest 6 months (DHS Comparative Studies, 1999).
Furthermore, 8 years after the birth of a child there is an overall tendency for women to report somewhat longer durations of breastfeeding than had been recorded at the time when the child was young (Vobecky et al, 1988). A review of the role of infant feeding in the aetiology of childhood diabetes mellitus showed that in studies where there was a substantially lower participation rate in controls than cases there was a tendency for the odds ratio associated with ever having been breastfed to be reduced (Norris and Scott, 1996). Participation in the UKCCS was lower in controls than cases (UK Childhood Cancer Study, 2000). Furthermore, participating controls were of higher socio-economic status than the first choice controls eligible for this study, but whether or not these factors produced bias is unknown.
In conclusion, it is unclear whether the apparent small reduction in the risk of each type of childhood cancer reported here is a nonspecific effect of breastfeeding on childhood cancer or whether it reflects some systematic bias shared by the majority of the casecontrol studies that have investigated aetiological factors in Numbers of cases and controls derived from data presented by the authors c "Never" includes breastfed for < 1 months (Magnani et al 1988, Shu et al 1995, Hardell et al 2001; <2 months (Smulevich et al 1999) Odds ratio for all childhood cancers except leukaemia and lymphoma by history of having been breastfe d Test for heterogeneity ≤ 6 months: χ 3 2 = 1.13; P > 0.1; NS Test for heterogeneity > 6 months: χ 3 2 = 11.64; P < 0.01 Test for heterogeneity between short-term and long-term breastfeeding: χ 1 2 = 0.05; P > 0.1; NS a Numbers of cases and controls derived from data presented by the authors b "Never" includes breastfed for < 1 month (Hardell et al 2001); < 2 months (Smulevich et al 1999) 0  Test for heterogeneity ≤ 6 months: χ 3 2 = 1.25; P > 0.1; NS Test for heterogeneity > 6 months: χ 3 2 = 9.30; P < 0.05 Test for heterogeneity between short-term and long-term breastfeeding: χ 1 2 = 1.15; P > 0.1; NS a Numbers of cases and controls derived from data presented by the authors b "Never" includes breastfed for < 1 month (Hardell et al 2001); < 2 months (Smulevich et al 1999) 0.0 1.0 2.0 3.0 0.0 1.0 2.0 3.0