This article is based on the Second Sir Richard Doll Lecture given at the International Agency for Research on Cancer, Lyon, France: May 11, 2005.
Epidemiology and the causes of breast cancer†
Article first published online: 15 DEC 2005
Copyright © 2005 Wiley-Liss, Inc.
International Journal of Cancer
Volume 118, Issue 10, pages 2373–2378, 15 May 2006
How to Cite
MacMahon, B. (2006), Epidemiology and the causes of breast cancer. Int. J. Cancer, 118: 2373–2378. doi: 10.1002/ijc.21404
- Issue published online: 28 FEB 2006
- Article first published online: 15 DEC 2005
- Manuscript Accepted: 14 JUN 2005
- Manuscript Received: 7 JUN 2005
- breast cancer;
- international causes;
- early life
This article describes the characteristics of 3 dominant features of breast cancer epidemiology. These characteristics include the association of disease risk with childbearing, its relationship to ovarian activity and its international variation (particularly as the latter differs in the years before and after menopause). Equivocal tests of one hypothesis that reconciled some of these features through variations in levels of the fractions of estrogen are described. Other hypotheses with a similar objective are needed. The 3 known causes of human breast cancer, ionizing radiation, exogenous ovarian hormones and beverage alcohol, offer some preventive possibilities but do little to explain the epidemiologic features of the majority of cases of the disease that occur in their absence. © 2005 Wiley-Liss, Inc.
For this lecture and article, acknowledgement is due to Sir Richard Doll for the inspiration he provided to so many in the field of epidemiology, and to the International Agency for Research on Cancer for establishing a lectureship in his honor. My article does not do justice to the extensive literature on the epidemiology of breast cancer, but is, as requested by the sponsors of the lecture, focused on the work of a group of investigators active in the Department of Epidemiology at the Harvard School of Public Health during the final quarter of the last century. Space does not permit listing all those who participated in the activities of this group at one time or another, but the central roles of Philip Cole (now at the University of Alabama) and Dimitrios Trichopoulos (now at the University of Athens) must be acknowledged.
Hankinson and Hunter1 provide a recent and comprehensive review of what is known of the epidemiology of cancer of the breast. Three features constitute a conundrum that lies at the center of understanding the etiology of this disease: (i) the role of childbearing; (ii) the place of the ovary; and (iii) the international variation in risks, particularly age-specific risks.
It has been known for many years that a woman's risk of breast cancer is reduced by bearing children, as suggested by high rates in nuns and low rates in married relative to single women. There was also the belief that the phenomenon might be explained by a protective effect of breastfeeding, breast cancer risk being low in areas of the world where prolonged breastfeeding was customary, and increasing in the West where breastfeeding was declining. At a time when thoughts of chemical carcinogenesis were prominent, the idea also had some appeal in the thought that lactation might wash out the bad stuff (whatever it was) along with the good. Some early studies seemed to support this view.2, 3 In the 1960s, we aimed to clear away this particular piece of under-brush to see whether or not it concealed a pheasant, and organized a case-control study in 7 parts of the world, including areas of high, intermediate and low breast cancer risk.4, 5 As shown in Figure 1, we found serendipitously a more or less linear relationship between a woman's risk and the age at which she bore her first child. The apparently protective relationship continued well down into the early teen years. One might ask whether the association is confounded by the fact that data from these very young age groups come disproportionately from countries of low breast cancer risk, but, with some irregularities due to small numbers, the relationship was seen in all study centers regardless of overall rates. Lane-Claypon described a difference between breast cancer patients and comparison patients in age at first birth as far back as 1924,2 and a number of later authors found the same,1 but before our report the phenomenon had been described as a difference in some measure of central tendency, usually a mean. Because the numbers of births at very young ages is small, they contribute little to such measures, and they do not show the strength of the relationship that becomes evident when risks are computed for individual ages.
For all women whose first birth occurred after age 35 risks were higher, not lower, than those of nullipara. Again, this was true in low as well as high-risk centers. For women with first births at 35 or older, risk overall was 60% higher than for nullipara. Multivariate analysis suggested some additional protection for births after the first, but the decrease in risk associated with later births was only one quarter of that associated with the first.6 Others have reported similar findings.7
An interesting twist to this story is that the effect of childbirth varies with the interval since the birth. This was best shown by Lambe et al.8 using data from a Swedish case-control study. Among women who had only one child, breast cancer risk was actually higher in the first 10 years after the birth than in women who had not borne a child, but after 10 years it was lower. The older a woman at the time of her first birth, the higher was her risk in the first 10 years after the birth. Because most of a woman's life after her first delivery comes more than 10 years later, the overall result is protection. This dual effect of pregnancy has also been found by others.9
With respect to lactation, there was considerable variation between the centers in our study in the frequency and duration of breastfeeding, but, after adjustment for relevant variables, there were no significant differences between cases and comparison patients in (i) the number of women who had children but never breastfed, (ii) the number of children who were not breastfed, or (iii) the total duration of lactation for each woman. These conclusions were generally supported by a number of other studies in developed countries,3 but with one refinement: there seems to be a modest decrease in breast cancer risk associated with breastfeeding in young women, sometimes characterized as premenopausal. First noted by Byers in 1985,10 this association has also been found by others.11 Because pre-menopausal women comprise only about 20% of breast cancer cases in Western populations, a relationship within this group does not affect the conclusion that breast-feeding does not explain the association between childbearing and breast cancer risk.
A different picture emerged from a re-analysis of data from 47 studies (that did not include our own studies, but included data from many areas of the world where breast cancer rates are low) undertaken by Oxford's Collaborative Group in 2002.12 The analysis suggested that the relative risk of breast cancer decreased by 4.3% for each 12 months of breastfeeding. On this basis, because in developed countries included in this analysis the mean lifetime duration of lactation was <9 months, it would be difficult to detect an association of such low magnitude in data from them. The estimates of effect at the longer durations of lactation in the group material are strongly driven by data from the developing countries where prolonged lactation is common; indeed, it is likely that there are some levels of lactation in developing countries at which women from developed areas are not represented at all. It seems inappropriate to apply an overall measure of effect that depends in large part on data from undeveloped areas to developed areas where exposures are not nearly so prolonged.
Our own study included 2 centers in which a quarter or more of the women had lifetime lactation of 5 years of more (Sao Paulo and Tokyo).4 Data from these 2 centers indicated only a small and non-significant difference between cases and controls in total duration of lactation, and no difference at all in the proportion of women who breastfed for 5 years or more. It therefore seems that an association of breast cancer risk with lactation is manifest only in areas where accumulate lifetime lactations of more than 5 years are common. It is unlikely that the concomitants of lactation are the same in such areas as they are in developed countries. We may even ask whether it is possible that the very long durations of lactation in undeveloped areas are only markers of some other characteristic of those areas that is more directly responsible for the areas' low breast cancer risks. Among the possibilities is some component of under-nutrition that might operate not through an effect on lactation itself but through a secondary effect in the reduction of some component of ovarian activity that might be more susceptible to alteration in areas of inadequate nutrition. Although studies in developed countries have not consistently identified specific nutrients as being either positively or negatively associated with breast cancer risk,13, 14 the comprehensive review of the World Cancer Research Fund found that an “energy-rich” diet is associated with early onset of menarche and, by implication, increased breast cancer risk,15 and the same has been suggested as a “unifying concept” underlying several features of breast cancer epidemiology.16 This and similar broad categories of nutritional adequacy at various stages of life need to be explored, particularly in developing countries.
Before leaving the subject of lactation, the question of whether women who themselves were breastfed as infants carry increased risk of the disease, a question prompted by the fact that certain strain of mice transmit mammary-tumor virus in breast milk, has been addressed in 2 studies. No association was found, either in the very large Four-State case-control study,17 or in the prospective Nurses' Health study, in which information was obtained both from the patients and their mothers.18 Moreover, in the case-control study, breast cancer risk was not increased among women breastfed by mothers who later developed breast cancer. The hypothesis of transmission through breast milk is unlikely to have relevance to humans.
It has long been known that ovariectomy reduces breast cancer risk in experimental animals. Evidence that this also occurs in humans first came from Lilienfeld in 195619 and was confirmed by us in 1960.20 Some years later, we compared information on surgical menopause from the US National Health Survey with data on 3500 cases of breast cancer reported to the Connecticut Cancer Registry.21 For all women whose menopause was surgically induced, breast cancer risk was less than half that of women whose menopause occurred naturally. Women whose surgery occurred before 35 years of age had only a third of the breast cancer risk of those whose menopause occurred naturally. Reduced risk did not appear in the first 10 years after the operation but thereafter persisted for many years. In humans, oophorectomy is rare before the age of 35 (more than half way through reproductive life), but neutered dogs, commonly neutered before reproduction, have only 12% of the breast cancer risk of intact animals, and those neutered before any estrus have <1% of that risk.22
Evidence of the role of the ovary in breast cancer is also seen in observations that increased risk is associated with early onset of menses and late natural menopause, of which there is abundant evidence.1 With respect to menopause, the shape of the age-incidence curve for breast cancer is interesting. In Western countries rates increase sharply with age up to the time of menopause, but at a slower rate after that. As noted by Boyle, the pattern becomes more interesting in international data.23
Overall, age-adjusted incidence and mortality rates of breast cancer have a range of the order of 5-fold.24 From a data-set on cancer incidence from the ‘Five Continents’ series,25 IARC provided data for 3 areas representing high, medium and low risk (Connecticut, Finland, Japan). Age-specific rates are shown in Figure 2. In all 3 areas, rates increase regularly with age up to the 50–54 age group; after that, although in Connecticut they continue to increase throughout life, in Japan there is no further increase with age. In Finland, rates continue to increase for 5 years later than in Japan, but again level off at ages above that.
The plot thickens if we add time to the mix, particularly in an area of low rates, such as Japan. There, as shown in Figure 3, rates have increased over time in all age groups. Note that the point of change from increasing to level rates remains in the same age group (50–54) throughout the 20-year period. It has been suggested that, in some countries, increases in rates have occurred in a pattern characteristic of generations or ‘birth cohorts,’ that is, as if the increases occurred very early in life and remained characteristic of the generation throughout life.26, 27 This is clearly not the case in Japan, for, if it were, the inflection in the age curve from increasing to level rates would have occurred in later age-groups in successive periods of time, as later-born generations came to occupy those age groups. This picture is more suggestive of a change that has affected all ages at the same time, regardless of the generation in which the members of the age group were born.
The data from Finland, shown in Figure 4, are a little more suggestive of a generation effect in that over time the point of change from increasing to level rates has become later in life as later-born generations came to occupy successive age groups. But, here again, the point of inflection remained in the 50–54 age-group throughout the first 20 years of the period (the lower 4 lines in the figure), and although the point of inflection came at a later age in the second half of the period, it is only 20 years later than that in the earlier period, not the 40 years later that would be expected if it were a purely generation phenomenon. We need to give much more attention to these age differences on an international basis, for the menopause, if indeed it is involved, is ubiquitous and occurs at approximately the same age in all populations. The IARC data-set will be invaluable for this purpose.
It is evident that triangulation of the inter-play of childbearing, the role of the ovary in general and of the menopause in particular, and the age-specific international variation presents a conundrum worthy of Sherlock Holmes. The problem is, however, not fictional. One attempt to reconcile at least 2 of 3 features focused on the paradox that pregnancy, with its associated increase in levels of ovarian hormones, leads to decreased rather than increased breast cancer risk. Lemon et al.28 suggested that the differences in carcinogenic potential of the several chemical forms of estrogen, the so-called ‘estrogen fractions,’ might be relevant. There are 3 major estrogen fractions: estrone, estradiol and estriol. Experimentally, estriol is less carcinogenic than estrone and estradiol and at one time was thought to compete with them for binding sites, so that its presence might reduce the carcinogenic potential of a mixture of the three. The ratio of estriol to the sum of the others, the ‘estriol ratio,’ might therefore have relevance as a cause, or at least a marker, of high breast cancer risk. The idea derived its appeal from the fact that circulating levels of estriol increase very sharply during the last months of pregnancy. We added the thought that the relationship might be particularly relevant in young women,29 and in collaboration with Dr. James Brown of the University of Melbourne, conducted a number of indirect tests of the hypothesis, focusing on estriol ratios in the urine of young women.
These included a study of North American and Asian women 15–39 years of age.30 Among Asian women, estriol ratios were broadly distributed over the spectrum, whereas those of North Americans were clustered in the unfavorable end. The difference was particularly striking in younger women and the follicular phase of the menstrual cycle. A larger study included three age groups in 2 centers in North America and 3 in Asia.44 There was a clear separation between the North American and the Asian centers, with substantially higher estriol ratios in the Asian than in the American samples. For follicular and luteal specimens, and for each of 3 age groups in the 18–34-year-old range, mean values for the estriol ratios in the North Americans were approximately half those in the Asian women in both cycle phases. The relationship of urine estriol ratios to childbearing was also explored.31 North American women between 19–34 years of age who had given birth within the previous 6–18 months were compared to age-matched women who had never given birth. Mean estriol ratios were significantly higher in the parous than the non-parous in women 19–23 years of age, but not in older women.
As knowledge of the biology of the estrogen fractions developed, the idea that the estrogen fractions compete for binding sites became less attractive. Moreover, a likely explanation of the protective effect of pregnancy emerged from the experimental work of Russo et al.32 In the rat, growth of mammary and pre-mammary cells begins in utero and continues throughout early life, with periods of augmented activity during puberty and pregnancy. Growth includes increase in the number of stem cells and differentiation from stem cells to functioning ductal and secretor cells, both of which involve the possibility of mutation and malignant transformation. The final differentiation occurs during pregnancy, particularly the first pregnancy, leading to increased probability of malignant transformation in the short term, but to a population of fully differentiated cells at lower risk of transformation thereafter. The parallels in the human data seem be close. Although this model offers a likely explanation of the protective effect of pregnancy, the association of high breast cancer risk with low urinary estriol ratios, as evidenced in international comparisons and in association with early first birth, remains unexplained. The observations seem to deserve more investigation than they have received. It is unknown whether the mechanism of the relationships observed in the rat parallels that postulated via estrogen fractions in humans.
In any speculation about the etiology of human breast cancer, one must at least mention its 3 known causes: exogenous estrogens, ionizing radiation and beverage alcohol. It remains unclear what relevance they have to the great majority of cases that occur in the absence of exposure to any of them.
Ovarian hormones are commonly taken exogenously, either for contraception, or as ‘replacement’ therapy for symptoms believed to be due to low levels of the natural products, usually during or after menopause. When oral contraceptives were introduced in the early 1960s there was considerable speculation, based on experimental work, that they might increase the risk of cancer of the breast and other sites. These concerns have been allayed by a great number of studies; although questions still remain about the use of OC in some groups of women (such as very young women) they involve only small numbers of women and small increases in risk, and generally relate to concerns other than breast cancer.1
Replacement hormones are another matter. In 1976, Hoover et al.33 published the first evidence of increased risk among women taking replacement estrogens. In a large gynecologic practice, there was a 30% excess of breast cancer among women taking Premarin, a kind of estrogen stew derived from the urine of pregnant mares, and among those taking the medication for 15 years or more the risk was doubled. Moreover, women taking the medication after removal of both ovaries, a group that would be expected to have reduced risk of breast cancer, had 4 times the expected risk after 15 years. Numbers were small in this study but the extensive subsequent literature generally confirms this association.34
The most convincing evidence of causation in epidemiology comes from randomized trials. The U.S. National Institutes of Health began a group of such trials in 1992 under the name of the Women's Health Initiative. In one component of this program, postmenopausal women were assigned randomly to receive either an estrogen–progestagen combination or a placebo.35 The trial was stopped after 5 years when a statistically preset boundary for an unacceptable increase in breast cancer, as well as other complications, was reached. A second component of the program addressed the effect of therapy with estrogens only.36 This trial was also stopped prematurely, but for reasons unrelated to breast cancer. The number of breast cancer cases was lower in the treated than in the placebo group. The duration of follow-up and therapy was smaller than that associated with increased risk in observational studies, but case-control data also show a higher breast cancer risk associated with estrogen-progestagen combinations than with estrogen-only formulations.37 In the observational British Million Women Study, relative risks and 95% confidence intervals for women taking estrogen-progestagen and estrogen-only were 2.0 (1.8–2.1) and 1.3 (1.2–1.4), respectively.38 Studies of exogenous hormones have strengthened the evidence for an ovarian role in the etiology of breast cancer, but also have shifted emphasis from a sole preoccupation with estrogens to include a concern for other ovarian products, such as the progestagens.
Mammary tissue is quite susceptible to malignant transformation by ionizing radiation. Excess breast cancer has been observed in patients given multiple fluoroscopies, radiotherapy for ankylosing spondylitis or enlargement of the thymus gland, and in survivors of the atomic bombings, painters of radium watch faces and X-ray technicians.39 It is relevant to the probability of the significance of early life events that among survivors of the atomic bombs in Japan, breast cancer risk was strongest among those exposed during childhood and was not elevated among women exposed after 40 years of age.40
The findings on beverage alcohol are summarized in a joint analysis by the Oxford Group of data from 53 epidemiologic studies.41 Women who had an average daily consumption of 4 or more drinks a day had a 50% higher breast cancer risk than those who did not drink alcohol. The association seems to be independent of the type of beverage in which the alcohol is consumed.
For both radiation and alcohol the evidence of association with breast cancer risk is undeniable. The association with radiation the association is certainly causal, and that with alcohol is probably so. Although each provides a basis for private action and public policy, their relevance to the etiology of the naturally occurring disease is not clear, and they offer no obvious avenues for new research.
An area of investigation that has not been part of the classical domain of breast cancer epidemiology but has recently come into focus is the possible importance of factors operating in early life. The association of increased risk with early menarche and the protective effect of early pregnancy point to the significance of events in the early stages of reproductive life, but recent writings by Trichopoulos et al.42, 43 have suggested that we might look even earlier. Ahlgren et al.44 report an impressive example of the value of record linkage. In a large cohort of girls born in Copenhagen, linkage of data from birth certificates, school health examinations and the Danish Cancer Registry identified over 3,000 cases of breast cancer. Risk was positively associated with birth weight, height at 8 and 14 years, growth rate around the time of puberty and body mass index at age 14. Previous authors had reported the association with birth weight, and this association, unlikely as it seemed initially, is now well established.45 This, and links to body size and childhood growth,46 direct suspicion toward etiologic factors associated with growth in early childhood as well as in intra-uterine life. They need to be explored further. Whatever significance events in early life may turn out to have, however, the effect of early childbirth and the variation in the age-association of the disease over the international picture leave no doubt that some factor or factors operating during and perhaps after reproductive life also have roles to play. The field of possibilities remains broad.
- 1Breast cancer. In: AdamiH-O, HunterD, TrichopoulosD, eds. Textbook of cancer epidemiology. New York: Oxford, 2002. p 301–39., .
- 2A further report on cancer of the breast with special reference to an association with antecedent conditions. London: Her Majesty's Stationery Office, 1926..
- 12Collaborative Group on Hormonal Factors in Breast Cancer and Valerie Beral. Breast cancer and breastfeeding: collaborative reanalysis of individual data from 47 epidemiological studies in 30 countries, including 50302 women with breast cancer and 96973 women without the disease. Lancet 2002; 360: 187–95.
- 15World Cancer Research Fund International. Nutrition and the prevention of cancer: a global perspective. Washington DC: American Institute for Cancer Research, 1997.
- 23Epidemiology of breast cancer. Baillière's Clin Oncol 1988; 2: 1–57..
- 25Cancer incidence in five continents. vol. I-VIII. IARC CancerBase No. 7. Lyon: IARC, 2005., , , .
- 29Oestrogen fractions during early reproductive life in the aetiology of breast cancer. Lancet 1969; i: 604–6., .
- 31Oestrogen profiles of parous and nulliparous women. Lancet 1976; ii: 596–9., , .
- 32Comparative study of human and rat mammary carcinogenesis. Lab Invest 1990; 62: 224–78., , , , , .
- 34Collaborative Group on Hormonal Factors in Breast Cancer. Breast cancer and hormone replacement therapy: collaborative reanalysis of data from 51 epidemiological studies of 52,705 women with breast cancer and 108,411 women without breast cancer. Lancet 1997; 350: 1047–59.