20.07fhp187_195

Articles
Breast cancer and breastfeeding: collaborative reanalysis ofindividual data from 47 epidemiological studies in 30 countries,including 50 302 women with breast cancer and 96 973 womenwithout the disease Collaborative Group on Hormonal Factors in Breast Cancer* IntroductionAlthough childbearing is known to protect against breast Background Although childbearing is known to protect against cancer, what contribution breastfeeding has on this breast cancer, whether or not breastfeeding contributes to protective effect, if any, has been difficult to determine.
Breastfeeding is closely related to many other aspects ofchildbearing—for example, women breastfeed only after Methods Individual data from 47 epidemiological studies in they have had a child, and the earlier they commence 30 countries that included information on breastfeeding childbearing, the more children they have and the longer patterns and other aspects of childbearing were collected, their lifetime duration of breastfeeding. No single study checked, and analysed centrally, for 50 302 women with has been large enough to reliably characterise the relative invasive breast cancer and 96 973 controls. Estimates of the contributions of such closely related factors in breast relative risk for breast cancer associated with breastfeeding in cancer. This study combines data from 47 epidemiological parous women were obtained after stratification by fine studies conducted in 30 countries, to examine the relation divisions of age, parity, and women’s ages when their first between breastfeeding and breast cancer, taking careful child was born, as well as by study and menopausal status.
account of the effects of other related aspects ofchildbearing.
Findings Women with breast cancer had, on average, fewerbirths than did controls (2·2 vs 2·6). Furthermore, fewer parous women with cancer than parous controls had ever Contributing studies and collection of data breastfed (71% vs 79%), and their average lifetime duration of The Collaborative Group on Hormonal Factors in Breast breastfeeding was shorter (9·8 vs 15·6 months). The relative Cancer has brought together worldwide data from risk of breast cancer decreased by 4·3% (95% CI 2·9–5·8; epidemiological studies of women with breast cancer to p<0·0001) for every 12 months of breastfeeding in addition to describe the relation between breast cancer and various a decrease of 7·0% (5·0–9·0; p<0·0001) for each birth. The reproductive, hormonal, and other factors.1–4 Case control size of the decline in the relative risk of breast cancer and cohort studies were eligible for the collaboration if they associated with breastfeeding did not differ significantly for had data for at least 100 women with incident invasive women in developed and developing countries, and did not breast cancer and had recorded information on each woman vary significantly by age, menopausal status, ethnic origin, the with respect to reproductive factors and use of hormonal number of births a woman had, her age when her first child preparations. For data from cohort studies, a nested case was born, or any of nine other personal characteristics control design was used, in which four randomly selected examined. It is estimated that the cumulative incidence of controls per case were matched for age at diagnosis and, breast cancer in developed countries would be reduced by where appropriate, broad geographical region. The methods more than half, from 6·3 to 2·7 per 100 women by age 70, if of identifying studies and of data collection, checking, and women had the average number of births and lifetime duration correction, have been described elsewhere.1–4 of breastfeeding that had been prevalent in developing Data were collated and analysed on individual women countries until recently. Breastfeeding could account for centrally so that analyses could be done with as similar almost two-thirds of this estimated reduction in breast cancer definitions across studies as possible. Details sought from principal investigators of each participating study includeddata collected regarding each woman’s total number of Interpretation The longer women breast feed the more they pregnancies, her age at each pregnancy, and the outcome are protected against breast cancer. The lack of or short of each pregnancy. A woman’s parity was defined as the lifetime duration of breastfeeding typical of women in total number of births, be they livebirths or stillbirths. In developed countries makes a major contribution to the high some studies, details of past births did not include incidence of breast cancer in these countries.
stillbirths, and for those studies a woman’s parity wastaken to be the total number of livebirths. Information was sought on the total number of children each woman hadbreastfed, her total (lifetime) duration of breastfeeding,and whether or not each individual live-born child hadbeen breastfed, and, if so, for how long. Included in theseanalyses are data from 45 published5–49 and two unpublished studies (Cancer Research UK, unpublished Correspondence to: Prof Valerie Beral, Cancer Research UK, data) that contributed data on lifetime duration of Epidemiology Unit, Gibson Building, Radcliffe Infirmary, Oxford breastfeeding, all but eight6–8,20,23,33,44,47 of which also provided information on the number of children breastfed.
THE LANCET • Vol 360 • July 20, 2002 • www.thelancet.com For personal use. Only reproduce with permission from The Lancet Publishing Figure 1: Details and results from studies that contributed data on breastfeeding and breast cancer*Results of two unpublished studies are also cited here. †Ten developing countries. ‡Three developed countries.
Statistical analysis and presentation of results mathematical models sacrifices some statistical power but The statistical methods used are similar to those used in has the advantage of avoiding assumptions about the previous reports.1–3 In this study, data from different precise forms of any relations in the data. The stratified studies are combined by means of the Mantel-Haenszel O–E values, together with their variances and covariances, stratification technique, the stratum-specific quantities yield both statistical descriptions (odds ratios, calculated being the standard observed minus expected subsequently referred to as relative risks) and statistical (O–E) numbers of women with breast cancer, together tests (p values). When only two groups are compared, with their variances and covariances.50 Use of these simple relative risk estimates are obtained from O–E values by the stratified O–E values in preference to more complex one-step method,50 as are their standard errors (SE) and THE LANCET • Vol 360 • July 20, 2002 • www.thelancet.com For personal use. Only reproduce with permission from The Lancet Publishing Lifetime duration of breastfeeding (months) Data are means unless otherwise indicated. *Excludes 7992 nulliparous cases, 13 379 nulliparous controls, and 3816 women with missing values.
Table 1: Relation between lifetime duration of breastfeeding and various other aspects of childbearing in parous cases and controls* CIs. When more than two groups are compared, variances with women whose total duration was 6 months or less are estimated by treating the relative risks as floating for some analyses. Where appropriate, a trend in the absolute risks (FARs).51 This approach yields floated relative risk of breast cancer with increasing duration standard errors (FSE) and floated CIs (FCI). The use of of breastfeeding is calculated. In such instances, the FARs rather than conventional methods does not alter the duration of breastfeeding associated with a particular relative risks but slightly reduces the variances attributed to category is taken to be the median duration within the relative risks that are not defined as 1·0, and also reduces unwanted covariances between them. Presentation In general, results in the text are presented as relative of the results in this way enables valid comparisons risks and their appropriate standard errors (SE or FSE).
between any two exposure groups, even if neither is the Where results are presented in the form of plots, relative baseline group. Any comparison between groups must take risks and their corresponding CIs or FCIs are represented the variation in each estimate into account by summing the by squares and lines, respectively. The position of the variances of the logarithms of the two FARs.
square indicates the value of the relative risk and its area is To ensure that women in one study are compared inversely proportional to the variance of the logarithm of directly only with similar women in the same study, all the relative risk, thereby providing an indication of the analyses are routinely stratified by study,by centre within study, by fine divisions of age (16–19, 20–24, 25–29, by singleyears from 30 to 79, 80–84, and 85–89years), by age at first birth (Ͻ20, 20–24, 25–29, у30), and by menopausal status(premenopausal, Ͻ5 or у5 years sincemenopause, hysterectomy before appropriate, parous women are furtherstratified by fine divisions of parity lifetime duration of breastfeeding was 6months or less, and the tendency forthe duration to be reported as multiples of 6 or 12 months (see the webfigure athttp://image.thelancet.com/extras/01art9187webfigure.pdf), this variable 7–18, 19–30, 31–54, Ͼ54 months) for certain analyses. Most studies thatrecorded information on breastfeeding period, such as a week, a month, oreven longer, before she was classified as Figure 2: Relative risk of breast cancer in parous women according to breastfeeding *Calculated as floating absolute risk (FAR), and stratified by study, age, age at first birth, and THE LANCET • Vol 360 • July 20, 2002 • www.thelancet.com For personal use. Only reproduce with permission from The Lancet Publishing amount of statistical information available for that increasing duration of breastfeeding, but this aspect of particular estimate. Owing to the large number of relative childbearing is less strongly related to lifetime duration of risk estimates calculated, results are generally given with breastfeeding than is parity or the number of children their appropriate 99% CIs or 99% FCIs, with 95% CIs or breastfed. Thus, there is considerable potential for any effect of breastfeeding on the risk of breast cancer to be To investigate the contribution of childbearing and confounded by the effects of each birth and, to a lesser breastfeeding patterns prevalent in developed countries to extent, by the ages women were when their children were the incidence of breast cancer in these countries, the incidence that would have occurred if the women had had To separate out the effects of breastfeeding from those the patterns of childbearing and breastfeeding that had of other aspects of childbearing, the first step was to been typical for Asian and African countries until describe the relation between breast cancer and certain recently,53,54 is estimated by applying the relative risks reproductive factors, in the absence of breastfeeding. Then obtained in this report to age-specific incidence rates for any additional contribution from breastfeeding is breast cancer in developed countries around 1990.1–4,52 The examined, taking account of the role of reproductive cumulative incidence of breast cancer up to age 70 years is patterns and of other potential confounding factors.
then calculated from the estimated age-specific results.
Having established, in the study population as a whole,what the independent effect of breastfeeding is, the consistency of the main results is examined across various The sponsors of the study had no role in study design, data subgroups of women and across studies and study designs. collection, data analysis, data interpretation, or writing ofthe report.
Breast cancer in relation to childbearing in women whonever breastfed 12 214 (29%) parous cases and 16 900 (21%) parous Altogether 50 302 women with invasive breast cancer controls had never breastfed (table 1). Analyses restricted (cases) and 96 973 women without breast cancer to these 29 114 women provide a description of the (controls) from 47 studies in 30 countries are included in relation between breast cancer and childbearing patterns, these analyses (figure 1). Among the cases, the median that is not affected by breastfeeding. The younger such year of diagnosis was 1988 and the average age at women were when they commenced childbearing, the diagnosis was 50·1 years. Cases had, on average, fewer lower was their relative risk of breast cancer; the relative births than did controls (2·2 vs 2·6) and a greater risk declining by 3·0% (SE 0·3%; p<0·0001) for each year proportion were nulliparous (16% vs 14%). The younger that women were when their first child was born.
proportion of parous women who had ever breastfed was After stratifying by women’s ages when their first child was also lower in cases than in controls (71% vs 79%). The born, as well as by study, age, and menopausal status, the average parity and average total duration of breastfeeding relative risk of breast cancer also decreased with the in parous women varied across countries, largely number of births a woman had (figure 2). In the absence of reflecting the small family size and short lifetime duration breastfeeding, each birth reduces the relative risk of breast of breastfeeding that have characterised women in many developed countries during the past century. Theproportion of parous women who had ever breastfed was Breast cancer in relation to breastfeeding lowest in the USA, at around 50%, whereas in Japan, Figure 2 shows the relative risk of breast cancer by parity, Scandinavia, and developing countries more than 90% of for women who had breastfed, as well as for women who parous women had ever breastfed. Overall, the average had never done so. Women with one child who had never lifetime duration of breastfeeding was 9·8 and 15·6 breastfed are taken to have a relative risk of 1·0. The months, respectively, for parous cases and controls. As relative risk of breast cancer declines with increasing parity expected, the lifetime duration of breastfeeding was much shorter for women in developed than developingcountries (average 8·7 and 29·2 months, respectively, incontrols). Table 1 shows, for parous cases and controls, the distribution of lifetime duration of breastfeeding and therelation of that factor to various other indices of childbearing. The mean parity and mean number ofchildren breastfed were greater for women with longerlifetime durations of breastfeeding. The age women were when their first child was born decreased slightly with Lifetime duration of breastfeeding (years) *Calculated as floating absolute risk (FAR), with corresponding floated standarderror (FSE), and stratified by study, age, parity, age at first birth, and Figure 3: Relative risk of breast cancer in parous women in relation to lifetime duration of breastfeeding Table 2: Relative risk of breast cancer in parous women, in *Calculated as floating absolute risk (FAR), and stratified by study, age, relation to lifetime duration of breastfeeding parity, age at first birth, and menopausal status.
THE LANCET • Vol 360 • July 20, 2002 • www.thelancet.com For personal use. Only reproduce with permission from The Lancet Publishing in women who had ever and who had never breastfed.
However, at each parity the relative risk is slightly lower for women who had breastfed than for women who had not(relative risk for ever versus never having breastfed, adjusted for parity and other factors shown in figure 2 is Separating out the unique contribution of breastfeeding to the risk of breast cancer is not straightforward; women breastfeed only after they have had a birth, the lifetime duration of breastfeeding increases with increasing parity (table 1); and the independent effect of each birth on the risk of breast cancer is substantial in the absence ofbreastfeeding (figure 2). Hence, the effect of each birth needs to be considered carefully when looking at the relation between breastfeeding and breast cancer.
Moreover, the reported lifetime duration of breastfeeding is not very accurate, with values often rounded to multiples of 6 or 12 months, especially for women who breastfed for long periods (see the webfigure at http://image.thelancet.com/extras/01art9187webfigure.
pdf). Additionally, comparatively few women in the studiesincluded here had breastfed for long periods—only 7% of the cases and 15% of the controls reported lifetime durations of breastfeeding of longer than 30 months(table 1). So, even in this large dataset, examination of the relation between breastfeeding and breast cancer is hampered by the potential for confounding, as well as by measurement errors and the limited numbers withsubstantial exposures. The potential confounding by parity can be virtually eliminated by stratification of all analyses by fine divisions of parity. The estimated relative risks of breast cancer, according to lifetime duration of breastfeeding shown in table 2, are stratified by parity from 1 up to 8+, as well as by study, age, age at first birth, and menopausal status.
The relative risk of breast cancer declines with increasingduration of breastfeeding, the estimated reduction in the relative risk per 12 months of breastfeeding being 4·5% (0·7%; p<0·0001; figure 3). Because there is some variation between studies in the classification of womenwhose lifetime duration of breastfeeding was short, sensitivity analyses were done, grouping together women with lifetime durations of 0 and 6 months or less. Whenthis was done, the estimated decline in the relative risk of breast cancer was virtually unchanged, at 4·3% (0·8%; p<0·0001) per 12 months of breastfeeding. Because this latter approach provides a more consistent classificationacross studies than the former, women with reported lifetime durations of breastfeeding of 0 and 6 months or less are grouped together subsequently when trends are Analyses similar to those in table 2 have been done separately for women of parity 1, 2, 3, 4, 5, and 6 or more (figure 4, and webtable 1 at http://image.thelancet.com/ extras/01art9187webtable1.pdf). The relative risk of breastcancer declined with increasing duration of breastfeeding at each parity, and the magnitude of the decline did not vary significantly across women of different parity (␹2 for heterogeneity 1·3; p=0·9). However, the standard errorsand hence the confidence intervals for each parity-specific Figure 4: Reduction in the relative risk of breast cancerassociated with breastfeeding in various subgroups of parous *Stratified by study, age, parity, age at first birth, and menopausal status, where appropriate. The dotted vertical line represents the overallresult for all parous women; information on each characteristic listed wasnot necessarily available for all women and averages of the subgroup-specific relative risks might therefore differ slightly from the result for all THE LANCET • Vol 360 • July 20, 2002 • www.thelancet.com For personal use. Only reproduce with permission from The Lancet Publishing but none of the estimates variedsignificantly according to the factors developing countries (␹2 0·2; p=0·7) by age at diagnosis (␹2 0·5; p=0·5), menopausal status (␹2 0·2; p=0·7), or family history of breast cancer (␹2 0·3; p=0·6). Although the magnitude of the decline of the relative risk designs (figure 1: ␹2 for heterogeneity breastfeeding for tumours localised to the beyond the breast (decline in relative risk of Cumulative incidence of breast cancer per 100 women [1·7%], respectively, for each year ofbreastfeeding). There was no significant difference in the extent of tumour spread among women with breast cancer (␹2 2·7; Figure 5: Estimated cumulative incidence of breast cancer in developed countries if women had family sizes and breastfeeding patterns typical for developingcountries Cumulative incidence of breast cancer *Cumulative incidence of breast cancer typical for women in developed countries around 1990;1–4,52 †estimated incidence of breast cancer in developed countries if women had, on breast cancer up to age 70 years was 5–7 average, 6·5 births instead of 2·5, and if women breastfed each child, on average, for 24 months instead of a lifetime mean of 8·7 months; such values have been typical of developingand developed countries until recently.53,54 1–2 per 100 in Asian and Africancountries.52 Women who would have beenof childbearing age during the 1950s make estimate are wide. Likewise, the decline in the relative risk substantial contributions to these estimates of cumulative of breast cancer associated with breastfeeding does not risk. In 1955–60 women in developed countries had, on vary significantly according to the age women were when average, between two and three births, whereas women in they began childbearing (webtable 1 on The Lancet Asian and African countries had, on average, between six website, figure 4; ␹2 for heterogeneity 1·5; p=0·7). and seven births.53 Among parous women from developed The relative risk of breast cancer declined by 3·4% countries in this study, the average duration of (0·9%; p<0·0001) for each child breastfed. However, this breastfeeding was about 3 months per child (the lifetime association seems to be secondary to breastfeeding average duration of breastfeeding was 8·7 months for duration, since additional stratification by lifetime duration parous controls from developed countries), which of breastfeeding substantially reduced the ␹2 test for trend contrasts with a median duration of breastfeeding of with number of children breastfed, from 15·9, down to a around 24 months per child in rural areas of Asia and non-significant value of 0·9. Conversely, the association of breast cancer with increasing duration of breastfeeding To assess the contribution of the small family sizes and persisted after adjustment for the number of children short lifetime duration of breastfeeding to breast cancer breastfed (3·8% [1·0%] decline in the relative risk for each incidence in developed countries around 1990, the 12 months breastfeeding; p<0·0001).
relative risks obtained here were applied to the age- The effect of ten other potential confounding factors specific incidence rates typical for developed countries at (ethnic origin, education, family history of breast cancer, that time.1,2 Figure 5 shows the cumulative incidence of age at menarche, height, weight, body-mass index, and breast cancer in developed countries and estimated use of hormonal contraceptives, alcohol, and tobacco) cumulative incidence under the assumption that each on the trends shown in figure 3 was examined. woman had, on average, 6·5 births instead of 2·5, and that Additional adjustment for each of these factors in turn did women breastfed each child for 24 months instead of not materially alter the magnitude of the effect of 3 months. The contribution to the estimated reduction in breastfeeding on the relative risk of breast cancer (see the incidence of breast cancer from the additional births— webtable 2 at http://image.thelancet.com/extras/01art9187 ie, without breastfeeding—is distinguished from the contribution from breastfeeding itself in figure 5. Overall,the larger family sizes and longer lifetime duration of breastfeeding typical in developing countries until recently The magnitude of the decline in the relative risk of breast are estimated to more than halve the cumulative incidence cancer associated with each year of breastfeeding was of breast cancer in developed countries, from 6·3 to 2·7 calculated separately for various subgroups of women, per 100 women by age 70 years. Part of this estimated including women from developed and developing reduction in incidence is due to the additional births, but countries, women of different ages, ethnic origins, familial almost two-thirds is due to the longer lifetime duration of patterns of disease, and 11 other possibly relevant factors, THE LANCET • Vol 360 • July 20, 2002 • www.thelancet.com For personal use. Only reproduce with permission from The Lancet Publishing Lifetime duration of breastfeeding is closely related to Our analyses here show that the relative risk of breast the number of children breastfed, and when the data are cancer is reduced by 4·3% (95% CI 2·9–5·8) for each year additionally stratified by the number of children breastfed that a woman breastfeeds, in addition to a reduction of the trend for the risk of breast cancer associated with 7·0% (5·0–9·0) for each birth. These relations are increasing duration of breastfeeding remained significant.
significant and are seen consistently for women from By contrast, the apparent association between breast developed and developing countries, of different ages and cancer and the number of children breastfed was no longer ethnic origins, and with various childbearing patterns and significant after the data were additionally stratified by The 47 studies that contributed data were of different Potential confounding by other factors such as age, designs and included women with a wide range of study (and centre within study), and menopausal status are reproductive and breastfeeding patterns. Since the minimised by stratification. The fine stratification used in decrease in the relative risk of breast cancer is these analyses means that no direct comparisons are made comparatively small for each year of breastfeeding, some between women in one study and women in another, and studies would, by chance alone, find significant that breastfeeding patterns in women with breast cancer associations and others would not; this possibility is are compared only with the pattern in women of exactly particularly true for studies done in North America and the same age and parity, with a similar age at first birth and many European countries, where women tend to age at menopause. Although the stratification was fine breastfeed for a short time, if at all. When all studies are enough to avoid substantial confounding, it was not considered together, there was no significant variation in excessively fine, since much of the statistical information the results across study design; there was, however, some content remained (an example of how to calculate the variation between studies, which could be due to amount of statistical information lost by stratification is differences between studies in the way that breastfeeding given elsewhere4). Potential confounding by ten additional was defined. The overall results were not affected unduly factors was examined, but separate adjustment for each in by a single study or group of studies.
turn did not materially alter the relative risk estimates As far as can be ascertained, more than 80% of the (see webtable 2 at http://image.thelancet.com/extras/ worldwide epidemiological data on breast cancer and 01art9187webtable2.pdf). Furthermore, the results did breastfeeding are included in this collaboration, and the not suggest that any of the factors examined, including findings from the ten studies that are not included55–64 are age, weight, family history of breast cancer, and generally consistent with these results. In the 1920s, Lane- menopausal status, significantly modified the magnitude of Claypon in the UK55 reported that the children of women the relative risks, although there is limited power to detect with breast cancer were less likely to have been breastfed for 1 year or longer than the children of controls (19%, Most women reliably report the number of children they 172 of 921, vs 33%, 457 of 1392). The findings of a have had, and hence differential reporting of births by cases parallel study by Wainwright in the USA56 showed a and controls, or misclassification of parity is unlikely to be a smaller difference in the same direction (28%, 472 of serious problem. Even though stillbirths are not counted 1714, vs 29%, 718 of 2451). Six of the eight other studies among the births for some studies, they represent about 1% not included here57–62 published results on ever versus never of all births, and so the omission of stillbirths would have breastfeeding adjusted by age, parity, and age at first birth, little effect on the findings with respect to parity. By and the combined relative risk for ever having breastfed contrast, women’s reporting of the length of time that they from these studies is 0·93 (95% CI 0·87–1·00; p=0·05).
breastfed is not so accurate, and there is a strong tendency The other two studies63,64 presented results according to for women to round to the nearest 6 months. Studies done duration of breastfeeding only, and the relative risk of in developing countries have shown that, even when breast cancer was seen to decline with increasing duration women gave birth in the previous 3–5 years, they still tend to report their duration of breastfeeding as multiples of 6 or12 months,54 and that women who are educated tend to overestimate the length of time that they breastfed.65 Most When studying the effect of each birth on the risk of breast women included in this collaborative reanalysis would have cancer, potential confounding by breastfeeding can be given birth decades before they were asked about their eliminated by looking at the relation between parity and breastfeeding practices, and no published data could be the relative risk of breast cancer in women who never found investigating reporting errors in lifetime duration of breastfed. When this is done, it is clear that, in the absence breastfeeding in such circumstances. The inevitable of breastfeeding, each birth has an independent effect, misclassification of women would, if anything, be expected reducing the relative risk of breast cancer by 7·0% per to lead to an underestimation of the true effect of birth. When studying the effect of breastfeeding, however, there is potentially extensive confounding by parity and, to There is no strong evidence in these data to suggest a lesser extent, by age at first birth. All analyses that there might be differential recall or reporting of examine the risk of breast cancer in relation to lifetime breastfeeding by cases and controls, since the results from duration of breastfeeding have therefore stratified women cohort studies, in which breastfeeding details were into eight groups according to the number of births they collected prospectively, are in line with results from case- had (1, 2 . . . 7, 8+), thereby effectively eliminating control studies, in which information was collected confounding by parity, and stratified further according to retrospectively (relative declines of 4·6% [1·8%] and 4·1% their age at first birth (<20, 20–24, 25–29, and у30), [0·7%], respectively, figure 1). The results for tumour thereby minimising counfounding due to that factor.
stage show little difference in the extent of tumour spread Stratification of the data by finer divisions of age at first according to duration of breastfeeding, which also suggests birth did not substantially alter the results. The trends that there is little or no differential detection of breast according to duration of breastfeeding do not vary cancer according to breastfeeding practices.
significantly by parity or age at first birth, indicating no About half the women included in these analyses had strong interaction with these factors (figure 4).
breastfed for a total of 6 months or less, with only 7% of THE LANCET • Vol 360 • July 20, 2002 • www.thelancet.com For personal use. Only reproduce with permission from The Lancet Publishing the cases and 15% of the controls breastfeeding for longer by mimicking the effect of breastfeeding therapeutically or than 30 months. No distinction was made between in some other way. In the meantime, important reductions exclusive breastfeeding and breastfeeding with in breast-cancer incidence could be achieved if women supplementary feeds; this, taken together with likely considered breastfeeding each child for longer than they do measurement errors in the reported lifetime duration of now. About 470 000 women in developed countries and breastfeeding, and the limited statistical power, means that 320 000 women in developing countries were diagnosed there remains some uncertainty about the magnitude of with breast cancer in 1990.67 Based on the estimates the protective effect of breastfeeding on the risk of breast obtained here, if women in developed countries had 2·5 cancer. Measurement errors would, if anything, be children, on average, but breastfed each child for 6 months expected to result in an underestimation of the true effect longer than they currently do, about 25 000 (5%) breast of breastfeeding on breast cancer. Because breastfeeding cancers would be prevented each year, and if each child patterns could well account for a large part of the variation were breastfed for an additional 12 months about 50 000 in breast-cancer incidence between developed and (11%) breast cancers might be prevented annually. There developing countries, there is a need for further research are obvious economic and social consequences to on this topic. Future epidemiological studies need to focus prolonging breastfeeding, and these results indicate that on populations in which breastfeeding was common for there are benefits to the mother, as well as the known relatively long durations, and should attempt to collect information on the use of supplementary feeds and onerrors in the reporting of lifetime duration of breastfeeding.
Members of the collaborative group on hormonal factors in breastcancer Although this collaboration was not set up to consider Analysis and writing committee—V Beral, D Bull, R Doll, R Peto, mechanisms, laboratory research to elucidate how breastfeeding protects against breast cancer is of direct Steering Committee—D Skegg (chairman), G Colditz, B Hulka, public-health relevance, since it might be possible to C La Vecchia, C Magnusson, T Miller, B Peterson, M Pike, D Thomas, F van Leeuwen.
prevent a substantial proportion of breast cancers in Collaborators are listed in the webappendix at developed countries if it were possible to mimic the effects http://image.thelancet.com/extras/01art9187webappendix.pdf Conflict of interest statementNone declared.
Public-health implicationsApplication of our results to incidence rates typical of developed countries around 1990 suggests that major We thank the women with and without breast cancer who took part in this reasons for the high incidence rates of breast cancer in research. Central pooling, checking, and analysis of data was supported by such countries are the small family size and the short Cancer Research UK, and the UNDP/UNFPA/WHO/World Bank specialprogramme of research, development and research training in human duration of breastfeeding that were characteristic of women in these countries during the past century. Indeed,if women had larger family sizes and longer lifetimedurations of breastfeeding that were typical of developing countries until recently, the cumulative incidence of breastcancer in developed countries is estimated to be reduced Collaborative Group on Hormonal Factors in Breast Cancer. Breastcancer and hormonal contraceptives: collaborative reanalysis of by more than half (from 6·3 to 2·7 per 100 women) by age individual data on 53 297 women with breast cancer and 100 239 70 years. Part of this estimated reduction in incidence is women without breast cancer from 54 epidemiological studies. Lancet due to the large family size, but almost two-thirds is due to 1996; 347: 1713–27.
breastfeeding. Thus, much of the difference in breast- Collaborative Group on Hormonal Factors in Breast Cancer. Breastcancer and hormonal contraceptives: further results. Contraception 1996; cancer incidence between developed and developing 54 (suppl 3): 1S–106S.
countries seems to be accounted for by these reproductive Collaborative Group on Hormonal Factors in Breast Cancer. Breast factors, since the cumulative risk of breast cancer by age 70 cancer and hormone replacement therapy: collaborative reanalysis of in most Asian and African countries around 1990 was data from 51 epidemiological studies of 52 705 women with breast between 1 and 2 per 100 women.52 Nevertheless, the cancer and 108 411 women without breast cancer. Lancet 1997; 350:
1047–59.
average family size and duration of breastfeeding are Collaborative Group on Hormonal Factors in Breast Cancer. declining rapidly in such countries,54,66 and breast-cancer Familial breast cancer: collaborative reanalysis of individual data incidence has subsequently begun to rise, especially in from 52 epidemiological studies including 58 209 women with breast cancer and 101 986 women without the disease. Lancet 2001;
358: 1389–99.
For women to have a lifetime duration of breastfeeding Vessey M, Baron J, Doll R, McPherson K, Yeates D. Oral that was typical of women in developing countries and yet contraceptives and breast cancer: final report of an epidemiologic study.
have the small family sizes typical for women in developed Br J Cancer 1983; 47: 455–62.
countries would be virtually impossible. However, in that 6 Lê MG, Bachelot A, Doyon F, Kramar A, Hill C. Oral hypothetical situation, these findings suggest that the contraceptive use and breast cancer or cervical cancer: preliminaryresults of a French case-control study. In: Hormones and sexual factors incidence of breast cancer in developed countries would be in human cancer aetiology. Amsterdam: Elsevier Science, 1984.
reduced by 42%, solely by the longer duration of Lubin F, Wax Y, Modan B. Role of fat, animal protein and dietary fibre breastfeeding. Thus, the short duration of breastfeeding in breast cancer aetiology: a case-control study. J Natl Cancer Inst 1986; typical of women in developed countries makes a major 77: 605–12.
contribution to the high incidence of breast cancer in these Meirik O, Lund E, Adami HO, Bergstrom R, Christoffersen T, BergsjoP. Oral contraceptive use and breast cancer in young women: a joint national case-control study in Sweden and Norway. Lancet 1986; 2:
To expect that substantial reductions in breast-cancer incidence could be brought about today by women McPherson K, Vessey MP, Neil A, Doll R, Jones L, Roberts M. Early returning to the pattern of childbearing and breastfeeding oral contraceptive use and breast cancer: results of another case-control that typified most societies until a century or so ago is study. Br J Cancer 1987; 56: 653–60.
10 Ewertz M, Duffy SW. Risk of breast cancer in relation to reproductive unrealistic. However, if in the future the mechanism of the factors in Denmark. Br J Cancer 1988; 58: 99–104.
protective effect of breastfeeding on breast cancer were 11 Rohan TE, McMichael AJ. Oral contraceptive agents and breast cancer: understood, it might be possible to prevent breast cancer a population-based case-control study. Med J Aust 1988; 149: 520–26.
THE LANCET • Vol 360 • July 20, 2002 • www.thelancet.com For personal use. Only reproduce with permission from The Lancet Publishing 12 Yuan J-M, Yu MC, Ross RK, Gao Y-T, Henderson BE. Risk factors for experience and breast cancer risk among postmenopausal women.
breast cancer in Chinese women in Shanghai. Cancer Res 1988; 48:
Cancer Epidemiol Biomarkers Prev 1998; 7: 365–69.
41 McCredie MRE, Dite GS, Giles GG, Hopper JL. Breast cancer in 13 Layde PM, Websterm LA, Baughman AL, Wingo PA, Rubin GL, Australian women under the age 40. Cancer Causes Control 1998; 9:
Ory HW, and The Cancer and Steroid Hormone Study Group. The independent associations of parity, age at first full term pregnancy, and 42 Hirose K, Tajima K, Hamajima N, et al. Comparative case-referent duration of breastfeeding with the risk of breast cancer. J Clin Epidemiol study of risk factors among hormone-related female cancers in Japan.
1989; 42: 963–73.
Jpn J Cancer Res 1999; 90: 255–61.
14 Rosero-Bixby L, Oberle MW. Fertility change in Costa Rica 1960–84: 43 Hopper JL, Chenevix-Trench G, Jolley D, et al. Design and analysis analysis of retrospective lifetime reproductive histories. J Biosoc Sci issues in a population-based case-control-family study of the genetic 1989; 21: 419–32.
epidemiology of breast cancer, and the Co-operative Family Registry for 15 Siskind V, Schofield F, Rice D, Bain C. Breast cancer and breast Breast Cancer Families (CFRBCS). Monogr Nat Cancer Inst 1999; 26:
feeding: results from an Australian case-control study. Am J Epidemiol 1989; 130: 229–36.
44 Magnusson C, Baron JA, Correia N, Bergstrom R, Adami H-O, Persson 16 UK National Case-Control Study Group. Oral contraceptive use and I. Breast cancer risk following long-term oestrogen- and oestrogen- breast cancer risk in young women. Lancet 1989; 1: 973–82.
progestin-replacement therapy. Int J Cancer 1999; 81: 339–44.
17 Paul C, Skegg DCG, Spears GFS. Oral contraceptives and risk of breast 45 Million Women Study Collaborative Group. The Million Women cancer. Int J Cancer 1990; 46: 366–73.
Study: design and characteristics of the study population. Breast Cancer 18 Clavel F, Andrieu N, Gairard B, et al. Oral contraceptives and breast Research 1999; 1: 73–80.
cancer: a French case-control study. Int J Epidemiol 1991; 20: 32–38.
46 Chang-Claude J, Eby N, Kiechle M, Bastert G, Becher H.
19 Weinstein AL, Mahoney MC, Nasca PC, Leske MC, Varma AO. Breast Breastfeeding and breast cancer risk by age 50 among women in cancer risk and oral contraceptive use: results from a large case-control Germany. Cancer Causes Control 2000; 1: 687–95.
study. Epidemiology 1991; 2: 353–58.
47 Gajalakshmi V. Diet and cancers of the stomach, breast and lung. Asian 20 Lee HP, Gourley L, Duffy SW, Esteve J, Lee J, Day NE. Risk factors for Pacific J Cancer Prev 2000; 1 (suppl): 39–43.
breast cancer by age and menopausal status: a case-control study in 48 Gao YT, Shu XO, Dai Q, et al. Menstrual, reproductive factors and Singapore. Cancer Causes Control 1992; 3: 313–22.
breast cancer risk in urban Shanghai, People’s Republic of China. Int J 21 Wang Q-S, Ross RK, Yu MC, Ning J-P, Henderson BE, Kimm HT. A Cancer 2000; 87: 295–300.
case-control study of breast cancer in Tianjin, China. Cancer Epidemiol 49 Tryggvadóttir L, Tulinius H, Eyfjord JE, Sigurvinsson T. Breastfeeding Biomarkers Prev 1992; 1: 435–39.
and reduced risk of breast cancer in an Icelandic cohort study on breast 22 Yang CP, Daling JR, Band PR, Gallagher RP, White E, Weiss NS. Non cancer. Am J Epidemiol 2001; 154: 37–42.
contraceptive hormone use and risk of breast cancer. 50 Peto R, Pike M, Armitage P, et al. Design and analysis of randomized Cancer Causes Control 1992; 3: 475–79.
clinical trials requiring prolonged observation of each patient. Br J 23 Ferraroni M, Gerber M, Decarli A, et al. HDL-cholesterol and breast Cancer 1976; 34: 585–612.
cancer: a joint study in Northern Italy and Southern France. 51 Easton DF, Peto J, Babiker AGAG. Floating absolute risk: an Int J Epidemiol 1993; 22: 772–80.
alternative to relative risk in survival and case-control analysis avoiding 24 Kalache A, Maguire A, Thompson SG. Age at last full-term pregnancy an arbitrary reference group. Stat Med 1991; 10: 1025–35.
and risk of breast cancer. Lancet 1993; 341: 33–36.
52 Parkin DM, Whelan SL, Ferlay J, et al. Cancer incidence in five 25 Morabia A, Szklo M, Stewart W, Schuman L, Thomas DB. Consistent continents. Lyon: IARC Scientific Publication, 1997.
lack of association between breast cancer and oral contraceptives using 53 Cho LJ. Estimated refined measures of fertility for all major countries of either hospital or neighborhood controls. Prev Med 1993; 22: 178–86.
the world. Demography 1964; 1: 359–74.
26 Thomas DB, Noonan EA, and the WHO Collaborative Study of 54 Anon. Breastfeeding and complementary infant feeding, and the Neoplasia and Steroid Contraceptives. Breast cancer and prolonged postpartum effects of breastfeeding. Demographic and Health Surveys lactation. Int J Epidemiol 1993; 22: 619–26.
Comprehensive Studies (1999) Number 30, Macro International Inc, 27 Land CE, Hayakawa N, Machado SG, et al. A case-control interview study of breast cancer among Japanese A-bomb survivors, II: 55 Lane-Claypon JE. A further report on cancer of the breast, with special interactions with radiation dose. Cancer Causes Control 1994; 5: 167–76.
reference to its associated antecedent conditions. In: Reports on public 28 Newcomb PA, Storer BE, Longnecker MP, et al. Lactation and a health and medical subjects. Number 32. London: Ministry of Health, reduced risk of premenopausal breast cancer. N Engl J Med 1994; 330:
56 Wainwright JM. A comparison of conditions associated with breast 29 Rookus MA, van Leeuwen FE, for the Netherlands Oral Contraceptives cancer in Great Britain and America. Am J Cancer 1931; 15: 2610–45.
and Breast Cancer Study Group. Oral contraceptives and risk of breast 57 Michels KB, Willett WC, Rosner BA, et al. Prospective assessment of cancer in women aged 20–54 years. Lancet 1994; 344: 844–51.
breastfeeding and breast cancer incidence among 89 887 women. Lancet 30 White E, Malone KE, Weiss NS, Daling JR. Breast cancer among young 1996; 347: 431–36.
US women in relation to oral contraceptive use. J Natl Cancer Inst 1994; 58 Romieu I, Hernández-Avila M, Lazcano E, Lopez L, 86: 505–14.
Romero-Jaime R. Breast cancer and lactation history in Mexican 31 Brinton LA, Daling JR, Liff JM, et al. Oral contraceptives and breast women. Am J Epidemiol 1996; 143: 543–52.
cancer risk among younger women. J Natl Cancer Inst 1995; 87:
59 Freudenheim JL, Marshall JR, Vena JE, et al. Lactation history and breast cancer risk. Am J Epidemiol 1997; 146: 932–38.
32 La Vecchia C, Negri E, Franceschi S, et al. Oral contraceptives and 60 Stuver SO, Hsieh C-C, Bertone E, Trichopoulos D. The association breast cancer: a cooperative Italian study. Int J Cancer 1995; 60:
between lactation and breast cancer in an international case-control study: a re-analysis by menopausal status. Int J Cancer 1997; 71:
33 Lipworth L, Katsouyanni K, Stuver S, Samoli E, Hankinson SE, Trichopoulos D. Oral contraceptives, menopausal estrogens, and the 61 Gilliland FD, Hunt WC, Baumgartner KB, et al. Reproductive risk risk of breast cancer: a case-control study in Greece. Int J Cancer 1995; factors of breast cancer in Hispanic and non-Hispanic white women.
62: 548–51.
Am J Epidemiol 1998; 148: 683–92.
34 Levi F, Pasche C, Lucchini F, La Vecchia C. Alcohol and breast cancer 62 Coogan PF, Rosenberg L, Shapiro S, Hoffman M. Lactation and breast in the Swiss Canton of Vaud. Eur J Cancer 1996; 32A: 2108–13.
carcinoma risk in a South African population. Cancer 1999; 86: 982–89.
35 Rossing MA, Stanford JL, Weiss NS, Habel LA. Oral contraceptive use 63 Furberg H, Newman B, Moorman P, Millikan R. Lactation and breast and risk of breast cancer in middle-aged women. Am J Epidemiol 1996; cancer risk. Int J Epidemiol 1999; 28: 396–402.
144: 161–64.
64 Zheng T, Duan L, Zhang B, et al. Lactation reduces breast cancer risk 36 Viladiu P, Izquierdo A, de Sanjose S, Bosch FX. A breast case-control in Shandong Province, China. Am J Epidemiol 2000; 152: 1129–35.
study in Girona, Spain: endocrine, familial and lifestyle factors. 65 Huttly SRA, Barros FC, Victora CG, Beria JU, Vaughan JP. Do Eur J Cancer 1996; 5: 329–35.
mothers overestimate breastfeeding duration? An example of recall bias 37 Enger SM, Ross RK, Bernstein L. Breastfeeding history, pregnancy from a study in Southern Brazil. Am J Epidemiol 1990; 132: 572–75.
experience and risk of breast cancer. Br J Cancer 1997; 76: 118–23.
66 Anon. Demographic and Health Surveys, 28: fertility levels, trends and 38 Thomas DB, Gao DL, Self SG, et al. Randomized trial of breast self- differentials. Calverton: Macro International, 1998.
examination in Shanghai: methodology and preliminary results. J Natl 67 Parkin DM, Pisani P, Ferlay J. Estimates of the worldwide incidence of Cancer Inst 1997; 89: 355–65.
25 major cancers in 1990. Int J Cancer 1999; 80: 827–41.
39 Thomas HV, Key TJ, Allen DS, et al. Reversal of relation between body 68 WHO Collaborative Study Team on the Role of Breastfeeding on the mass and endogenous estrogen concentrations with menopausal status.
Prevention of Infant Mortality. Effect of breastfeeding on infant and J Natl Cancer Inst 1997; 89: 396–97.
child mortality due to infectious diseases in less developed countries: a 40 Enger SM, Ross RK, Paganini-Hill AL, Bernstein L. Breastfeeding pooled analysis. Lancet 2000; 355: 451–55.
THE LANCET • Vol 360 • July 20, 2002 • www.thelancet.com For personal use. Only reproduce with permission from The Lancet Publishing

Source: http://www.geburtsstaette.ch/stillen_brustkr.pdf