Lactation and cancer of the breast

Bull. Org. mond. Sant 1970, 42, 185-194 Bull. Wld Hlth Org. Lactation and Cancer of the Breast A Summary of an International Study * B. MACMAHON,' T. M. LIN,2 C. R. LOWE,3 A. P. MIRRA,4 B. RAVNIHAR,r E. J. SALBER,6 D. TRICHOPOULOS,7 V. G. VALAORAS 8 & S. YUASA 9 An international collaborative study has been carried out to test the hypothesis that prolonged lactation protects women against cancer of the breast. While pregnancy itself seemed to confer some protection against breast cancer in all areas studied, no consistent differences in duration oflactation werefound between breast cancer patients and unaffected women, once the fact that breast cancer patients have fewer pregnancies had been allowed for. Even in areas where some women had lactated for a total of S years or more, such women occurred proportionately no less frequently among breast cancer patients than among unaffected women. In the light of this and other recent evidence, it is unlikely that lactation has any protective effect against breast cancer in women, and other explanations must be sought for the remarkable international differences in the frequency of this disease. The hypothesis that prolonged lactation protects against the development of cancer of the breast is one of the oldest and most enduring hypotheses concerning the etiology of this neoplasm. It could account for several of the major epidemiological features of the disease, including the following: (1) Most of the countries in which very low rates of breast cancer have been observed are in areas of * This study was supported by Grant E-385A from the American Cancer Society, Grant 402-C-200 from the Boris Kidric Fund of Yugoslavia, a grant from the Medical Research Council of Great Britain, a grant from the Ministry of Health and Welfare, Japan, Grant 5 P01 CA 06373 from the US National Cancer Institute, a grant from the National Council for Science of China (Taiwan) and Grants R/00057, R/00062 and C2/181/24 from the World Health Organization. 1 Professor and Head, Department of Epidemiology, Harvard School of Public Health, Boston, Mass., USA. 2 Associate Professor, Department of Epidemiology, Col- lege of Medicine, National Taiwan University, Taipei, Taiwan. ' Mansel Talbot Professor, Department of Social and Occupational Medicine, Welsh National School of Medicine, University of Wales, Cardiff, Wales. 'Director, Central Cancer Registry, Sao Paulo, Brazil. 'Professor, Institute of Oncology, Medical Faculty, University of Ljubljana, Yugoslavia. ' Senior Research Associate, Department of Epidemio- logy, Harvard School of Public Health, Boston, Mass., USA. 7Lecturer, Department of Epidemiology, Harvard School of Public Health, Boston, Mass., USA. ' Professor, Department of Hygiene and Epidemiology, University of Athens, Athens, Greece. ' Lecturer, Department of Epidemiology, Institute of Public Health, Tokyo, Japan. the world where lactation is customary and frequently prolonged. (2) Breast cancer risk is inversely associated with high parity, a status that provides increased oppor- tunity for lactation. (3) There has been a significant increase in breast cancer rates in young women in the USA in recent years (MacMahon, 1958; Feinleib & Garrison, 1969). The fact that this increase has affected successive cohorts of women, rather than particular time periods, suggests that it results from some experience during early life; the decreasing frequency of breast feeding is one possibility. Moreover, credible mechanisms can be postulated to explain a relationship between long lactation and low breast cancer risk. These include local effects of breast feeding on mammary tissue, as well as hormonal mechanisms dependent on the effect of lactation in changing endocrine patterns. Recent studies designed to test this hypothesis by comparing the lactation experience of breast cancer patients and controls have given conflicting results. Most of these studies have been undertaken in the USA and in other countries where breast cancer rates are high. Even if studies in the countries of highest risk were consistently negative, their inter- pretation could be questioned. It is conceivable, for example, that the duration of lactation necessary to 2463 -185- B. MACMAHON AND OTHERS confer protection against breast cancer is so long that few women in such countries experience it; virtually all women inay therefore be at maximum risk with respect to this factor, and the variables that discriminate between affected and unaffected women would be other components of the causal complex. Therefore, before considering alternative explana- tions for the international variation in breast cancer rates, it seemed desirable to provide a definitive test of the lactation hypothesis. We considered that such a test should: (1) include areas exhibiting the whole range of observed breast cancer rates, and (2) be based on series of cases known to be representative of all those occurring in the areas rather than on the patients coming to specialized, or otherwise selected, hospitals. METHOD The basic design of the study was that of a case- control study, information being obtained by inter- viewing breast cancer patients and a comparison group of unaffected women. This approach was used in 7 centres, 2 being areas of high breast cancer rates, 2 areas of low rates, and 3 areas where inter- mediate levels of risk were thought to prevail. The centres were as follows: High risk: the city of Boston, USA, including certain suburbs, and the county of Glamorgan, Wales, including the 3 county boroughs of Cardiff, Swansea and Merthyr Tydfil. intermediate risk: the city of Greater Athens, Greece, the Socialist Republic of Slovenia, Yugoslavia, and the city of Sao Paulo, Brazil. Low risk: the prefecture of Tokyo (Tokyo-to), Japan, and the city and county of Taipei, Taiwan. As judged from available mortality data, these centres represented a 6-fold range of breast cancer rates-from about 26 per 100 000 in the USA to about 4 per 100 000 in Japan and Taiwan. The features of the study design which were to be common to all centres included the following: (1) For each centre, a geographically and admin- istratively defined population would be selected. (2) An attempt would be made to identify all residents of the defined population having a diag- nosis of breast cancer established for the first time during a specified period. (3) Interviews would be attempted with such patients during the period of hospitalization. Patients who left hospital prior to interview would not be interviewed at home, although they would be recorded as incident cases. (4) As controls, patients under treatment for dis- eases other than breast cancer were to be interviewed. For each breast cancer patient the 3 eligible patients in the beds closest to the cancer patient at the time of her interview would be selected. Eligibility required also being a resident of the defined population and over 35 years of age (unless the breast cancer patient was under 35 years of age, in which event a specific age match was required). In some instances, 3 eligible controls could not be obtained for each cancer case. (5) Generally, the same interviewer would inter- view all members of a " set" consisting of 1 breast cancer patient and 3 controls. (6) The same interview form, translated as re- quired, would be used in all centres. (7) At the end of the study, hospital records would be searched for patients who might have been missed during the interviewing period. (8) A histological specimen would be obtained from each patient and sent to the co-ordinating centre for review by a single pathologist. (9) Coding, data processing and analysis for all centres would be carried out in the co-ordinating centre. In addition, investigators in each centre would be free to carry out whatever analyses of their own data they wished. The extent to which these procedures could be followed in the individual centres, as well as the details of the populations surveyed and numbers of cases detected, are described in individual reports (Lowe & MacMahon, 1970; Salber, Trichopoulos & MacMahon, 1969; Valaoras et al., 1969; Yuasa & MacMahon, 1970; and Lin, Chen & MacMahon, Mirra & Cole, Ravnihar, MacMahon & Lindtner, unpublished data). The only major discrepancy bearing on the data to be presented in this paper is that in Tokyo and in Sao Paulo it proved impos- sible to identify all cases occurring in the defined population during the study period, and incidence cannot be estimated for these two centres. Never- theless, the numbers of cases identified in Tokyo and Sao Paulo, and the variety of institutions from 186 LACTATION AND CANCER OF THE BREAST TABLE I TOTAL FEMALE POPULATIONS, NUMBERS OF CASES DETECTED AND PATIENTS INTERVIEWED IN THE CO-OPERATING CENTRES N. No. of patients Bes Total female Study period of breast cinterviewed ancerCentre population (month and year cancer Cor- patientsinclusive) cases Breast paio intervieweddetected cancer patients | Boston 438 264 1.65-12.66 758 606 1 807 79.9 Glamorgan 629 278 5.65- 4.67 710 619 1 850 87.2 Athens 1 093 269 1.65- 6.67 956 799 2 470 83.6 Slovenia 865 829 10.64-12.67 831 772 2 308 92.9 Sao Paulo 2 458 556 12.65- 6.68 537 537 1 555 -a Taipei 1 265 984 11.64-10.67 232 213 648 91.8 Tokyo 5 304 661 4.65- 3.67 861 849 2 250 -a Total 12 055 841 - 4 885 4 395 12 888 a Total incident cases not known. which they came, give reasonable assurance that the series are representative of all affected women in the two centres. The numbers of cases detected and patients inter- viewed are given in Table 1. Altogether, 4885 cases were identified and 4395 patients were interviewed. Nearly 13 000 control patients were also interviewed. The proportion of all new cases identified for which interviews could be arranged was 80% or higher in all centres except the two where total ascertainment of cancer cases was not feasible. In these two areas, on the basis of other information, it is estimated that the patients interviewed comprised about 50% of the total incident cases in Tokyo (Yuasa & MacMahon, 1970) and about 60% of those in Sao Paulo. The interviewers were asked to give an opinion on the validity of the information obtained at each interview. In the analyses, interviews rated as unreli- able have been excluded. The numbers of such interviews were small (56 cases, 128 controls) and they occurred in similar proportions in the cases and controls. The highest frequencies were in Boston, where 4.3% of the cases and 3.9% of the controls were rated as unreliable. For 3805 patients (78% of the total of 4885 cases), the diagnosis of cancer was confirmed in histological material sent to the co-ordinating centre. The remaining patients have been included on the basis of local diagnoses alone. In no centre did the pro- portion of cases without histological confirmation, either local or central, exceed 6% of the total. The statistical method used was the comparison of observed values in the case series with expected values based on the control series for the correspond- ing centre. The expected numbers were computed by applying, to the number of cases in each 10-year age-group, the values for the controls in the same age- group, and summing these over all age-groups. All the expected values shown are adjusted for age in this manner. In certain tables, as indicated in the footnotes, the expected values also take account of differences in parity distributions of the compared series. Summary X2 tests with 1 degree of freedom, as described by Mantel & Haenszel (1959) and Mantel (1963) were used. In these tests, the expected values are not those shown in the table but are, of course, based on the distributions of the case and control groups combined. INCIDENCE OF BREAST CANCER For the five centres in which detection of breast cancer is believed to have been reasonably complete, the average annual incidence rates per 100 000 females in the population, age-adjusted to the stan- dard world population, suggested by Segi (1960) and Doll, Payne & Waterhouse (1966) were as follows: ':g7 B. MACMAHON AND OTHERS Centre Boston Glamorgan Athens Slovenia Taipei Incidence rate 55.0 38.8 28.9 24.4 9.1 The range between centres-from 55.0 per 100 000 in Boston to 9.1 per 100 000 in Taipei-is similar to that observed in mortality data. As already noted, it is not possible to give a direct estimate of incidence in Tokyo, the other low-risk centre. If the breast cancer incidence rates from the cancer registry in Miyagi Prefecture, Japan (Doll, Payne & Water- house, 1966), are adjusted to the same population, a rate of 12.9 per 100 000 is obtained. The actual rate in Tokyo is probably a little higher than this, since Miyagi Prefecture includes both urban and rural areas and breast cancer rates tend to be higher in urban areas. The accompanying figure shows the age-specific incidence curves for the five centres. These centres illustrate a phenomenon that is also evident in reports from cancer registries (Doll, Payne & Water- house, 1966)-namely, that the international dif- ferences in breast cancer risk are particularly strik- ing in the older age-groups. In Boston and Glamor- gan, rates increase progressively with age throughout life, but in the other centres the rates reach a plateau or decline after middle-age. AGE-SPECIFIC INCIDENCE RATES OF BREAST CANCER IN 5 STUDY CENTRES, 1965-67 O cu cn E 2 a) C= =1 4= FL -cc 20 30 Age (years) wHO 00087 MARITAL STATUS AND PARITY The analysis of lactation histories is restricted to married women who had at least 1 liveborn child. Single women were excluded because, in most cen- tres, they were not asked whether they had borne children. Within the married group, only women who had a liveborn child had an opportunity to lactate, and it is within this group that the effect of lactation can be separated from other concomitants of pregnancy. However, before turning to this group, the risks of breast cancer in excluded groups have been examined. The relationships involved are generally well known for this disease, although they have not previously been shown to prevail in all the areas included in this study. Observed and expected numbers of single women among the interviewed patients and estimates of the risks for single women relative to those for women ever married are shown in Table 2. Except in Taipei, TABLE 2 ESTIMATES OF THE RISK OF BREAST CANCER IN SINGLE WOMEN RELATIVE TO THAT IN WOMEN EVER MARRIED No. of cases in single Centre Nof women Relative ______ Observed Expected __ Boston 580 115 89.0 1.36 Glamorgan 611 61 44.3 1.42 Athens 796 81 61.0 1.37 Slovenia 756 122 106.1 1.18 SAo Paulo 536 73 52.2 1.46 Taipei 210 1 3.4 - Tokyo 849 66 45.6 1.49 a Number of interviewed patients after exclusion of un- reliable interviews and patients of unknown marital status. b Risk for the single women relative to a risk of 1.00 for the married women. where over-all numbers are small and those of single women particularly so, the risks were higher in single than in ever-married women. The excess amounts to between 35% and 50% in all centres except Slovenia, where it was somewhat lower- namely, 18%. It is relevant that in Slovenia 25% of the single women reported having had 1 or more children. This may account for some, though prob- ably not all, of the discrepancy between the data for Slovenia and the other centres. 188 LACTATION AND CANCER OF THE BREAST TABLE 3 ESTIMATES OF THE RISK OF BREAST CANCER IN NON-PAROUS MARRIED WOMEN RELATIVE TO THAT IN PAROUS MARRIED WOMEN No. No. of patients Centre o; non-parous b Relative patients a Observed Expected risk c Boston 463 88 68.0 1.36 Glamorgan 550 100 65.3 1.65 Athens 713 133 106.1 1.31 Slovenia 634 63 55.9 1.14 SAo Paulo 459 49 41.8 1.19 Taipei 209 32 12.9 2.75 Tokyo 783 158 109.3 1.56 a Cases of unknown parity excluded. b "1 Non-parous " defined as never having borne a full-term child, liveborn or stillborn. c Risk for the non-parous married women relative to a risk of 1.00 for the parous married women. Among the ever-married women, observed and expected numbers who never delivered a full-term child are given in Table 3. In all centres, risks were higher for the non-parous than for the parous women. Except in Sao Paulo, where the risk for single women relative to that for married women (1.46) was appreciably greater than that for non- parous relative to parous women (1.19), the relative risks for the individual centres given in Tables 2 and 3 are quite similar, suggesting that most, or perhaps all, of the excess risk of the single women can be attributed to their non-parity rather than to some other aspect of their marital status. In two centres, there was a statistically significant excess of stillbirths among the children of breast cancer patients (Table 4). While this relationship was not found in the other centres, it seemed prefer- able to omit women who had only stillborn children from the analysis of lactation, and they have in fact been excluded. It is true that women who have stillborn children usually do not lactate. However, even if a high stillbirth rate were a characteristic of women who develop breast cancer, mechanisms other than lack of opportunity for lactation can be postu- lated to explain such a relationship. Therefore, the inclusion of stillbirths in a test of the lactation hypo- thesis would, if they occurred in excess in sufficient numbers in one or the other series, necessarily TABLE 4 OBSERVED AND EXPECTED NUMBERS OF STILLBIRTHS AMONG THE CHILDREN OF THE BREAST CANCER PATIENTS; EVER-MARRIED WOMEN ONLY No. of No. of stillbirths Centre women full-term Ex -xwm births Observed pEcte Boston 465 1 069 37 35.3 0.1 Glamorgan 550 1189 74 57.1 3.8 Athens 715 1 564 65 46.5 4.4 Slovenia 634 1 809 51 55.8 0.3 Sao Paulo 459 1 748 49 46.2 0.0 Taipei 209 758 19 16.5 0.0 Tokyo 784 1 766 77 73.7 0.1 a Expected values based on the stillbirth rates in the appro- priate control group, with adjustment for age at the time of i nterview. b x2 values are Mantel-Haenszel summary x2 tests with 1 degree of freedom. For P = 0.05, x2 = 3.84. Values equal to or greater than 3.84 are in italics. introduce a difference in lactation patterns. In fact, the number of such women is too small to influence the data appreciably, and the question of their inclu- sion or exclusion is somewhat academic. LACTATION HISTORY The numbers of married women who had at least 1 livebom child, and a summary of their lacta- tion experience, are given in Table 5. Four measures of lactation were examined: (1) The number of women who had 1 or more liveborn children but who had never lactated. (2) The number of liveborn children who were not breast-fed. (3) The mean duration of lactation for children who were breast-fed. (4) The number of women whose total lactation time was unusually long. With respect to (4), lactation practices varied substantially between centres, and in order to obtain a group that was unusual for the centre, but yet large enough for meaningful analysis, it was neces- sary to adopt different values to define an " unusually long" total lactation experience. The use of " more than 5 years (60 months)" in Sao Paulo, Taipei and Tokyo and of " more than 2 years (24 months) " 189 B. MACMAHON AND OTHERS TABLE 5 SUMMARY OF FINDINGS WITH RESPECT TO SELECTED MEASURES OF LACTATION EXPERIENCE Boston Glamorgan Athens Slovenia Sco Paulo Taipei Tokyo No. of cases a 375 445 574 564 410 410 610 No. of controlsa 1259 1479 1893 1773 1 253 586 1817 No. who never lactated: Observed b 89 99 24 31 20 17 36 Expected 99.1 108.6 24.4 22.3 17.5 17.6 38.2 No. who lactated for long periods: c Observedb 49 58 231 166 104 50 127 Expected 46.1 60.6 246.3 156.2 113.0 57.5 130.3 No. of children:d of cases 1 034 1 115 1 489 1 759 1 699 739 1 690 of controls 3980 4113 5271 5597 5957 2547 6050 No. not breast fed: Observedb 384 340 90 133 121 104 130 Expected 418.2 378.4 79.2 107.5 129.8 111.4 120.6 Mean duration of breast feeding: e Observed b 6.2 6.5 11.6 7.4 12.3 13.2 15.2 Expected 5.7 6.6 12.3 7.2 12.0 14.6 15.2 a Married women with at least 1 liveborn child. b Observed values are those in the breast cancer patients. Expected values are based on the control series adjusted to the age and parity distribution of the patients. The observed values in italics differ significantly (P <0.05) from the expected. c In SAo Paulo, Taipei and Tokyo this is the number who nursed for more than 5 years (60 months). In the other centres it is the number who nursed for more than 2 years (24 months). d Liveborn children born to breast cancer patients or controls. e Mean duration of lactation (months) among children who were breast fed. in the other centres identified approximately 25% of the total group, rather less in some centres (13 % in Boston and Glamorgan) and rather more in others (40% in Athens). For each of these measures, the observed numbers in the breast cancer series were compared with expected values computed from rates in the corre- sponding control series specific for age (in decades) and parity (single parities up to 5 or more) simul- taneously. There are 28 comparisons of observed and ex- pected values in the table. In 13 of these 28 com- parisons the difference between observed and expected was in the direction favouring a protective effect of lactation (i.e., less lactation in the cases than expected), and in 14 instances the difference was in the opposite direction. In 1 instance the observed and expected values were identical. Summary x2 values were computed for all 28 comparisons. In 3 instances there was a statistically significant dif- ference between observed and expected values- 1 was in the direction favouring the hypothesis and 2 were contrary. However, all three statistically signifi- cant differences occurred in measure (2). An objec- tion may be raised to the application of the signifi- cance test to this particular measure, namely, that the probabilities of each of the included children being breast-fed are not independent since there is a tendency for all the children of a given mother to be treated similarly. The X2 tests, which assume independence, therefore underestimate the prob- abilities of the differences being due to chance. Effect of socio-economic status After parity and age, socio-economic status is per- haps the variable most likely to confound an analysis of lactation and breast cancer risk, since it has been shown on many occasions to be associated with both characteristics. The measure of socio-economic status most clearly related to breast cancer risk in our data was duration of the patient's schooling. There was indeed a strong inverse relationship between duration of schooling and duration of lacta- tion in the control series of each centre (Table 6). 190 LACTATION AND CANCER OF THE BREAST TABLE 6 PERCENTAGE OF MARRIED CONTROL PATIENTS WITH AT LEAST 1 LIVEBORN CHILD WHO LACTATED FOR RELATIVELY LONG PERIODS,a BY YEARS OF SCHOOLING Years of schooling b Total >16 6.7 11.1 (95) (1 269) - 12.7 (12) (1 493) - 44.5 (18) (1 914) - 28.2 (9) (1 781) - 32.7 (6) (1 261) 8.7 33.2 (38) (588) - 28.1 (17) (1 836) a In So Paulo, Taipei and Tokyo, this is the percentage who lactated for more than 5 years (60 months). In the other centres it is the percentage who lactated for more than 2 years (24 months). The percentages are adjusted to allow for differences in parity distribution between educational classes within centres but not between centres. b Values in parentheses are the numbers on which the estimates are based. No estimates are given for cells containing less than 25 women. This relationship was in the direction contrary to that which would lead us to overlook a true protec- tive effect of lactation against breast cancer. That is to say, breast cancer tended to occur more frequently in the higher socio-economic groups and, since women in such groups lactate less, it might have been expected that the patients would have had shorter total periods of lactation on this basis alone. Indeed, the relationships in Table 6 are so strong that we must inquire whether, by neglecting socio- economic status (Table 5), an association of lacta- tion with increased breast cancer risk has been over- looked. In Table 7, the observed and expected values from the bottom row of Table 5 are shown, together with the expected values computed with adjustments for age, parity and duration of schooling. The effect of the additional adjustments is trivial, and in 5 of the 7 centres serves only to move the expected values somewhat closer to the observed values. Individual age- and parity-groups A previous report from Japan has suggested that lactation may be a significant protective factor in TABLE 7 OBSERVED NUMBERS OF BREAST CANCER PATIENTS WHO BREAST FED AT LEAST I CHILD AND WHO LACTATED FOR RELATIVELY LONG PERIODS,a AND EXPECTED VALUES ADJUSTED FOR AGE, PARITY AND DURATION OF SCHOOLING a ~~~~Expected Centre Observed Ab BC Boston 49 46.1 45.2 Glamorgan 58 60.6 60.2 Athens 231 246.3 239.6 Slovenia 166 156.2 157.2 SAo Paulo 104 113.0 110.5 Taipei 50 57.5 59.2 Tokyo 127 130.3 128.5 a Same criteria as in Tables 5 and 6. b Expected with adjustment for age and parity, as in Table 5. c Expected with adjustment for age, parity and duration of schooling. Centre Boston Glamorgan Athens Slovenia Sao Paulo Taipei Tokyo None (24) (3) 50.5 (373) (13) 40.4 (406) 43.2 (160) (18) <8 28.8 (128) 14.8 (37) 44.5 (1 169) 29.4 (1 162) 29.7 (728) 34.7 (232) 39.6 (420) 8-11 13.0 (389) 12.7 (1 368) 43.0 (190) 22.9 (438) 8.3 (95) 21.6 (82) 25.4 (1 164) 12-15 4.4 (627) 5.8 (71) 26.4 (163) 27.7 (149) 1.4 (26) 18.5 (76) 7.8 (215) 191 192 B. MACMAHON AND OTHERS breast cancers developing in women over 40 years of age, but not in those occurring in younger women (Kamoi, 1960). It is necessary therefore to examine characteristics of the lactation experience in indi- vidual age-groups. Examination of individual pari- ties is also suggested by the observation, to be reported subsequently,1 that breast cancer risk is strongly associated with the age at which a woman has her first pregnancy. In the more detailed reports from the individual centres analyses of lactation characteristics for separate age- and parity-groups have therefore been shown. No significant or con- sistent differences between cases and controls were found within any classification by parity or age at diagnosis of the cancer. DISCUSSION The hypothesis that lactation reduces breast can- cer risk has been the subject of more polemics than studies. The support to be found in the literature is not as strong as the popularity of the hypothesis would suggest. In early studies, Lane-Claypon (1926) and Wainwright (1931) found both absence of lacta- tion and very prolonged lactation to be more fre- quent in breast cancer patients than in controls. In neither study was the selection of cases and controls satisfactory by present-day standards. MacDonald (1942) came to a most definite conclusion in support of a protective effect of lactation, but the comparison on which the conclusion was based is quite indefen- sible. More recently, two studies in the USA (Levin et al., 1964; Kaplan & Acheson, 1966), one in France (Schwarz, Denoix & Rouquette, 1958), two in Japan (Segi et al., 1957; Kamoi, 1960) and one involving several areas (Wynder, Bross & Hirayama, 1960) have indicated either less frequent or less prolonged lactation among breast cancer patients than among controls. However, the differences in all these studies have been quite small, and in none have the interrelationships of parity, socio-economic status and nursing been adequately separated. Two studies in the USA (MacMahon & Feinleib, 1960; Shapiro et al., 1968), and a small but very carefully controlled study in Israel (Abramson, 1966), have not revealed any appreciable differences in the lactation experience of breast cancer patients and unaffected women. It is noteworthy that only one of these past studies (Shapiro et al., 1968) is based on all cases occurring 1 MacMahon, B., Cole, P., Lin, T. M., Lowe, C. R., Mirra, A. P., Ravnihar, B., Salber, E. J., Valaoras, V. G. & Yuasa, S., Age at first birth and breast cancer risk (in preparation). in a defined population. The opportunities for the introduction of differences between patients and controls in socio-economic status, in addition to the evident tendency of breast cancer to affect higher economic groups disproportionately, are therefore manifold. The relatively small effect of adjustment for this variable in the present study does not exclude the existence of more important confounding effects in other studies using different methods of case or control selection. Furthermore, few past studies have adequately separated the effect of childbearing from that of breast feeding. Thus, the use of the measure " mean duration of lactation per child" does not adequately allow for parity differences between patients and controls since women of high parity tend to have a long lactation, quite independently of their greater oppor- tunity for lactation. For example, in the Boston control series in the present study, the mean duration of lactation per child breast-fed was 3.0 months for women of parity 1 and 6.4 months for women of parity 5 or more (Salber, Trichopoulos & Mac- Mahon, 1969). Adding to these difficulties the small numbers involved in some previous studies, it is not surprising that findings have been inconsistent. It is believed that the data reported in this paper compare favourably with those previously available with respect to biases introduced in the selection of the material. In addition, the large numbers of cases included, even in areas of low breast cancer risk, such as Japan, have allowed more comprehensive analyses to be made of the role of factors such as socio-economic status, parity and age. However, two methodological aspects of the present study deserve discussion. They are the use of other hospital patients as the source of the control groups and the reliance on women's memories regarding the duration of their lactation. Other hospital patients were used in this study because no other control group could conveniently be assembled in all centres. Many investigations for which a hospital control group might be unsuitable can be visualized. Intuitively, there seems little reason to suppose that the study of lactation may be among these, particularly if, in the analysis, allowance is made for the distributions of the case and control series by age, parity and socio-economic status. However, to assess the validity of this sup- position, a second control group was assembled in Boston from town census records. Findings with respect to lactation in this second control group drawn from the general population did not differ LACTATION AND CANCER OF THE BREAST significantly, or consistently, from either the breast cancer cases or the hospital control group (Salber, Trichopoulos & MacMahon, 1969). Regarding the reliability of lactation histories, it was possible, in some centres, to re-interview a sample of women 3-12 months after the original interview. The numbers of patients re-interviewed were: Boston controls 44, Glamorgan controls 49, Slovenia cases 48 and Taiwan cases and controls 29. There were a number of discrepancies in the infor- mation given at the two interviews. However, these were not of the magnitude, nor did they carry suf- ficient bias, to influence substantially the conclusions presented. For example, the two most critical mea- sures of our analyses are the proportion of women who never lactated, and the proportion who lactated for a relatively long period (as defined in Table 5). With respect to the former, among the 170 re-inter- viewed women, 51 were originally coded as having never lactated and 5 of these would not have been so coded on the basis of the re-interview; 3 women not originally assigned to this class would have been so coded on the basis of re-interview. Regarding the number who lactated for long periods, 34 of the 170 were originally coded to this class; 3 of these would not have been so coded on the re-interview, find there were 3 not so coded originally who would have been coded in this category on the re-interview. Of course, these data reflect on the reliability, rather than on the validity, of the lactation histories. How- ever, the strong relationships with socio-economic status noted in Table 6 do suggest that real differ- ences between women with respect to lactation experience were revealed by the interviews. In short, while both the use of a hospital control group and the reliance on patients' memories have no doubt introduced inaccuracies and inefficiences into our material, it is not thought that these can have been sufficient to lead to any appreciable excess or deficiency of lactation in the case series being overlooked. The failure to confirm the hypothesis of the pro- tective effect of lactation in these centres representing such a wide range of breast cancer risk, and includ- ing areas where appreciable numbers of women have very long lactation experiences, indicates that other hypotheses must be developed to account for the observed international differences in breast cancer rates. The finding also has relevance to at least one hypo- thesis concerning the mechanism underlying the relationship between ovarian function and breast carcinogenesis. It has been suggested that it is the cyclic nature of ovarian activity that is particularly significant in breast cancer etiology, and that circum- stances which interrupt cyclic ovarian activity reduce breast cancer risk (Wynder, Bross & Hirayama, 1960; Levin et al., 1964; Feinleib, 1968). In several of the centres in this study, the mean duration of lactation per child exceeded the usual duration of pregnancy (Table 5), and in such areas the inter- ruption of cyclic ovarian activity associated with lactation was presumably as long as that associated with pregnancy.- Since in these centres lactation was not, but pregnancy was, associated with de- creased breast cancer risk, the cyclic activity hypo- thesis seems an inadequate explanation of the mecha- nism of reduced risk associated with pregnancy. ACKNOWLEDGEMENTS This study would not have been possible without the to-operation of our medical colleagues in the several centres who gave access to their records and permission lo interview their patients. More detailed recognition than can be given here of our indebtedness to these individuals and to our staff who conducted the inter- views is given in the reports from the individual centres. Dr Harry Carter, New England Deaconess Hospital, undertook the central review of histological material. Mrs Hazel Coven, Mrs Joyce Berlin and Miss Linda Rosenstein have been responsible for the coding, data- processing and computer programming. For encouragement and counsel at all stages of the investigation, we are indebted to the World Health Organization. RESUME LACTATION ET CANCER DU SEIN: COMPTE RENDU SUCCINCT D'UNE ETUDE INTERNATIONALE On a procede parmi sept groupes de population 'a une de cette affection. Deux des groupes ont dtd choisis dans tenqu&e sur les ant6c6dents en matiere de lactation de des rdgions oiu le risque de cancer du sein est eleve malades atteintes de cancer du sein et de femmes indemnes (Boston, Etats-Unis d'Amerique, et Glamorgan, Pays 193 B. MACMAHON AND OTHERS de Galles), deux dans des regions oLu ce risque est faible (Taipeh, Taiwan, et Tokyo, Japon) et trois dans des regions oiu les taux d'incidence de I'affection sont de valeur intermediaire (Athenes, Grece; Slovenie, Yougo- slavie, et Sao Paulo, Bresil). On s'est efforce, dans chacun de ces echantillons de population, d'interroger toutes les femmes atteintes d'un cancer du sein diag- nostique pour la premiere fois au cours de l'enquete. Aux fins de comparaison, on a egalement recueilli, pour chaque cas de cancer du sein, les ant6c6dents de trois malades, d'age correspondant, hospitalis6es pour d'autres raisons. Dans cinq des centres d'etudc, les investigations ont pu etre menees avec une rigueur suffisante, les enqueteurs ayant reussi a s'entretenir avec plus de 80% des patientes atteintes de cancer du sein. Les taux d'incidence annuelle de la maladie atteignaient, par 100 000 femmes, 55,0 a Boston, 38,8 a Glamorgan, 28,9 ai Athenes, 24,4 en Slovenie et 9,1 at Taipeh. On estime que, dans les autres secteurs, 1'enquete a porte sur environ 50% (Tokyo) et 70% (Sao Paulo) des nouveaux cas. Au total, quelque 4400 malades atteintes de cancer du sein et pres de 13 000 femmes souffrant d'autres affections ont e interrogees sur leurs antecedents en matiere de lactation. Ainsi qu'il etait prevu, on a releve une plus forte proportion de femmes celibataires et de femmes mariees nullipares parmi les patientes atteintes de cancer du sein que parmi les malades temoins. Cependant, chez les meres d'un ou plusieurs enfants vivants, et compte tenu des variations relatives A I'age, au nombre d'accouche- ments et au niveau socio-economique, on ne notait aucune difference nette sous le rapport de la duree de la lactation - appreciee selon divers criteres - entre les femmes atteintes de cancer du sein et les autres malades. On a procede chez les femmes atteintes de cancer du sein a 28 comparaisons distinctes portant sur la duree de la lactation telle qu'elle etait observee et sur sa duree escomptee, evaluee d'apres les -chiffres releves chez les malades temoins presentant des carac- teristiques identiques en ce qui concerne l'age et le nombre d'accouchements. Dans 13 de ces comparaisons, la difference entre la valeur observee et la valeur escomp- tee plaidait en faveur d'un role protecteur de la lactation; dans 14 autres, la discordance donnait i penser que la lactation a un effet favorisant sur l'apparition du cancer du sein; une comparaison enfin montrait une concordance entre la valeur observ&e et la valeur escomptee. La conclusion tiree de la presente etude est que la lactation, meme prolongee pendant de longues periodes, n'a aucune influence sur le risque de cancer du sein auquel est exposee une femme. D'autres mecanismes sont en jeu qu'il faut elucider si l'on veut expliquer les variations considerables de l'incidence de l'affection dans differents pays. Les constatations faites au cours de cette enquete semblent par ailleurs infirmer la theori selon laquelle le risque de cancer du sein serait dependan de la duree des phases d'activite cyclique de l'ovaire. REFERENCES Abramson, J. H. (1966) Israel J. med. Sci., 2, 457-464 Doll, R., Payne, P. & Waterhouse, J. (1966) Cancer incidence in five continents, New York, Springer Feinleib, M. (1968) J. nat. Cancer Inst., 41, 315-329 Feinleib, M. & Garrison, R. J. (1969) Cancer, 24, 1109-1116 Kamoi, M. (1960) Tohoku J. exp. Med., 72, 59-65 Kaplan, S. D. & Acheson, R. M. (1966) J. chron. Dis., 19, 1221-1230 Lane-Claypon, J. E. (1926) A further report on cancer of the breast with special reference to its associated ante- cedent conditions. In: Report on Public Health and Medical Subjects No. 32, London, H.M. Stationery Office Levin, M. L., Sheehe, P. R., Graham, S. & Glidewell, 0. (1964) Amer. J. publ. Hlth, 54, 580-587 Lowe, C. R. & MacMahon, B. (1970) Lancet, 1, 153- 157 MacDonald, I. (1942) Surg. Gynec. Obstet., 74, 75-82 MacMahon, B. (1958) Cancer, 11, 250-254 MacMahon, B. & Feinleib, M. (1960) J. nat. Cancer Inst., 24, 733-753 Mantel, N. (1963) J. Amer. statist. Ass., 58, 690-700 Mantel, N. & Haenszel, W. (1959) J. nat. Cancer Inst., 22, 719-748 ( Salber, E. J., Trichopoulos, D. & MacMahon, B. (1969 J. nat. Cancer Inst., 43, 1013-1024 Schwartz, D., Denoix, P. F. & Rouquette, C. (1958) Bull. Ass. franc. Cancer, 45, 476-493 Segi, M. (1960) Cancer mortality for selected sites ij 24 countries (1950-57) Sendai, Department of Publi Health, Tohoku University School of Medicine Segi, M., Fukushima, I., Fujisaku, S., Kurihara, M. Saito, S., Asano, K. & Kamoi, M. (1957) Gann, 48 suppl., pp. 1-63 Shapiro, S., Strax, P., Venet, L. & Fink, R. (1968) Amer. J. publ. HIth, 58, 820-835 Valaoras, V. G., MacMahon, B., Trichopoulos, D. & Polychronopoulou, A. (1969) Int. J. Cancer, 4, 350-363 Wainwright, J. M. (1931) Amer. J. Cancer, 15, 2610-264 Wynder, E. L., Bross, I. J. & Hirayama, T. (1960 Cancer, 13, 559-601 Yuasa, S. & MacMahon, B. (1970) Bull. Wld Hlth Org. 42, 195-204