|
|
| |||||||||||||||||||||||||||||||||||||||||
Background The effect of pregnancy on the risk of breast cancer is not clear. We tested the hypothesis that the risk of breast cancer increases transiently after pregnancy but then falls to a level below that of age-matched nulliparous women.
Methods We conducted a case-control study of a nationwide cohort in Sweden, using a computerized record linkage between the Cancer Registry and the Fertility Registry. The study subjects were women born from 1925 through 1960 who were resident citizens of Sweden at the time of the 1960 census. A total of 12,666 patients with breast cancer were compared with 62,121 age-matched control subjects. We used conditional logistic regression to estimate odds ratios for the development of breast cancer at different ages, according to maternal age at first delivery (in uniparous as compared with nulliparous women) and age at second delivery (in biparous as compared with uniparous women).
Results Uniparous women were at higher risk of breast cancer than nulliparous women for up to 15 years after childbirth and at lower risk thereafter. The excess risk was most pronounced among women who were older at the time of their first delivery (odds ratio 5 years after delivery among women 35 years old at first delivery, 1.26; 95 percent confidence interval, 1.10 to 1.44). Women who had two pregnancies had a less striking increase in risk.
Conclusions Pregnancy has a dual effect on the risk of breast cancer: it transiently increases the risk after childbirth but reduces the risk in later years. In women with two pregnancies, the short-term adverse effect is masked by the long-term protection imparted by the first pregnancy. A plausible biologic interpretation is that pregnancy increases the short-term risk of breast cancer by stimulating the growth of cells that have undergone the early stages of malignant transformation but that it confers long-term protection by inducing the differentiation of normal mammary stem cells that have the potential for neoplastic change.
Pregnancy, therefore, may have opposing influences on the risk of breast cancer: a detrimental effect caused by enhancement of the growth of occult cancer cells and a protective effect mediated by the terminal differentiation of stem cells13,14. This view, which is supported by experimental data,15,16 can accommodate the fragmentary epidemiologic evidence. However, no study has explicitly examined the hypothesis of a dual effect of pregnancy in a data set large enough to generate high statistical power. The linkage of two nationwide Swedish registries gave us the opportunity to undertake such a study. The results demonstrate the dual effect of pregnancy and support the concept that pregnancy-associated hyperestrogenemia or another factor related to pregnancy transiently increases the risk of breast cancer17.
Methods
Population Registries
The data base for this study was created by linking the Swedish Cancer Registry and a nationwide Fertility Registry. In both registries female residents of Sweden can be identified by a unique registration number assigned at birth to every Swedish citizen (beginning in 1947) and assigned to immigrants at the time of their first residency.
The Swedish Cancer Registry includes cancer diagnoses beginning in 1958, as reported separately by clinicians and by pathologists or cytologists. Almost all diagnosed cancers are recorded, and in 95 percent of cases the registry receives two notifications18.
The Fertility Registry includes data on the reproductive history of approximately 1.7 million women born from 1925 through 1960 who were resident citizens of Sweden at the time of the 1960 census. Among this group of women, more than 2.8 million births were registered from 1943 through 1984. Information on the dates of birth of biologic and adopted children born from 1943 through 1960 was collected retrospectively at the time of the 1960 census (with adopted children representing less than 2 percent of the total)19. Beginning in 1961, only biologic children were included, with all new births added annually from records of vital statistics. For each woman, parity status (including nulliparity) was recorded, with the dates of all live births, as well as the date of the woman's emigration or death20. In the oldest cohorts, primarily including women born from 1925 through 1929, fertility was underestimated, since children born before 1943 and those who were not living with their mothers or who died before the 1960 census were not recorded.
Subjects
At time of their linkage for the purposes of the present study, the Cancer Registry and the Fertility Registry were updated through 1984. Among women born from 1925 through 1960 who were listed in the Cancer Registry, 16,243 breast cancers were recorded between 1958 and 1984, of which 14,273 had matching reproductive information in the Fertility Registry. Thus, no reproductive data could be retrieved for 1970 breast cancers (12.1 percent). Part of the discrepancy is accounted for by women given a diagnosis of cancer between 1958 and 1960 who died or emigrated before the 1960 census. Most of the discrepancy is due to the listing of non-Swedish citizens in the Cancer Registry but not in the Fertility Registry. An audit of the 1960 census data showed that in the oldest cohorts, the proportion of women for whom fertility information was missing was similar to the proportion of foreign nationals residing in Sweden21,22.
A total of 595 cancers listed in the registry involved a second diagnosis of breast cancer in the same woman (in the contralateral breast) and were excluded from the study. Of the remaining 13,678 women for whom concomitant fertility information was available, 895 women had been given a coding in the Cancer Registry indicating that their tumors had been found to be benign, and they were excluded from the study, as were 117 women who had births listed in the registry subsequent to a diagnosis of breast cancer. Hence, the final analysis included 12,666 women with a first diagnosis of breast cancer for whom reproductive information was available in the Fertility Registry. Seventy-nine percent of the patients were 50 years old or younger at the time of their diagnosis of breast cancer (mean, 44.4 years; range, 16 to 59).
To evaluate possible time-dependent associations between pregnancy and the risk of breast cancer, a case-control study nested within the large nationwide cohort defined by the Fertility Registry was undertaken. For every woman with breast cancer, five women were randomly selected as controls from the Fertility Registry. Matched controls were born in the same year and month as the corresponding case patient, were residents of Sweden when the case patient received her diagnosis, and had not had breast cancer before that date.
Statistical Analysis
Since the long-term protective effect of pregnancy occurs after the first birth, masking any short-term adverse effect of subsequent pregnancies, we focused our examination on a comparison of uniparous and nulliparous women. If age at first delivery is an important risk factor and if the effect of the first pregnancy varies with time since delivery (with a short-term increase in risk followed by a long-term reduction), we would expect different odds ratios for different categories of age at first delivery. According to this working hypothesis, age should be considered a potential interacting factor (effect modifier) when nulliparous women and uniparous women with a particular age at first delivery are compared with respect to the risk of breast cancer. The hypothesis of a dual effect would predict the odds ratio in the younger age groups to be greater than 1 (indicating a short-term increase in risk after pregnancy), whereas in the older age groups it would be less than 1 (indicating a long-term reduction in risk).
The hypothesis was also examined by comparing biparous women with uniparous women, after adjustment for age at first delivery. We expected the effect of age in this comparison to be less striking than in the analysis of uniparous and nulliparous women because of a measure of protection imparted by the first pregnancy.
Data on the case patients and control subjects were tabulated according to age and parity, with age at the time of delivery divided into five-year groups. Odds ratios were estimated on the basis of a conditional logistic-regression analysis that took into account the matched design (Table 1)23. In another conditional logistic-regression model, age and age at the time of delivery were treated as continuous variables. This model assumed a log-linear (i.e., a multiplicative) relation between the variables in the model and the odds ratio (an estimate of the rate ratio) of breast cancer. For the comparison between uniparous and nulliparous women, the model included age, parity (with scores of 1 for uniparous and 0 for nulliparous), age at first delivery, and the interaction term between age and age at first delivery (Table 2 and Figure 1). Table 3 shows the distribution of uniparous and biparous case patients and control subjects according to age at diagnosis and age at second delivery, with the odds ratios derived from the conditional logistic regression. Figure 2 compares the biparous and the uniparous women in a model that included age, age at first delivery, an indicator of parity (with scores of 1 for biparous and 0 for uniparous), age at second delivery, and the interaction term between age and age at second delivery.
|
|
|
|
|
We restricted the first set of analyses to the 4949 patients with breast cancer (39.1 percent) and the 21,253 control subjects (34.2 percent) who were nulliparous or uniparous. Among women 39 years of age or younger, 1209 cases of breast cancer were diagnosed. Table 1 shows the distribution of these women according to age and age at first delivery and the age-specific odds ratios derived from the conditional logistic regression. Odds ratios larger than 1 were found in younger women, whereas those for older women were mostly smaller than 1.
Table 2 shows that as compared with nulliparous women, uniparous women have elevated odds ratios of breast cancer soon after delivery but that the odds ratios decline later. The interaction term between age and age at first delivery was significant (P<0.001; 
coefficient, -6.7 x 10-4). Figure 1, in which years since delivery (age minus age at first delivery) is the time axis, shows this pattern. The increased risk immediately after delivery is most pronounced in women who were 30 years old or older at the time of their first delivery. For women of all ages at their first delivery, however, the odds ratios are lower than those for nulliparous women after approximately 15 years.
The number of births may have been underreported in a few women, mainly those born before 1930 (see the Methods section). Because such misclassification would affect the case patients and the control subjects equally, the true association may be underestimated. We therefore repeated the analysis, excluding the women born from 1925 through 1929. If anything, the short-term increase and the long-term decrease in relative risk became marginally stronger, as predicted (data not shown).
Biparous women, grouped according to age at the time of their second delivery, are compared with uniparous women in Table 3 and Figure 2. With the exception of the women who were 35 years old or older at the time of the second delivery, the biparous women had lower odds ratios than the uniparous women for all periods after second delivery, and the odds ratios tended to decline further with time.
The interaction term between age and age at second delivery was not significant (P = 0.39; coefficient, -1.3 x 10-4).
Discussion
Our results indicate a dual effect of pregnancy on the risk of breast cancer that was more evident among uniparous as compared with nulliparous women than among biparous as compared with uniparous women. These findings confirm predictions about the pathogenesis of breast cancer that are derived from theoretical models as well as from fragmented but converging epidemiologic evidence5,9,10.
The population we studied was a large, dynamic cohort and in theory could be analyzed as such. However, a nested case-control sampling design incorporates all the superior attributes of a cohort design with respect to validity, with substantial advantages in computational convenience24. A special but resolvable problem in the present context arises because the variable for exposure, time since first delivery, equals the difference between age at diagnosis (or control ascertainment) and age at first delivery -- two factors that are controlled for in the analysis. The standard solution to this problem is to consider current age (age at diagnosis) or time since first delivery as a factor that interacts with age at first delivery in the modulation of the odds ratio (the relative risk of breast cancer)25. This approach results only in a loss of statistical power, an acceptable compromise in a study of this size.
With data from large nationwide registries, our analysis achieved high statistical power and virtually eliminated selection and information bias. Possible sources of residual bias include the misclassification of parity. Such an error is nondifferential and probably negligible, because it applies to a small number of subjects in the oldest birth cohorts. We could not examine and control for confounding by age at menarche, age at menopause, and somatometric factors. Abundant external information about the magnitude of collective confounding by these factors suggests, however, that this problem would be negligible26. By contrast, there are substantial confounding influences among age, parity, and age at first delivery -- factors that were fully accounted for in this study.
In a mathematical model, Pike et al. related the hormonal changes during a woman's life to the aging of breast tissue, and thus to the risk of breast cancer27. Rosner et al. recently applied this model to data from the Nurses' Health Study and extended it to accommodate multiple births. They found a significant short-term increase in the risk of breast cancer immediately after the first delivery28. The results of our study are also compatible with data showing that pregnancy brings about both transient and permanent structural changes in the breast tissue of laboratory animals15,16. Thus, the short-term increase in risk that follows a first birth may reflect a growth-enhancing effect of high estrogen levels during pregnancy on tumor cells whose malignant transformation has already begun. Pregnancy also exerts a long-term protective effect, presumably by causing stem cells to differentiate and become resistant or less sensitive to carcinogenic stimuli13,16. We found a larger short-term increase in risk among women who were older at first delivery. We would expect this result if susceptible breast parenchyma is exposed to carcinogenic stimuli for a relatively long time before the pregnancy.
The differentiation of mammary cells induced by the first full-term pregnancy may have masked a short-term adverse effect of subsequent pregnancies. It is also possible that most of the growth-enhancing effects on cells that have undergone the early steps of malignant transformation actually take place during the first pregnancy. The absence of a strong transient increase in risk after the second birth may also reflect, to a certain extent, the fact that hormonal changes are more marked in primigravidas; higher levels of estrogens have been found in first than in second pregnancies29,30.
Our study explains why nulliparous women have a lower risk of breast cancer than parous women during the childbearing years31 or before the age of 35 or 405; why the average age at the diagnosis of breast cancer is lower among parous than among nulliparous women,32 even though the cumulative incidence of the disease is higher in the second group; why the age at diagnosis interacts with parity to modulate the risk of breast cancer, with crossover points between the ages of 45 and 60 years, as shown by studies that did not directly examine the effect of age since delivery3,6,7,11; and why studies of the short-term increase in the risk of breast cancer after the most recent pregnancy that combined women with two or more pregnancies9,10 or all parous women11 into one category found a shorter-lasting increase in risk9,10 or no increase at all,11 as compared with the increase lasting 15 years found in the present study (in those studies, the short-term increase was masked by the refractory state imparted by the earlier pregnancies). We have confirmed that a woman's age at the time of a pregnancy subsequent to the first pregnancy also affects the risk of breast cancer,4 although considerably less so than her age at the time of the first pregnancy (Figure 1 and Figure 2), with a crossover point in risk for women 30 to 35 years of age during pregnancy. Furthermore, our study explains why older age at the last delivery,8,12 particularly among multiparous women,12 would affect the risk of breast cancer, even after adjustment for age at first delivery. A late last pregnancy is likely to reflect the late occurrence of the previous pregnancies, all of them having cumulative effects on the risk. Our study could not fully assess the influence of pregnancy on the long-term risk of breast cancer, because the cohort comprised mostly younger women; such women, however, are best suited for the study of short-term effects.
Supported by grants from the Swedish Cancer Society, the Swedish Society of Medicine, the Lions Cancerforskningsfond, and the Stavros S. Niarchos Fund of the Harvard School of Public Health. Dr. Hsieh was the recipient of a Suzanne Sheats Breast Cancer Research Grant from the Massachusetts Department of Public Health.
We are indebted to Mr. Jan Qvist and Mr. Lars Nordin at Statistics Sweden for their valuable advice and assistance in preparing the original data.
Source Information
From the Department of Social Medicine (M.L.) and the Cancer Epidemiology Unit (M.L., A.E., H.-O.A.), University Hospital, Uppsala, Sweden; the Department of Epidemiology, Harvard School of Public Health, Boston (C.H., D.T., A.E., M.P., H.-O.A.); and the Cattedra di Igiene, Facolta di Medicina di Catanzaro, Universita di Reggio Calabria, Catanzaro, Italy (M.P.).
Address reprint requests to Dr. Lambe at the Department of Social Medicine, University Hospital, S-751 85 Uppsala, Sweden.
References
| |||||||||||||||||||||||||||||||||||||||||
Related Letters:
The Risk of Breast Cancer after Giving Birth
Thompson W. D., Lambe M., Hsieh C.-c.
Extract |
Full Text
N Engl J Med 1995;
332:63-64, Jan 5, 1995.
Correspondence
Survivors of Breast Cancer
Potter J., Ship A. N., Love R. R., Burstein H. J., Winer E. P.
Extract |
Full Text |
PDF
N Engl J Med 2001;
344:309-310, Jan 25, 2001.
Correspondence
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2008 Massachusetts Medical Society. All rights reserved. |
Previous
