Does birth history account for educational differences in breast cancer mortality? A comparison of premenopausal and postmenopausal women in Belgium
This study investigates the impact of reproductive factors on the association between education and breast cancer mortality in Belgium. The role of reproductive factors has been investigated in several studies, with mixed results. Reproductive factors are either completely or partially responsible for the association between education and breast cancer mortality. The data consist of the 1991 census linked to registration data on cause-specific mortality during the period 1991–1995, including all breast cancer deaths in Belgium during the observation period. The study population includes all women aged 35–79 at time of the census. Age-standardized mortality rates and mortality rate ratios (Poisson regression) are computed for educational groups with and without control for reproductive factors. The population is stratified according to age (women aged 35–49 and 50–79) and according to nulliparity. The relationship between education and breast cancer is significant among postmenopausal women. Breast cancer mortality is higher among the higher educated women. These results are consistent with international findings, the gradient not being negative as in most other causes of death, but positive. Statistical control for parity and age at first birth reduces the association largely. In addition, among nonparous women, differences in breast cancer mortality by education are not consistent and generally not significant. Reproductive factors are largely responsible for the positive association between education and breast cancer mortality among postmenopausal women in Belgium. Among premenopausal women, the relation is not significant, a pattern consistent with international studies.
Breast cancer is the most common cause of premature death among women in industrialized countries. In 2006, Belgium has among the highest breast cancer incidence and breast cancer mortality in Europe. Thus, Belgium represents a challenging setting to investigate differences in breast cancer mortality. Nulliparity or postponement of childbirth—factors known to be positively associated with breast cancer disease—are not particularly high and yet incidence and mortality rates are remarkably high for this cause of death. In 2006, the mean age at childbirth is equal to 29.4 in Belgium compared to 30.1 in Europe. Similarly, childlessness is not particularly pronounced in Belgium.
Differences in breast cancer are of particular interest. Unlike other causes of death, breast cancer incidence and breast cancer mortality show a positive association with socio-economic status, especially educational level.1–5 The picture is becoming more complex, however. Recent studies have shown that the gradient in breast cancer mortality is moving from being positive over nonexistent to negative.6–8
There are several pathways that link education to breast cancer. With respect to breast cancer incidence, education enhances physical activity and dietary and health behavior patterns, factors known to reduce breast cancer incidence. On the other hand, there are also factors linked to education that increase breast cancer incidence. Education increases nulliparity and leads to postponement of parenthood. Both these reproductive factors—nulliparity and late age at first birth—have been held responsible for the positive educational gradient in breast cancer occurrence. Evidence is not entirely consistent on this point. Some studies show that the gradient can be explained entirely by age at first birth and nulliparity, other analyses report a residual excess incidence9 or mortality,10 allowing for the conclusion that other factors too are responsible for the positive association between education and breast cancer. With respect to breast cancer survival, education generally improves survival chances by its impact on stage of breast cancer at diagnosis (participation in screening programs), type of cancer treatment, psychosocial support, postmenopausal hormone therapy and estrogen—receptor status.11
This study focuses on the joint outcome of breast cancer incidence and breast cancer survival, i.e., breast cancer mortality. The analyses will investigate whether educational differences in breast cancer mortality can be explained by reproductive factors. Our dataset is exhaustive and includes all cases of breast cancer mortality during the observation period in Belgium. To our knowledge, only Nordic countries dispose of individual linked data on breast cancer mortality, education and reproductive factors on a nationwide scale. The data for this analysis are thus quite exceptional outside the Nordic context. Investigation into the relationship between breast cancer mortality and education may provide more information on the etiology of the disease and on the reasons why breast cancer increases with education. It should be stressed that the analyses deal with breast cancer mortality. The data do not allow for analyzing the educational impact on breast cancer incidence and breast cancer survival separately. If education is associated positively with both incidence and survival, differences in breast cancer mortality only reflect part of the differences in breast cancer incidence between lower and higher educated women.
Material and Methods
The data consist of a linkage between the 1991 census and register information on survival status, emigration status and cause of death for the period March 1, 1991 to December 31, 1995. The dataset contains information on all women (legally) residing in the country on the March 1, 1991. Each woman is classified by attained educational level and by fertility history (number of children and date of birth of offspring). For the period under investigation, breast cancer diagnosis was defined by ICD-9-code 174. The research population comprises all women aged 35–79 at the time of the census. In total, 2,247,699 women are included in the analyses (excluding women with missing values), 8,224 breast cancer deaths and 10,615,471 person-years lived.
Education is categorized according to the International Standard Classification of Education (ISCED): preprimary and primary education (ISCED 0–1); lower secondary education (ISCED 2); upper secondary education (ISCED 3) and tertiary education (ISCED 4–6). Educational differences in breast cancer mortality are controlled for two reproductive factors, number of children and age at first birth. Each variable is categorized into six groups (no children, 1 child, 2 children, 3 children, 4 children and 5 or more children and <20 years, 20–24 years, 25–29 years, 30–34 years, 35–39 years and >40 years). The effect of both reproductive variables is estimated through a combined indicator, allowing for the inclusion of all women whether they have children or not.
Mortality rates in breast cancer are calculated for each educational group as a function of age, number of children and age at first birth. The association between these variables is measured in terms of mortality rate ratios estimated using a Poisson model. All models are controlled for age at the 1991 census, introduced as a continuous variable (1-year age groups). The association between breast cancer and education is controlled for number of children and for age at first birth.
As risk factors of breast cancer differ by menopausal status,12, 13 analyses are stratified by age in order to distinguish between premenopausal and postmenopausal women (women aged 35–49 and women aged 50–79). Age at time of census is used as a proxy variable for menopausal status because there is no information on the menopausal status of the women in the dataset. Taking into account, the time lag between breast cancer incidence and breast cancer mortality, some breast cancer deaths observed for postmenopausal women originated in the premenopausal stage of life. Conclusions concerning breast cancer incidence strictly related to menopausal status are therefore not drawn.
In a second series of models, analyses are presented for parous and nonparous women separately. This allows for estimating the separate impact of both reproductive factors—number of children and age at first birth—on breast cancer mortality among parous women. Among nulliparous women, educational differences in breast cancer are controlled for age at census only.
In our dataset, more than half of the women have had no education or primary education only (labeled from now on as “lower education”). About 30% has a diploma of secondary education (16% lower and 14% upper secondary education) and somewhat more than 10% has a certificate of tertiary education (“higher” education). The remaining 9% procured no information for educational level. This group counts proportionally more women with a foreign nationality, more residents of the Brussels-Capital Region and more single women than women for whom educational level is known. Their age structure is comparable, their mortality somewhat higher, but not for breast cancer.
Fertility clearly differentiates according to education. Women in our dataset with lower education have on average 2.27 children compared to 1.81 among women with higher education. Women with secondary education have intermediate levels: 1.91 for lower secondary and 1.83 for upper secondary. The largest differentiation is between women with lower education and other women. In the timing of births, a more linear pattern can be observed. Lower-educated women have their first child at 24.1 years, lower secondary educated women at 24.5 years, upper secondary educated women at 25.1 years and higher educated women at 26.6 years.
Table 1 gives the mortality rates for breast cancer according to education and age (the highest rates within each row being highlighted). The crude rates decrease with educational level (first row in Table 1). This is due to the age composition: women with lower education being on average 10.5 years older than women with higher education (mean age of 58.4 compared to 48.0). Age-specific breast cancer mortality rates are obviously lower among women with lower education (except for women aged 35–39 at baseline) and higher among women with higher education. To investigate the role of reproductive factors in generating this pattern, Poisson models are used regressing age, education, parity and age at first birth on breast cancer mortality.
Table 1. Mortality rates for breast cancer by education and age, women aged 35–79, Belgium
Table 2 presents the rate ratios (R.R.) resulting from the Poisson regressions. The association between education and breast cancer is not significant among premenopausal women but significant among postmenopausal women. Women aged 50–79 with higher education have a 16% higher mortality risk due to breast cancer, women with upper secondary have a 14% higher mortality risk compared to women with lower education. Controlling for parity and especially for parity in combination with age at first birth explains much of the excess mortality observed among tertiary educated women. Among women with a diploma of upper secondary education, the excess mortality remains significant.
Table 2. Breast cancer mortality rate ratios by education, controlling for age, parity and parity combined with age at first birth, 95% confidence intervals (significant values in italics)
The association between breast cancer mortality and parity is significant for both premenopausal and postmenopausal women. The mortality risk is systematically lower for women with children (although not always statistically significant). Among postmenopausal women, the rate ratio decreases gradually with the number of children. After introduction of age at first birth, the association remains consistent within each category of age at first birth for postmenopausal women giving birth to their first child before the age of 30.
The gradient according to age at first birth is consistent across all parity categories for postmenopausal women who had their first child before the age of 30. Among postmenopausal women, the risk of dying from breast cancer declines systematically with a younger age at first birth, independently of the number of children. The lowest risk is observed among women who have 4 or more children and who had their first child before the age of 20. The highest risk is observed among women with one child born when they were 35 or older. They have a mortality risk higher than nulliparous women (although not significant).
To investigate the impact of number of children and age at first birth separately, analyses have been limited to parous women. The analyses in Table 3 confirm the above conclusion that there is no significant association between education and breast cancer mortality among premenopausal women. Among postmenopausal women, differences are smaller when nulliparous women are excluded from the analysis. All higher educated postmenopausal women have a higher rate ratio of breast cancer mortality, although only statistically significant for the upper secondary educated. Among these women with upper secondary education, the rate ratio remains significant after control for number of children and for age at first birth. In other words, among women who have children, highly educated women do not have a higher risk of dying from breast cancer, whereas women with upper secondary education have an excess mortality of 10%. There is a significant association between breast cancer mortality and number of children: the relative risk of dying from breast cancer diminishes almost linearly as the number of children increases among postmenopausal parous women. Age at first birth clearly leads to considerable differences in breast cancer mortality: women who start procreating at age 40 and above have a mortality risk that is almost twice as high compared to women who start to procreate before the age of 20. Among premenopausal parous women, the risk amounts to three times the risk of women starting their fertility career before the age of 20. Clearly, postponement of motherhood goes together with higher breast cancer mortality.
Table 3. Breast cancer mortality rate ratios among parous women by education, controlling for age, parity and parity combined with age at first birth, 95% confidence intervals (significant values in italics)
Among nulliparous women, differences in breast cancer mortality by education are generally not significant (Table 4). In both premenopausal and postmenopausal women, the results do not show a clear pattern or gradient, but among postmenopausal women those with a lower secondary education do have a significantly lower risk to die from breast cancer.
Table 4. Breast cancer mortality rate ratios among nonparous women by education, controlling for age, 95% confidence intervals (significant values in italics)
In line with previous findings, highly educated women have higher breast cancer mortality rates than women with lower educational levels.1, 3–7 The relationship is clearly confirmed for postmenopausal women, but less obvious for premenopausal women. The analyses show that reproductive factors—number of children and age at first birth—are largely responsible for the excess mortality of highly educated postmenopausal women. Separate analyses for parous and nulliparous women demonstrate that educational differences are generally not significant among nulliparous women. In addition, the comparison of Tables 2 and 3 allows for the conclusion that nulliparity plays an important role in the observed excess mortality among higher educated women, whereas the impact of nulliparity is almost inexistent among lower and upper secondary educated women.
Controlling for reproductive factors reduces the educational differences almost entirely. Given the large number of unobserved factors in our analyses, the residual variation is remarkably small. Upper secondary educated women still have an excess mortality compared to the other educational levels, hinting toward the possibility of other intervening factors such as differences in smoking behavior and other life style factors.
A consistent relationship appears between breast cancer mortality and number of children. Mortality declines with the number of children among both premenopausal and postmenopausal women. This finding confirms the important role of childbearing in preventing breast cancer.
Simultaneous control for the number of children and age at first birth generally reinforces this pattern among postmenopausal women. For women aged 30 or older at first childbirth, differences by parity are neither consistent nor significant. Apparently, the number of children no longer has a protective effect for women who had their first child at an older age. As shown in other studies, age at first birth generates larger mortality differences than the number of births.10 Postponement of childbearing clearly takes high toll in terms of breast cancer mortality. Premenopausal parous women who had their first child at age 30 or older have a three times higher risk of dying from breast cancer than those who had their first child at age 20 or before; postmenopausal women have a two times higher risk.
To our knowledge, few studies on breast cancer mortality have differentiated between parous and nonparous women. Our results suggest that there is no significant educational gradient among nonparous women. These results contrast with a Norwegian study, showing that nulliparous women have larger educational difference than parous women and that other confounders—other than reproductive factors—are important to explain the larger educational differences among nulliparous women, such as life style factors and health behavior.10
When concentrating solely on parous women, educational differences in breast cancer mortality are smaller. This underlines the role of reproductive factors and suggests that childlessness plays a significant role in generating educational differences.
This study has some weaknesses that have implications for the interpretation of educational differences in premenopausal and postmenopausal breast cancer mortality. The main point resides in the impossibility to take into account differences in incidence, prevention and treatment that may be responsible for less obvious patterns in premenopausal breast cancer mortality. Another important point is the fact that reproductive data are based on self-reported live born children. Stillbirths and prematurely interrupted pregnancies are not reported, although they may be as relevant by influencing hormonal levels. Different prevalence of stillbirths according to educational level cannot be excluded and is probably more important in the cohorts observed in the premenopausal group. Finally, the research design used implies that menopausal status is confounded with birth cohort effect and period effect.
The results show that premenopausal breast cancer mortality is more strongly related to reproductive factors than postmenopausal breast cancer mortality, but shows no association with education. This pattern is consistent with international studies that indicate that a high level of education is more strongly associated with breast cancers diagnosed at age 50 or older, compared to age 49 or less.11, 14 In these studies, breast cancer incidence is associated with education among premenopausal women, but to a smaller extent than with postmenopausal women. To explain this pattern of smaller educational inequalities with premenopausal women, several factors have been put forward in the literature.
Premenopausal breast cancer seems more strongly associated with endogenous factors—age at menarche12 and family history13—that are either not or much less related to educational level than the risk factors that have been identified in postmenopausal breast cancer. Breast cancer is hormone related and there is evidence that risk factors differ for premenopausal and postmenopausal cancers.15 Breast tissue, as well as hormones and hormone-receptor status, vary at different stages of life and risk factors may have a differential impact according to the stage of life.16–20
To explain the smaller differences among premenopausal women, literature also focused on the changing distribution of reproductive behavior according to educational level, differences having decreased among the younger cohorts.10, 14 The premenopausal women in this study consist of the cohorts that are largely responsible for the baby bust, characterized by a generalized decline of fertility and smaller educational differences in realized parity (birth cohort effect). In the age group 50–79, tertiary educated women have—in our study population—on average 0.40 fewer children compared to lower educated women (1.77 vs. 2.17). Among the younger women, differences are smaller (1.80 compared to 2.02 children). Differences in age at first birth seem to have increased instead of decreased in our study population. The rejuvenation of fertility is obvious when comparing postmenopausal and premenopausal women in our data. Among postmenopausal women, the lower educated had their first child on average at 24.6 years, tertiary educated at 27.3 years, giving a difference of 2.7 years. Among premenopausal younger women, the difference amounts to 3.7 (22.6 compared to 26.3 years). Differences in age at first birth by educational level are larger among premenopausal women and cannot explain the smaller differences in breast cancer mortality among younger women.
The fact that we do not find educational differences in breast cancer mortality among premenopausal women does not necessarily imply that there are no differences in breast cancer disease. Incidence is positively related to educational levels,2, 4, 9, 11 whereas studies have shown that survival from breast cancer is generally higher among the higher educated11, 21–23 reducing or even diluting the consequences of the lower incidence of breast cancer among women with lower education.
Furthermore, period-related differences in breast cancer risk factors as well as in breast cancer detection may exist, explaining the difference between premenopausal and postmenopausal women.11 This effect could explain our research results partly. If self-inspection, screening and prevention are more prevalent among higher educated women, this should not only result in a lower mortality, but also in a more pronounced time lag between incidence and mortality. Because of this differential time lag between diagnosis and mortality by level of education, part of the differentiation could disappear among premenopausal women.
It is possible that the positive association between breast cancer mortality and education observed for postmenopausal women will disappear and maybe turn into a negative association. Several countries have witnessed such an inversion of the gradient.6, 7 This may be explained through the inverse equity theory: inequality will increase if individuals within higher socio-economic groups access new technologies such as improved breast cancer treatments, prior to those within lower groups. Comparing postmenopausal and premenopausal women shows that there might be a trend toward declining or inversing inequalities. Conclusions cannot be drawn on this point however, as the disease of breast cancer during premenopausal and postmenopausal stage of life is quite different. Longitudinal data are necessary to study whether inequalities in breast cancer are moving from being positive to negative.