Perinatal Factors and Mortality from Breast Cancer

Inverse associations have been reported between birthweight and subsequent mortality from circulatory disease and diabetes among women. In the current study, we assessed whether perinatal factors were associated with mortality from breast cancer. This follow-up study consists of breast cancer cases who participated in two population-based case-control studies of breast cancer in women under age 45 years conducted between 1983 and 1992 in three western Washington counties. This analysis is restricted to the 1,024 cases or their proxies who completed a supplementary questionnaire on perinatal factors from 1994 to 1996. The mean and median length of follow-up among living cohort members were 153 and 148 months, respectively. Relative to women who were firstborn, women who were born second or higher in the birth order seemed to have lower mortality from breast cancer [hazard ratio (HR), 0.2; 95% confidence interval (95% CI), 0.2-0.3]. In contrast, maternal age of ≥35 years (HR, 1.7; 95% CI, 1.1-2.8) was associated with higher breast cancer mortality relative to a maternal age of <25 years. Birth order modified the effect of maternal age on mortality from breast cancer (P = 0.03). There was evidence of increased breast cancer mortality for birthweight of ≥4,000 g (HR, 1.8; 95% CI, 1.0-3.1) and twin membership (HR, 2.5; 95% CI, 1.0-6.2). The protective effect of being born second or higher in the birth order against breast cancer mortality regardless of maternal age is striking and needs to be confirmed in future studies. (Cancer Epidemiol Biomarkers Prev 2006;15(10):1984–7)

Reported associations between perinatal factors and breast cancer incidence have been mixed. In a review article, Potischman and Troisi (1) identified a strong elevation in breast cancer risk for being a twin and a moderate increase in premenopausal breast cancer risk for high birthweight. Results were inconsistent for breast cancer associated with being firstborn, older maternal age, and gestational age, whereas there was no relation for maternal smoking. Our previous studies conducted among the same group of younger women as in the present study (2, 3) and studies conducted since the review article have been in general agreement. Trichopoulos (4) hypothesized that exposure to high pregnancy estrogen levels in utero could lead to subsequent breast cancer, whereas exposure to low pregnancy estrogen levels could protect against subsequent breast cancer. In studies utilizing cord blood, Troisi et al. (5) and Shibata et al. (6) found no association between estrogen levels and high birthweight or being firstborn. Troisi et al. (5) did not report any association between estrogen levels and older maternal age; however, Shibata et al. (6) reported a positive association between estrogen levels and older maternal age. Alternative mechanisms that may explain associations between high birthweight and breast cancer incidence include higher intrauterine exposure to insulin (7-9), insulin-like growth factor I (7-9), or leptin (9) based on cord blood levels. To our knowledge, there have been no studies of associations between cord blood levels of insulin, insulin-like growth factor I, or leptin and other perinatal factors.

Hypothesized pathways for breast cancer etiology could also influence the risk of dying although little research has been conducted to date. Goodwin et al. (10) found that elevated insulin levels, but not estrogen or insulin-like growth factor I levels, were associated with breast cancer mortality among women diagnosed pre- and postmenopausally independent of obesity. However, Borugian et al. (11) reported increased mortality from breast cancer associated with elevated insulin levels among women diagnosed postmenopausally only. To date, there have been no studies of leptin and breast cancer mortality. Only two studies have assessed the association between a perinatal factor and breast cancer mortality. In the Hertfordshite Cohort Study, Syddall et al. (12) failed to find an association between high birthweight and breast cancer mortality. In the American Cancer Society Cancer Prevention Study 1, Holmberg et al. (13) reported a nonsignificant elevation in breast cancer mortality associated with older maternal age. The present study was conducted to investigate the association between perinatal factors that may reflect estrogen, insulin, insulin-like growth factor I, and leptin levels and mortality from breast cancer.

Breast cancer cases (or proxies for deceased cases) from two previous population-based case-control studies of breast cancer, among women under age 45 years diagnosed between January 1983 and December 1992 in three western Washington counties, were recontacted and asked to provide information pertaining to their birth. Detailed methods of the two studies appear elsewhere (14, 15). Briefly, women were eligible for the first study if they were diagnosed with primary invasive breast cancer between January 1983 and April 1990, were born after 1944, and resided in King, Pierce, or Snohomish counties of Washington State at the time of diagnosis. Using the same methods as the first study, eligible study participants in the second study consisted of women who were diagnosed with primary invasive breast cancer from May 1990 to December 1992, were under age 45 years, and resided in the three county area. A total of 83% of eligible breast cancer cases ascertained through a population-based cancer registry completed a standardized personal interview in the initial studies. After obtaining ethical approval for the study of human subjects, all invasive breast cancer cases and proxies for deceased cases were targeted for the follow-up study. Between May 1994 and December 1996, a total of 1,024 (82.3%) mailed questionnaires or telephone interviews were completed for 1,244 eligible cohort members. Response rates were comparable for the 852 living cases (82.4%) and the 172 proxies for deceased cases (81.9%).

The study was approved by the institutional review board of the Fred Hutchinson Cancer Research Center. Although information on perinatal factors was also collected from mothers of cases, the small number of cases (n = 510) with maternal information would have decreased the precision of estimates; therefore, this analysis is based on self-report of cases or their proxies.

After obtaining informed consent, women were asked about their birthweight in pounds and ounces or, if they could not recall their exact birthweight, whether they were <5.5 or ≥9 pounds. Pounds and ounces were converted to grams; <5.5 pounds was classified as <2,500 g and ≥9 pounds was classified as ≥4,000 g. Maternal age was in exact years. Birth order was the combination of live and still births before the subject's birth. To classify gestational age, women were asked whether they were born >4 weeks before they were due, around the time they were due, or >4 weeks after they were due. Twin birth, maternal smoking, and maternal hormone use were yes or no questions. Maternal hormone use could have been use of diethylstilbestrol or any other hormonal formulation, such as oral contraceptives, during pregnancy with the subject. We conducted a validity study among women born in Washington state and found very high correlations comparing self-report with birth certificate for maternal age (r = 0.95) and comparing self-report with mother report for birth order (r = 0.89) and for birthweight (r = 0.85; ref. 16).

Detailed methods of the follow-up for mortality appear elsewhere (17). Briefly, the primary source of information on deaths was the cancer registry, which attempts to update disease status and vital status on an annual basis. Information on death was collected for women who currently did and did not reside inside the cancer registry catchment area. Secondary sources of information on deaths were death certificates, National Death Index, Health Care Financing Administration tapes, and relatives of patients. Death certificates were abstracted to ascertain cause of death. After comparing the 92% of women whose death was breast cancer related with all-cause mortality among the entire cohort of women, we found similar results; thus, we report results for all-cause mortality. Subjects were followed until the earliest of the date of death, date last known to be alive, or end date of the follow-up period, which was June 2002. Of women not reported to be dead in June 2002, 93% of women had been located within the previous year and 96% had been located within the previous 3 years. The mean and median length of follow-up for living cohort members were 153 and 148 months, respectively.

Cox proportional hazards was used to estimate the relative risk of dying from breast cancer and its 95% confidence interval (95% CI) associated with perinatal factors. This hazard ratio (HR) and left truncation were used to adjust for the median lag of 7 months between diagnosis and interview. To account for the left truncation of survival times, women were considered to be at risk of death from the time they were interviewed rather than from the time of diagnosis. Observations were censored on either the date of last known follow-up or the end date of the follow-up period. Based on a 10% change between crude and adjusted HRs, stage at diagnosis and birth order confounded the association between perinatal factors and mortality from breast cancer. For covariates that were missing a substantial percentage of data, we created a missing category for multivariate analyses. There was no evidence of confounding by family history of breast cancer, race, smoking, alcohol use, parity, recency of last birth before diagnosis, oral contraceptive use, exercise, education, income, or lactation. Nor was there evidence of confounding by variables that may be in the causal pathway between perinatal factors and breast cancer mortality, including age at menarche, body mass index, mammogram history, treatment history, birthweight, gestational age, twin birth, maternal smoking, or maternal hormone use. All analyses were initially adjusted for age of subject (continuous), diagnosis year (exact year), and stage at diagnosis (I, IIA, IIB, III+), and further adjusted for birth order (first, second, third+) or maternal age (<25, 25-29, 30-34, 35+ years). We stratified by stage at diagnosis, birth order, and maternal age, which are in the causal pathway between perinatal factors and breast cancer mortality, to determine the effect of perinatal factors on breast cancer mortality beyond their effect on intermediate variables. We assessed linear trend by treating categorical variables as continuous variables. Because birth order may modify the effect of maternal age on breast cancer mortality, we added an interaction term between maternal age (<30 and 30+ years) and birth order (first and second+) to the proportional hazards model with the main effects and did the likelihood ratio test to examine whether there was evidence of effect modification. To determine whether the associations between perinatal factors and breast cancer mortality were mediated by treatment, we stratified by use of any adjuvant therapy, chemotherapy, radiation therapy, and hormone therapy.

Table 1 presents the HRs and 95% CIs for breast cancer mortality associated with perinatal factors. After adjustment for age at diagnosis, diagnosis year, and stage at diagnosis, women who were ≥4,000 g at birth had somewhat higher mortality from breast cancer relative to women whose birthweights were 2,500 to 3,999 g, which was more pronounced after further adjustment for birth order (HR, 1.8; 95% CI, 1.0-3.1). Before further adjustment for birth order, women whose mothers were of ages ≥35 years at their birth had somewhat lower mortality from breast cancer compared with women whose mothers were of ages <25 years, which was reversed after further adjustment for birth order (HR, 1.7; 95% CI, 1.1-2.8). In addition, there was a trend of increasing mortality with increasing maternal age (P = 0.01) evident after further adjustment for birth order. Women who were born second or higher in the birth order had substantially lower mortality from breast cancer (HR, 0.2; 95% CI, 0.2-0.3) and there was a trend of decreasing mortality with increasing birth order (P < 0.01). Although based on very few cases, further adjustment for birth order resulted in higher mortality from breast cancer for being a twin (HR, 2.5; 95% CI, 1.0-6.2), which was not evident before adjustment for birth order. There was a slight reduction in mortality from breast cancer among women whose mothers used hormones during their pregnancy (HR, 0.6; 95% CI, 0.4-1.0).

Table 2 shows the joint effect of maternal age and birth order on mortality from breast cancer. The reference group is firstborn women whose mothers were of ages <30 years at their birth. Birth order modified the effect of maternal age on mortality from breast cancer (P = 0.03). The substantially reduced risk of mortality from breast cancer associated with higher birth order persisted regardless of maternal age. However, the increased risk of death from breast cancer among women whose mothers were older was seen among firstborn women only (HR, 1.6; 95% CI, 1.1-2.2). For all perinatal factors, our findings were similar among women who did and did not receive any adjuvant therapy, chemotherapy, radiation therapy, or hormone therapy (data not shown).

Although the Hertfordshire Cohort Study failed to find an association between birthweight and breast cancer mortality (12), we found a borderline increase in breast cancer mortality among women who were ≥4,000 g at birth relative to women whose birthweights were 2,500 to 3,999 g. We saw higher mortality from breast cancer among women whose mothers were of ages ≥35 years at their birth (HR, 1.7; 95% CI, 1.1-2.8) relative to women whose mothers were of ages <25 years at their birth. Although not significant, Holmberg et al. (13) reported an elevation in breast cancer mortality associated with a maternal age of ≥45 years (HR, 1.30; 95% CI, 0.85-1.98). We were unable to investigate advanced maternal age in our data because the mothers of only three women, all of whom survived, were of ages ≥45 years at their birth. Although the American Cancer Society Cancer Prevention Study 1 included women diagnosed pre- and postmenopausally, the majority of women were diagnosed after menopause (18). Both the Hertfordshire Cohort Study (12) and the American Cancer Society Cancer Prevention Study 1 (13) included primarily women who were breast cancer free in the comparison group whereas our comparison group was women diagnosed with breast cancer, which may explain differences between their studies and the present study. Our most striking finding was the substantially lower mortality from breast cancer among women born second or higher in the birth order relative to firstborn women regardless of maternal age. In the present study, birth order modified the effect of maternal age on breast cancer mortality, with the higher breast cancer mortality associated with older maternal age seen among firstborn women only. To our knowledge, no other study has investigated the association between birth order and breast cancer mortality or its effect modification with maternal age.

In agreement with studies of perinatal factors and breast cancer incidence, we found higher mortality from breast cancer for high birthweight (1); however, we also found higher mortality from breast cancer for older maternal age and lower mortality from breast cancer for higher birth order. The HR of 1.8 we found for birthweight of ≥4,000 g is close to the range of relative risks (1.5-1.7) reported in the review article (1), and is similar to the relative risk (1.7) we found in our previous study based on self-report (2). The HR of 1.7 we found for maternal age ≥35 years is higher than the relative risks for previous studies, which tended to report weak positive or no association (1), and is higher than the relative risk (1.0) we reported in our previous study (2). The HR of 0.2 we found for being second or higher in the birth order is lower than the relative risks of previous studies, which tended to report weak inverse or no association (1), and is lower than the relative risk (1.0) we reported in our previous study (2). Neither the review article nor our previous study accounted for effect modification, which has been seen in two (19, 20) of three (21) previous studies that investigated the joint effect of maternal age and birth order on breast cancer incidence.

If biological mechanisms proposed for the associations between perinatal factors and breast cancer incidence are similar for breast cancer mortality, a possible explanation for high birthweight is higher intrauterine exposure to insulin (7-9), insulin-like growth factor I (7-9), or leptin (9) based on cord blood levels. One (6) of two (5) studies that investigated cord estrogen levels found a positive association with older maternal age; thus, the Trichopoulos hypothesis (4) may hold for older maternal age and breast cancer mortality. An alternative explanation for higher breast cancer mortality associated with older maternal age is the greater likelihood of germ-cell mutations in the offspring (22) due to genetic damage of older oocytes (23) coupled with the inability of older mothers to repair DNA.⁴

We did not assess older paternal age, which in a study of Hemminki and Kyyronen (22) resulted in a similar increase in sporadic breast cancer (10%) as that seen for older maternal age. Because cord estrogen levels were not associated with being firstborn (5, 6), an alternative explanation for lower breast cancer mortality associated with late rank in the birth order relates to the fetal antigen hypothesis whereby a pregnancy-induced immune response, initiated by fetal antigens produced by paternal genes during the first pregnancy, occurs between the mother and the fetus in subsequent pregnancies (24). The effect modification we saw of maternal age by birth order would support both of these hypotheses.

There were several limitations of this study. We were unable to assess maternal preeclampsia, history of having been breastfed, and paternal age. Data on some perinatal factors were missing for a substantial percentage of women, reducing statistical power to detect associations. Reporting of perinatal factors is prone to misclassification. We are confident that our loss of accuracy in using self-report rather maternal report was not that great because our validity study among women born in Washington state showed very high correlations comparing self-report with mother report for birth order (r = 0.89) and birthweight (r = 0.85; ref. 12). Proxy respondents were used for 172 deceased cases and few of these proxies were subjects' mothers. There was no validation of proxy reports and, to our knowledge, there have been no studies of proxy reporting of perinatal factors other than maternal reporting. For birthweight, we used a larger reference group of women (2,500-3,999 g) than the typical reference group (2,500-2,999 g) because we did not want to exclude women who were able to report that they were of low or high birthweight but did not know their exact birthweight.

This study has many strengths. This is the first study to investigate mortality from breast cancer associated with a range of perinatal factors. The population-based nature of the original study and its high response rates among living cases (82.4%) and proxies for deceased cases (81.9%) minimize selection bias. We evaluated effect modification of maternal age by birth order and assessed confounding by known risk factors for breast cancer mortality, including early screening and treatment. The mean and median length of follow-up of over 10 years should have been sufficient time to detect associations if they existed.

In all likelihood, these perinatal factors interact with genetic or environmental factors leading to a poorer prognosis for some women. The protective effect of being born second or higher in the birth order against breast cancer mortality is striking and needs to be confirmed in future studies.