Increased Mortality in Women with Breast Cancer Detected during Pregnancy and Different Periods Postpartum

Pregnancy-associated breast cancer (PABC) commonly refers to a breast malignancy diagnosed during pregnancy or within one or 2 years following delivery (1). Because of a continued trend toward postponed childbearing, the incidence of PABC is likely to increase in many Western countries.

Pregnancy and breast cancer has often been viewed as an incurable combination. Several studies have found evidence of a diagnostic delays in pregnant women (2–7), which may explain why results from some, but not all, studies show that women with PABC have larger tumors and are more likely to have positive nodes, metastases and vascular invasion compared with nonpregnant women with breast cancer (2, 5, 8–10). A poor prognosis in PABC (4, 8–14) has previously primarily been attributed to diagnostic delays (7, 15). However, few studies have investigated whether the prognosis varies by time of diagnosis (during pregnancy and different postpartum time intervals). Some studies have distinguished between pregnant and lactating women (2, 4, 5, 10, 14), while most investigators have focused on cancers diagnosed after childbirth in 1-year intervals following delivery (8, 9, 12, 13). In all these studies, ranging in size from 35 to 708 cases of PABC, proximity to pregnancy was associated with a worse prognosis. Another question of interest relates to whether mortality varies by time since diagnosis. Guinee and colleagues (10) found evidence of a mortality peak occurring 2 years after diagnosis in women with PABC.

In this large nation-wide Swedish study encompassing more than 1,100 cases of PABC, we used a detailed classification of PABC to assess all-cause mortality in relation to time of diagnosis (during pregnancy and in postpartum intervals) and also whether mortality varies over time from diagnosis.

We conducted a population-based study using data from the Swedish National Cancer Register. The study cohort was defined by women born 1948 and later with a breast cancer diagnosis between ages 15 to 44 years during 1963 to 2002. To determine whether the diagnosis was pregnancy-associated (i.e., occurring during or within 2 years after pregnancy), information on dates of all births was retrieved from the Multi-Generation Register from 1961 and onwards. Information in the 2 registers was matched by means of record linkage using the national registration number, a unique identifier assigned to all Swedish residents (16).

We further linked the study cohort to the Swedish Cause of Death Register and the Population Register to obtain outcome and follow-up information (deaths, migrations). Information on educational level for each woman was obtained from the Education Register. Only first breast cancers (n = 15,800) were included. We excluded 71 women with inconsistent migration history (immigration or emigration events not recorded in logical order) and 8 women diagnosed on the same day as they died. The final cohort for analyses encompassed 15,721 women with a breast cancer diagnosis between ages 15 to 44, of whom 1,110 (7%) had a PABC.

In this study, PABCs were defined as malignancies diagnosed within 9 months before the delivery date (during pregnancy) or within 2 years after delivery. Breast cancers occurring outside this time window were referred to as non-PABC. For women with 2 childbirths within the time window (n = 49), we chose the second pregnancy as the index birth related to the PABC.

Parity status was defined as number of births prior to diagnosis. For women experiencing PABC, the index birth was not counted in the parity variable.

On the basis of information in the Cause of Death Register, 91% of PABC cases and 90% of non-PABC cases who died had breast cancer listed as the underlying cause of death. For the purpose of this study, all-cause mortality was chosen as the main outcome.

Person-time at risk was counted from date of breast cancer diagnosis until date of death or censoring. Censoring events were first emigration after diagnosis, end of study (December 31, 2003) or 15 years after diagnosis, whichever came first.

We estimated crude mortality rates, defined as the number of events over total person-time at risk, with 95% CI based on the Poisson distribution. Mortality rates were calculated using days as time unit and reported per 1,000 person-years. The underlying timescale was time since breast cancer diagnosis. The mortality rates were modelled and adjusted for confounders using flexible parametric survival models (17, 18), which use a spline function for the baseline mortality rate. These models are proportional hazards models and provide HRs with 95% CIs as the measure of association between exposures and outcome. It is possible to estimate various fitted curves from the models, such as mortality rates and HRs over time.

In the modeling, we firstly investigated possible associations between mortality and the main exposure non-PABC (reference) and PABC. We also investigated associations between mortality and proximity of PABC diagnosis to delivery (categorized as breast cancer diagnosis during pregnancy (0–9 completed months prior to delivery), during months 1 to 3, 4 to 6, 7 to 12, 13 to 18, or 19 to 24 after delivery, or non-PABC (reference)). Models were adjusted for age at breast cancer diagnosis (categorized as 15–29, 30–34, 35–39, or 40–44 years), calendar year at diagnosis (categorized as 1963–1974, 1975–1989, or 1990–2002) and educational level (≤9 years, 10–13 years, university undergraduate level, or postgraduate level). We investigated parity prior to breast cancer (categorized as nulliparous, 1 childbirth or ≥2 childbirths), and the interaction between PABC and parity (categorized as PABC nulliparous, PABC parous, non-PABC nulliparous, and non-PABC parous). All variables were categorized prior to the analysis. The significance level was 5% and tests were 2-sided.

Second, we modelled the mortality rates for PABC and non-PABC allowing the HRs to vary over time since diagnosis. Within the flexible parametric modeling framework, it is possible to allow exposures to have time-dependent effects (nonproportional hazards) by fitting separate spline functions for the different levels of the exposure. Differences between groups are still reported as HRs, but now represent a function over time rather than a constant. The time-dependent HRs are reported at 2, 5, and 10 years after diagnosis. The baseline hazards were modelled using splines with 5 degrees of freedom (4 internal and 2 boundary knots), while the time-varying effects used splines with 3 degrees of freedom. A sensitivity analysis showed that our spline estimates were robust for different choices of knots.

Only women with complete information on all variables were included in the models (the only variable with missing values was education, 563 (4%) missing of 15,721). The data was analyzed using Stata Intercooled 11.0 (19). The flexible parametric models were estimated using the Stata command stpm2 (17).

The study was approved by the Research Ethics Committee at Karolinska Institutet.

The mortality rate was higher in young (<40 years at diagnosis) compared with older breast cancer patients (40–44 years at diagnosis) and was higher among patients diagnosed in the earlier periods (1963–1974 and 1975–1989) compared with the most recent period under study (1990–2002). The mortality rate increased with decreasing educational level (Table 1).

Among 1,110 women with PABC, 515 (46%) died during 15 years of follow-up, while among 14,611 women with non-PABC, 4,911 (34%) died (Table 1). The average time of observation was 7.5 years in women with PABC and 8.9 years in women with non-PABC. The crude mortality rate among PABC was 61.9 per 1,000 person-years (95% CI: 56.7–67.4) and 37.6 (95% CI: 36.6–38.3) among non-PABC, indicating a much poorer prognosis among women diagnosed in connection with childbearing. Compared with non-PABC, PABC was associated with a significantly increased mortality, both before (unadjusted HR 1.61, 95% CI: 1.47–1.77) and after adjustment for age at diagnosis, calendar period and education (adjusted HR 1.51, 95% CI: 1.36–1.68).

The PABC category was further subdivided by time between diagnosis and delivery. The poorest prognosis was observed in patients diagnosed during months 4 to 6 after delivery who had a more than 2-fold increased mortality rate (adjusted HR 2.45, 95% CI: 1.83–3.29) compared with non-PABC women. An almost doubled mortality rate was observed in women diagnosed during pregnancy (adjusted HR 1.85, 95% CI: 1.34–2.56), while corresponding increases in mortality rates among women diagnosed during 1 to 3 and 13 to 24 months after delivery were around 30 percent.

Parity prior to breast cancer was not associated with the risk of death after adjustment for age, calendar period, and education, and did not modify the effect of PABC on mortality, neither before nor after adjustment. Among non-PABC cases, mortality rates were similar in parous and nulliparous women (adjusted HR 0.95, 95% CI: 0.88–1.03). Compared with nulliparous women with non-PABC, women with PABC and no history of earlier births had an adjusted HR of 1.30 (adjusted, 95% CI: 1.05–1.62) while parous women with PABC had an adjusted HR of 1.51 (95% CI: 1.33–1.72).

The mortality rates varied with time from diagnosis, with a peak about 2 years after diagnosis both among PABC and non-PABC cases (Fig. 1). The difference in mortality between PABC and non-PABC cases was most pronounced during the first 5 years of follow-up, and was not detectable after around 8 years. The highest mortality rate was observed in women diagnosed during months 4 to 6 following delivery, with almost 250 deaths per 1,000 person-years at 2 years of follow-up. Women diagnosed during months 1 to 3 after delivery had a peak rate of approximately 150 deaths per 1,000 person-years (which occurred at around 18 months after diagnosis), whereas women diagnosed during months 7 to 24 after delivery had a peak rate just above 100 deaths per 1,000 person-years. Among women with non-PABC, the peak mortality was around 50 deaths per 1,000 person-years.

Figure 2 shows how the HRs for women with PABC compared with non-PABC (adjusted for age, calendar period, and education) vary by time since diagnosis and by proximity of breast cancer to delivery. Among PABC diagnosed during pregnancy (panel A), the HR was 2-fold increased during the first 2 years after diagnosis and decreased after approximately 10 years to the same level as non-PABC. Among PABC diagnosed during months 1 to 3 after delivery (panel B), the same pattern was observed. Among PABC diagnosed during months 4 to 6 after delivery (panel C), there was a 4-fold peak in HR at 2 years following diagnosis. Much of the mortality increase was diminished after 5 years, but some difference still remained after 10 years, albeit close to nonsignificance. Among PABC diagnosed during months 7 to 12, 13 to 18, and 19 to 24 after delivery (panels D, E, and F, respectively), there were similar patterns with significantly increased mortality in the first 5 years, but HRs were lower compared with PABC diagnosed during months 4 to 6 after delivery. Among PABC diagnosed 13 to 18 and 19 to 24 months after delivery, there was also an indication of a protective effect beyond 10 years after diagnosis. These groups were larger in numbers, as indicated by the narrower confidence bands.

To obtain point estimates from the curves, we extracted estimates of HRs and 95% CIs at 2, 5, and 10 years after diagnosis from the model adjusted for age, calendar period, and education (Table 2). Compared with non-PABC, the mortality was almost doubled among PABC 2 years after diagnosis (HR 1.8, 95% CI: 1.6–2.2). The mortality difference between PABC and non-PABC was reduced over time, and no difference was present after 10 years (HR 1.0, 95% CI: 0.8–1.3). When PABC was further subdivided by proximity of breast cancer diagnosis to delivery, the highest mortality was observed in women diagnosed during months 4 to 6 after delivery, who had an almost 4-fold increased mortality rate compared with non-PABC at 2 years after diagnosis (HR 3.7, 95% CI: 2.4–5.8). After 10 years, the increased mortality was eliminated in all time windows of PABC, except in patients diagnosed during months 4 to 6 after delivery (HR 2.1, 95% CI: 1.2–3.8).

The impact of PABC on mortality over follow-up time was the same among women with and without childbirths prior to breast cancer diagnosis (Table 2). Thus, regardless of parity status, mortality rates were strongly driven by whether the breast cancer was pregnancy-associated or not.

We found evidence of a higher mortality in women with PABC compared with women with non-PABC of the same age, educational level, and calendar year of diagnosis. We observed a mortality peak around 2 years after diagnosis for both PABC and non-PABC cases, when the largest difference in mortality between the 2 groups also occurred. A second important finding was that the elevated mortality among PABC women varied markedly with timing of diagnosis in relation to delivery. Compared with non-PABC patients, we observed nearly 4 times higher peak mortality in women diagnosed with a breast cancer 4 to 6 months postpartum. The second highest mortality rate was found in women diagnosed during pregnancy or within 3 months after birth. The mortality decreased with increasing time from the index delivery, and 8 years following diagnosis there was no detectable mortality difference between PABC and non-PABC cases.

Strengths of this study included the use of data from population-based registers that ensured virtually complete ascertainment of breast cancer cases, deliveries, and deaths during 4 decades. This provided high statistical power and minimized selection and information bias. Data in the Swedish Multi-Generation Register allowed us to identify PABCs, but we were unable to identify pregnancies that were electively terminated due to a breast cancer diagnosis. Also, PABC cases occurring after age 44 years were not included in the study, but more than 95% of Swedish women have completed their childbearing at age 40 years (27).

The analyses were adjusted for several known confounders, including age at diagnosis, calendar period at diagnosis, and education. A major weakness of this study was the lack of information in the Swedish National Cancer Register on prognostic factors, such as tumor characteristics (e.g., tumor size, lymph node status, hormone receptor status, stage, grade, and lymph-vascular invasion) and treatment. Also, no information was available on breast feeding. From an international perspective, the frequency of breastfeeding in Sweden is high, with more than 90% of all infants being exclusively or partially breastfed at 2 months in the mid-1990s (28). We could not examine and control for confounding from physical activity, body mass index or the use of oral contraceptives. However, these life-style factors are unlikely to have varied between PABC and non-PABC cases. Finally, we had no information on family history of breast cancer.

There may be several explanations for a poorer prognosis in women with PABC, including delayed diagnosis, lower treatment intensity, stage and aggressiveness of the tumor, promotion by hormonal factors, and a tumor promoting microenvironment.

First, because changes in the breast tissue may be attributed to the pregnancy or lactation rather than a malignancy, patients with PABC may be more likely to experience diagnostic delays (7), leading to a more advanced stage at diagnosis. Our finding of a higher mortality among women diagnosed during and shortly after pregnancy indirectly supports this hypothesis. A recent review reported that the diagnostic delay may range from 1 to 9 months (7). However, the increased mortality observed in women diagnosed up to 2 years after delivery is unlikely to be explained by diagnostic delays related directly to pregnancy. Second, there may be postponement of the initiation and the aggressiveness of treatment, because of conflicting interests between the cancer treatment and maintaining the pregnancy. Such delay and lower treatment intensity are likely to be reflected in an increased mortality for breast cancers diagnosed during pregnancy, but not those diagnosed after delivery. Third, a biological mechanism for a more aggressive potential of pregnancy-related tumors may be a selection and growth advantage of particularly malignant cells due to exposure to high levels of pregnancy hormones (20–22), primarily produced by the placenta. High placental weight, an indirect marker of pregnancy hormones, was recently associated with increased PABC mortality (23). This mechanism would give rise to increased mortality among all PABC cases, but primarily among those diagnosed close to delivery when hormone levels are high. Lastly, our finding of a particularly high mortality in women diagnosed with breast cancer during months 4 to 6 after delivery, when many women begin to wean the infant, may reflect a tumor promoting tissue microenvironment following involution of the breast, that is, the regression of the mammary gland to its prepregnant state. According to this novel hypothesis, the involution process is inflammatory-like, that is, with increased immune cell influx and breakdown of the stroma surrounding the gland and ducts, changes which have been postulated to enhance tumor growth and metastasis (24).

A peak in mortality at around 2 years after diagnosis is possibly reflecting tumors which do not respond to treatment. However, the maintained increased mortality for up to 8 years after diagnosis in PABC cases is likely to reflect some other prolonged effect of pregnancy.

Childbearing has been associated with a transient increase in risk for breast cancer that is more pronounced after a first pregnancy (25, 26). However, similar to a previous study (12), we found no influence of previous parity on mortality among PABC women; an increased mortality among PABC women was present in both parous and nulliparous women.

On the basis of previous Swedish estimates of PABC incidence (29) and U.S. birth statistics (30), we estimate that around 1,600 U.S. women who gave birth in 2006 were diagnosed with pregnancy associated breast cancer. Applying the present mortality estimates, 480 of these women would be expected to die from their cancer within 5 years after diagnosis. If they would have experienced the same mortality rates as non-PABC cases, only 290 would succumb to the disease during the same time period.

Although PABC accounts for a small proportion of the total breast cancer burden in women aged 15 to 44 years, it accounts for a relatively high proportion of all cases diagnosed in women aged 25 to 29 years, where at least one in 5 breast cancers coincide with pregnancy or lactation (3). A poor prognosis in women with PABC highlights the importance of optimizing management of this special group of young patients. An improved understanding of the underlying mechanisms for the high mortality in women diagnosed shortly after delivery may open up new venues for prevention and treatment.

In conclusion, we found evidence of an increased mortality among women with PABC which peaked around 2 years after diagnosis. Because the mortality increase was most pronounced in the subgroup of PABC women diagnosed shortly after delivery, these results give indirect support to the hypothesis that postpartum changes in the mammary gland microenvironment may enhance tumor growth and metastases. Thus, events following pregnancy may be more important than events during pregnancy. Our findings may also reflect a detrimental influence of diagnostic delays during pregnancy and lactation, as well as a promoting effect on breast cancer development from exposure to high levels of pregnancy hormones.

No potential conflicts of interest were disclosed.

This work was supported by grants from the Swedish Cancer Society (grant number 07-0526; Mats Lambe) and Susan G. Komen for the Cure (grant number KG 100116; Mats Lambe).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.