The role of pregnancy in maternal cancer risk: Epidemiologic evidence from the Nordic Countries Linked Birth and Cancer Registries Cohort Project

Rebecca Troisi, Ingrid Glimelius, Tom Grotmol, Mika Gissler, Cari M. Kitahara, Anne Gulbech Ording, Solbjørg Makalani Myrtveit Sæther, Camilla Sköld, Henrik Toft Sørensen, Britton Trabert and Tone Bjørge: The role of pregnancy in maternal cancer risk: Epidemiologic evidence from the Nordic Countries Linked Birth and Cancer Registries Cohort Project


INTRODUCTION
Over half a century ago, epidemiologic studies documented reduced breast cancer risk among parous women (1), particularly those who delivered children at an early age, compared with nulliparous women (2). More recently, studies have focused on various characteristics of pregnancy as markers for exposures that occur while the breast is vastly remodelling in structure and morphology. Shifts in child-bearing patterns (i.e., having fewer children and giving birth at older ages), and rising incidence of some pregnancy complications, for example preeclampsia (3), may affect cancer risk in later life. Record linkage studies, largely from the Scandinavian countries, have successfully used data from health and other registers to obtain exposure information over meaningful time frames. These large population-based datasets also address the relative rarity of most cancers. This brief review covers the epidemiology of pregnancy and its characteristics and subsequent maternal cancer risk (summarized in Table  1 (4)), based on findings from a Nordic-National Institutes of Health (NIH) collaboration of studies conducted in the Nordic Countries Linked Birth and Cancer Registries Cohort Project (Nordic Project), an effort that combined linked registry birth and cancer data from Denmark, Finland, Norway, and Sweden. While this is not a comprehensive review of the literature on this topic, we have tried to put the Nordic findings in the context of other major epidemiological work. This large collaboration and its resulting resources provided data to explore in greater detail whether profound physiological states in pregnancy are linked with subsequent cancer risk.

Breast cancer
Breast cancer, because of its higher incidence than other tumours and clear association with reproductive hormones, has been most studied regarding the influence of pregnancy. The effect of parity on the mother's breast cancer risk depends on her age at pregnancy and tumour subtype (5). For several years following delivery, there is a slight increase in breast cancer risk, mainly of estrogen receptor (ER) negative tumours. In contrast, if the pregnancy occurs at a young age, the long-term risk of ER positive tumours is reduced (reviewed in Behrens (6)). To provide benefit, the pregnancy must be fullterm, as miscarriages do not provide equivalent protection (7); it has been unclear if gestational length of pregnancy resulting in live birth is inversely associated with risk (8). Pregnancy-associated breast cancer (PABC), likely from an abundance of somatotropic hormones affecting tumour progression, is rare, but often diagnosed at an advanced stage (9). Data from the Nordic Project (10) demonstrated, as expected, that parity was inversely associated with breast cancer risk, with an approximate 25% risk decrease for ≥ 4 births vs. 1 birth. Age at birth also was strongly associated with risk: Older age at first birth and at last birth were positively associated with risk, while time since first birth and time since last birth were inversely associated with risk. In contrast to inconsistent results for whether duration of pregnancy is associated with breast cancer risk (8), in the Nordic Project data (10), gestational length was inversely associated with breast cancer risk both for the first and last birth. The association of the mother's weight gain in pregnancy with her later breast cancer risk is complicated by the strong correlations between maternal pregnancy weight and gestational length of the pregnancy, the baby's birth weight, and the mother's later adult weight. Studies for the most part have been inconsistent, however, a Finnish cohort study found that pregnancy weight gain was positively associated with subsequent breast cancer risk independent of weight at diagnosis (11). Birth weight has been positively associated with maternal breast cancer risk, but there have been few studies, with findings limited to subgroups (12). We were able to assess multiple correlated variables simultaneously in the Nordic Project (10). Having had a premature first birth (<32 weeks' gestation vs. 37-41 weeks' gestation) or premature last birth was associated with a 10-15% increase in risk. Risk was reduced by approximately 20% among women with a high maternal BMI (≥30) compared with average BMI (18.0-24.9). The breast cancer risks were increased slightly (<10%) with low birth weight (<1500 g) for the first birth and for the last birth compared with average weight (2,500-3,499 g). In the subset of pregnancies with information on gestational age and birth weight, breast cancer risk was slightly lower for small for gestational age in the first birth and slightly higher for large for gestational age compared with appropriate size for gestational age. With simultaneous adjustment, the results for multiple births, maternal body mass index (BMI), and premature birth (for the first birth) remained the same, but the elevated risk for low birth weight (for the first birth) was attenuated.
Whether twin pregnancies are a risk factor for breast cancer is unclear, although several Scandinavian cohort studies, mainly in younger women, have found a modest risk reduction in mothers of twins compared with mothers of singletons (reviewed in 13), while results from the Nordic Project (10) showed that having delivered twins (n=2,463 cases) or higher-order multiples compared to singletons was associated with a <10% increase in risk.
The effects of other factors related to pregnancy or the offspring on maternal breast cancer risk also have been investigated. Offspring sex has been hypothesized to affect maternal breast cancer risk through differences in maternal hormones, but results of studies have been inconsistent (13), and there was no association in the Nordic Project (10). Despite the paucity of information on placental size, two studies have demonstrated positive associations between placental weight and diameter (14), and an inverse association with maternal floor infarctions (15) and subsequent breast cancer in the mother. Breastfeeding duration is inversely associated with breast cancer risk, with a stronger and more consistent relation for hormone receptor-negative tumours (16).
Certain pregnancy complications also may be associated with subsequent maternal breast cancer risk, for example preeclampsia, but because these conditions are rare, the investigations require large studies like the Nordic Project to reach the statistical precision required for accuracy. Several studies have shown a reduced risk of breast cancer among women with a history of preeclampsia, a hypertensive disorder of pregnancy, compared with those with normotensive pregnancies (17). Interestingly, one study demonstrated a marked reduction with elevated mean arterial pressure (MAP) (18) and with systolic blood pressure increasing from mid-to late pregnancy below the diagnostic criterion for hypertension (i.e., in normotensive pregnancies) (19). A prevailing feature of preeclampsia is anti-angiogenesis, which is also essential in restricting tumour growth, although there are several other biological hypotheses that have been posited to explain the lower breast cancer risk.
Our findings from the Nordic Project (10) were consistent with about a 10% reduction in breast cancer risk in women who developed hypertension during pregnancy, or preeclampsia. The number of cases were too few to assess severe preeclampsia or eclampsia. Breast cancer risks were similar for gestational hypertension and preeclampsia among women without pre-existing (before the pregnancy) hypertension. The results were also similar among women without pre-existing diabetes, and with further adjustment for gestational length. No other pregnancy complication including macrosomia, hyperemesis, abruptio placenta, placenta previa, antepartum haemorrhage and poor fetal growth, were associated with breast cancer risk, except for an approximate 15% increase in risk associated with retained placenta.
Further adjustment for age at first birth did not change any of the results for pregnancy complications and breast cancer risk. The associations for hypertensive conditions and breast cancer risk were similar in women who experienced preeclampsia or gestational hypertension in the first pregnancy or in a subsequent pregnancy and were similar for all three hypertensive conditions by age at breast cancer diagnosis, and time since last and first birth. Breast cancer risk appeared slightly lower for gestational hypertension among women with one live or still birth pregnancy carrying a male compared with a female fetus, but the difference was not statistically significant. Results were similar for the three hypertensive conditions by calendar year of breast cancer diagnosis (before 2002, 2002-2008 and 2009-2013). Adjusting for smoking and BMI did not affect the results in women with this information.

Colorectal cancer
Several prospective studies have evaluated the role of pregnancy in relation to colorectal cancer risk, with inconsistent results. The EPIC (European Prospective Investigation into Cancer and Nutrition) study found little evidence for associations between parity or age at first full-term pregnancy and colorectal cancer (20). On the other hand, a US study (National Institute of Health-American Association of Retired Persons Diet and Health Study) found that age at first childbirth was positively associated with colorectal cancer, while parity was inversely associated (21). A Swedish registrybased study found parity to be positively associated with adenocarcinoma of the proximal colon (22). In a report from the Women's Health Initiative Observational Study (US), greater parity was associated with a reduced colorectal cancer risk (23). The Million Women Study (UK) reported lower colorectal cancer risk in parous than in nulliparous women, but no risk trend was observed by parity (24).
Our large population-based case-control study (25) including more than 22,000 cases within the Nordic Project found no evidence for associations between parity, age at first and last birth, and time since first and last birth and colorectal adenocarcinoma in parous women overall, by specific subsites (proximal and distal colon and rectum), or in analyses stratified by mother's year of birth, parity, and proxies for menopausal status. The study population was relatively young (mean age at colorectal cancer diagnosis was 57 years), as it was restricted to women with a prior birth recorded in national birth registries.
Future studies would benefit from the inclusion in analyses of possible confounders, such as use of exogenous hormones, obesity, alcohol, smoking, and aspirin and NSAID (nonsteroidal anti-inflammatory drug) use. Larger studies would also be valuable by allowing comparison of risks across morphologically and molecularly defined subtypes and anatomical subsites, and by taking information on colorectal cancer screening into account.

Endometrial cancer
Much of what is known about the epidemiology of uterine cancer relates to endometrial cancer, as uterine sarcomas comprise only 3-7% of uterine malignancies (26). The risk of endometrial cancer rises sharply among women in their late forties to mid-sixties and is strongly dependent on lifetime hormonal exposures. Various aspects of reproduction have been explored extensively regarding endometrial cancer.
Pregnancy is known to confer long-term protection against endometrial cancer. Conversely, nulliparity is associated with elevated endometrial cancer risk (27,28). Primary and secondary infertility are also associated with increased endometrial cancer risk, with independent effects (29). The pregnancy history of women with endometrial cancer has been examined in depth, including timing of births (27,28,(30)(31)(32), twin births, and sex of offspring (33). Several studies have reported reduced risks with either older age or shorter time since last birth. Investigators have hypothesized that this reflects a protective effect of the mechanical clearance of initiated cells (27,28,(30)(31)(32). However, existing studies have been unable to evaluate timing of pregnancy associations by histologic subtype (32). Birth of twin boys, but not twin girls or non-sexconcordant twins, appears to put women at increased risk of endometrial cancer (33). Pregnancy complications such as preeclampsia also have been explored in relation to endometrial cancer, with inconclusive results (34)(35). Gestational diabetes has been associated with increased risk of both endometrial hyperplasia and endometrial cancer (36).
Recent data indicate that the associations between reproductive factors and endometrial cancer differ by dualistic subtype (Types I and II) (32,37) motivating us to investigate endometrial cancer risk overall and by subtype (dualistic and histologic subtype if sufficiently powered) in relation to pregnancy-related factors, pregnancy complications, and birth characteristics in the Nordic Project (38). Pre-existing and pregnancy-related hypertensive conditions were associated with increased endometrial cancer risk (approximately 50-90%), with consistent associations across dualistic type (Type I and II). Increasing number of pregnancies and shorter time since last birth were associated with markedly reduced endometrial cancer risk, with consistent associations across most histologic subtypes. These findings support the role for both hormonal exposures and cell clearance as well as immunologic/inflammatory aetiologies for endometrial cancer.

Ovarian cancer
Epithelial ovarian cancer has the worst prognosis of all gynaecological malignancies with an approximate 45% overall 5-year relative survival rate (39). Parity is well known to be associated with decreased risk of ovarian cancer. Parous women have a 30-40% lower risk of developing ovarian cancer, and an additional protective effect is seen with increasing parity (39,40). Spontaneous or induced abortion does not seem to influence the risk of ovarian cancer (41), suggesting that a longer period of hormone exposure or anovulation is required for risk reduction. Findings from studies on pre-and post-term delivery have been inconsistent (42,43), and additional studies could shed light on the effect of pregnancy length. Other pregnancy-related factors such as high or low birth weight (42), preeclampsia (34,44), offspring gender (43,45), twin birth (43), and placental weight (46) have not been conclusively associated with ovarian cancer. Pregnancies at older ages seem to provide stronger protection against ovarian cancer than pregnancies at younger ages (39). Lactation is also protective, and the effect size increases with duration of breastfeeding (47).
The inconsistencies in previous studies led us to indepth study associations between pregnancy and birth characteristics and risk of epithelial ovarian cancer by histologic subtype in parous women in the Nordic Project (48). Increasing number of pregnancies was associated with reduced ovarian cancer risk. The strongest risk reductions were for clear cell and endometrioid histologic subtypes, but reduced risks with increasing number of pregnancies were observed for all subtypes. We found that preterm delivery was associated with increased risk of ovarian cancer, and the shorter the pregnancy, the stronger the association. Older age at both first and last birth was associated with a decreased risk. We found no association with twin pregnancies, preeclampsia or offspring size with overall epithelial ovarian cancer risk.
While parity is an established risk factor for epithelial ovarian cancer, the association with non-epithelial ovarian tumours has not been studied. Non-epithelial ovarian cancers are divided into sex cord-stromal tumours (SCSTs) and germ cell tumours (GCTs). We investigated the associations of pregnancy characteristics by these subtypes in the Nordic Project (49). Like our findings in epithelial ovarian cancer, the risk of SCSTs, but not GCTs, decreased with older age at first and last birth. SCST risk decreased gradually with increasing age at first or last birth. A recent childbirth (shorter time since first or last birth) was also associated with decreased risk of SCSTs (but not GCTs). Number of births, preterm birth, preeclampsia, and offspring size were not associated with risk of SCSTs or GCTs. Even with population data from four countries, the number of these tumours was limited, which could explain the lack of associations. Recently, parity was found to be associated with a better prognosis for germ cell ovarian tumours, while it did not impact survival for other ovarian cancer subtypes (50).

Leukaemia/lymphoma, sarcomas, and other solid tumours
The relations between pregnancy characteristics and other malignancies have been much less investigated than breast, colorectal, endometrial, and ovarian cancers. This is likely due to the less pronounced hormonal aetiology of these malignancies and their lower incidence.
While malignant melanoma is the most common cancer arising during pregnancy, accounting for about a third of malignancies among expectant mothers, parous women are not at higher risk of subsequently developing melanoma than nulliparous women (51), or of developing non-melanoma skin cancer (52). Women have a lower incidence of leukemias and lymphomas overall than men, but there is little evidence that pregnancy factors are associated with non-Hodgkin's lymphoma (53). One study, however, showed that parous women had about a two-and-a-half-fold increased risk of lymphoid neoplasm compared with nulliparous women, but no associations were observed between lymphoid neoplasms and other reproductive factors, including age at first birth and breastfeeding (54).
Parity is unlikely to play an important role in the aetiology and disease progression of Hodgkin's lymphoma (55,56) and the focus of research is on the ability to conceive after treatment (57,58). Pregnancyrelated hormonal or immunological changes seem to have only a minor influence in the aetiology of leukemias. However, one study did find a small tendency toward reduced risk of chronic myeloid leukaemia with higher parity (59) and another reported short-term protection against acute myeloid leukaemia with pregnancy (60). Risk of sarcoma was not associated with parity and number of abortions in one study (61), but another suggested an increased risk in women who were older at first birth (62).
Age at first birth, parity, and number of live births are not consistently related to female thyroid cancer risk (63,64). Like breast cancer, thyroid cancer risk appears to be elevated in the first few years after pregnancy but not subsequently (65,66), which may indicate a promotional effect of pregnancy hormones in thyroid carcinogenesis. Diagnosis of hyperemesis gravidarum has been associated with increased risk of maternal thyroid cancer (67). Higher fetal growth and birth weight also have been associated with an increased risk of maternal thyroid cancer (68), while longer duration of breastfeeding has been associated with a modest reduction in risk (65,69). It is not clear whether any or all of these findings are confounded by healthcare access or thyroid cancer screening, leading to the identification and diagnosis of subclinical thyroid cancers (63). These issues are being addressed in data from the Nordic Project. Meta-analyses of pancreatic cancer show a risk reduction in parous women compared with nulliparous women (70), with two children being most protective (71). Higher risk of pancreatic cancer has been associated with older age at first birth (72). In contrast, a meta-analysis demonstrated an increased risk of kidney cancer in parous compared with nulliparous women, and an increase in risk with each subsequent birth (73).

USE OF NORDIC REGISTERS IN RESEARCH
Understanding the role of pregnancy on subsequent maternal cancer risk is challenging because of a relatively long latency period between exposure and disease, the possibility of bias in recall of information about pregnancy in case-control studies based on interviews or questionnaires, and the large numbers of cases required for stable estimates when studying rare exposures or cancer types.
For several decades, the Nordic countries have collected administrative health and welfare data (74,75), which can be leveraged in studies of pregnancy characteristics and subsequent cancer risk. Registrybased research in the Nordic countries has benefited from Personal Identification Numbers, making it possible to link various data sources, including all other registers, also for example medical records and biobank samples. This has allowed the study of rare exposures, such as eclampsia or other pregnancy complications recorded in medical birth registries or rare malignancies recorded in cancer registries.
The use of registries also has the advantage of minimizing bias. As all residents who had a birth or a cancer diagnosis in the Nordic countries are captured in the registries, selection bias introduced by including patients treated by specific hospitals or with specific insurance plans is virtually eliminated.
Standardized data from registries also eliminate bias from participant self-selection and recall. Information on perinatal factors based on mandatory reporting of birth information supplements the analysis, and while it can be missing for some variables, lack of data is unlikely to be related to subsequent cancer risk. Information on pregnancy complications especially in the early years of the birth registries is not complete, and could be misclassified, but any misclassification would tend to bias results towards the null.

CONCLUSIONS
An understanding of the origin of cancer is crucial for cancer screening, prevention, and treatment. Complex biological mechanisms promote carcinogenesis, and there is increasing evidence that pregnancy-related exposures may have long-lasting impact on health and disease susceptibility in the mother. In addition, understanding the role of pregnancy in the subsequent health of the mother is important as women are experiencing pregnancy at older ages and are having less children. This review has provided evidence that some pregnancy and pregnancy-related factors are involved in the carcinogenesis of several cancer types. The abilities to link various health registries and to pool data across the Nordic countries has provided opportunities to conduct high-quality research of pregnancy exposures and subsequent maternal cancer risk. Unfortunately, this is becoming increasingly difficult due to legal restrictions, although research can employ metaanalysis approaches in joint projects. The future should also aim to integrate biological data into these large studies, further deepening the understanding of the differences in aetiology for malignancies in reproductive organs.
The next step in increasing the utility of these resources will involve linking with other population data, such as prescription or in-vitro fertilization registers, and with biospecimens, most likely through birth cohort biobanks (76,77). The latter may lead to the identifycation of persistent changes in epigenetic markers that could represent pregnancy or in utero exposures, which could then be associated with cancer risk. These approaches could improve our understanding of potential preventable causes of cancer.