A Linked-Registry Study of Gestational Factors and Subsequent Breast Cancer Risk in the Mother

Exposures that occur during pregnancy have important consequences for the subsequent breast cancer risk of the mother; however, the underlying mechanisms are not understood. An early age at first birth is associated with one of the largest risk reductions, and each subsequent pregnancy confers an additional though smaller benefit (1). Pregnancy complications such as preeclampsia (2–5) and placental abruption (6) also may be related to decreased risk of maternal breast cancer and gestational diabetes with increased risk (7), although results are inconsistent across studies.

Despite the likely importance of these pregnancy characteristics, there have been few attempts to comprehensively evaluate several potentially relevant pregnancy conditions. Interview studies may be prone to bias in maternal recall of events or misclassification of reported conditions. Pregnancy complications are rare and large datasets are required to identify sufficient numbers of cases. Furthermore, associations between pregnancy characteristics and breast cancer risk may differ by age at diagnosis or by time since the index pregnancy, as risk has been shown to increase among parous women during the first 2 decades following delivery before ultimately declining, in comparison with nulliparous women (8). We used linked cancer-birth registry data to examine maternal, gestational, and neonatal factors in relation to maternal breast cancer risk.

Human Subjects Protection Committee approval was granted by the appropriate agencies before the study was conducted.

Incident cancer cases have been registered in the Cancer Surveillance System (CSS) of western Washington since 1973 and the Washington State Cancer Registry (WSCR) since 1993. The CSS records data of all newly diagnosed cancers (except nonmelanoma skin cancer) occurring in residents of the 13 counties of western Washington State, with a base of more than 3.5 million people representing approximately 70% of the state population. Since 1993, WSCR, established as part of the Centers for Disease Control's National Program of Cancer Registries, has collaborated with the CSS [affiliated with the National Cancer Institute's Surveillance Epidemiology and End Results (SEER) Program] to coordinate surveillance of incident cases statewide. WSCR also exchanges data with 30 other states, to ascertain residents diagnosed/treated for cancer out-of-state. Both registries participate in comprehensive quality control activities at the national level for completeness and accuracy (9, 10)_. Registry data abstracted from medical records at diagnosis include demographic information, diagnosis date and extent and stage of disease, first course of treatment, and annual follow-up to assess vital status.

Washington State Health Department birth records linked to the hospital discharge records for delivery admissions were accessed, and linkage of these data to the State cancer registries was conducted for this project, similar to previous studies (11–13).

All women younger than 85 years with primary, in situ, or invasive breast cancer diagnosed from 1974 to 2009 were identified in the cancer registries (N = 111,914). From among these, 22,758 women were identified with deliveries occurring before cancer diagnosis, based on linkage of their cancer registry records to Washington State live birth and fetal death certificates for 1968 to 2009. A case's “index” delivery was defined as the first live birth delivery recorded in the birth registry, regardless of parity (not necessarily her first-ever delivery); for a majority of cases (64%), this represented a first delivery. For each case, 10 controls were selected from the birth records among women without breast cancer, matched on parity at the time of the index delivery (0, 1, 2, 3, and 4+), calendar year of delivery, age at index delivery (±1 year), and race/ethnicity [White, African American, Native American, Asian (Japanese, Chinese, Korean, Vietnamese, Asian Indian, Other Asian), other]. A total of 224,721 controls were identified. The majority 22,091 (97.1%) of cases had 10 controls identified; 2.0% had 5 to 9 controls identified; and 0.9% had less than 5 controls identified. For 53 cases (0.2%), only a single control could be identified after relaxing race/ethnicity requirements into White/non-White.

Maternal information available from the birth certificate (check box, or as numeric fields) included: age; parity; prior fetal loss; singleton/multiple gestation; complications such as preeclampsia, chronic hypertension, gestational or established diabetes, placental abruption, placenta previa, bleeding during pregnancy, and postpartum hemorrhage; type of delivery (vaginal and c-section); pregnancy weight gain and prepregnancy weight, height and body mass index [BMI; weight (kg)/height (m)²]. Data on potential confounders such as marital status, race/ethnicity, and education were available. Neonatal characteristics included birth weight and gestational age, and offspring sex. Sex-specific, ultrasound-based growth curves (14) were used to calculate predicted birth weight for gestational age using the coefficients associated with the curves with the present data, to create expected values for birth size based on gestational age.

Additional information for deliveries from 1987 onward included the discharge diagnosis codes (up to 9 fields per hospitalization of ICD-9-CM codes) and insurance payer billed at discharge based on linkage of the birth certificate to the hospital discharge record for the mother's birth hospitalization. This information combined with the birth certificate check boxes allowed more accurate identification of maternal medical conditions and pregnancy complications (15–17). The following conditions were considered to be present in the pregnancy if the discharge record contained these codes and/or they were indicated by check box in the birth certificate: established/chronic diabetes (250.0–250.9x, 362.0x, 648.01–648.02); gestational diabetes (648.8); chronic hypertension (401–405, 642.0–642.2, 642.7, 642.9); preeclampsia/eclampsia (642.4–642.6; referred to henceforward as preeclampsia), hemorrhage in early pregnancy (640.x), and late or antepartum bleeding (641.3, 641.8, 641.9); postpartum hemorrhage (666.x); placenta previa (641.0, 641.1, 762.0); and placental abruption (641.2, 762.1). Additional conditions that were identified by hospital discharge record alone, as no birth certificate check box existed, were: gestational hypertension (642.3); placental anomalies/insufficiency (656.7–656.9, 762.2); and placenta accreta (667.x).

To evaluate the associations of maternal, gestational, and neonatal characteristics in relation to breast cancer among parous women, a history of having an index pregnancy affected by the exposures of interest was compared among breast cancer cases and their controls. Analyses were restricted to time periods for which relevant data were available for each exposure. Initially, stratified analysis was conducted to examine data quality and missing values, and to estimate preliminary odd ratio (OR) estimates of the relative risk and 95% confidence intervals (CI). Subsequently, unconditional logistic regression was used to estimate ORs, with adjustment for the matched variables (parity, calendar year of delivery, age at delivery, and race/ethnicity). Other variables considered for their possible effects on the associations of interest included marital status, type of medical insurance (Medicaid, other) as a proxy for socioeconomic status/access to care, number of prior (to the index delivery) pregnancies, and total deliveries identified before diagnosis.

The characteristics of multiple gestation pregnancies (or of women carrying them) may differ from those of singleton pregnancies, and thus, after measuring the association of multiple gestation pregnancy in relation to breast cancer, the remaining analyses were restricted to women with singleton deliveries. Analyses of gestational factors and breast cancer risk were repeated by age at diagnosis (<50 years/50+ years) and time since the index pregnancy (<15 years/15+ years). Differences by these potential effect modifiers were assessed using a likelihood ratio test, adding interaction terms to a model containing the main effects. Additional analyses were restricted to the first pregnancy, to women who had only one pregnancy on record and no fetal deaths, to cases diagnosed within 5 years of index pregnancy, and to invasive breast cancer. α level was 0.05.

Twenty-eight percent of cases were younger than 45 years of age at breast cancer diagnosis, 39% were 45 to 54 years, and 33% were 65 years or older (data not shown). The majority of cases had local (47%) or in situ (19%) disease, 31% were staged as regional, and 3% had distant disease. The elapsed time from index delivery to breast cancer diagnosis was less than 5 years for 4%; 5 to 9 years for 9%; 10 to 19 years for 28%; 20 to 29 years for 36%; and 30+ years for 24% of cases.

A majority of cases and controls were White and married at the time of the index pregnancy (Table 1). Cases and controls were similar in education level but cases were slightly less likely to have Medicaid insurance. Similar proportions of cases and controls had fetal deaths. Index pregnancies were multiple gestations for 1.2% of cases and 1.8% of controls (OR, 0.65; 95% CI, 0.57–0.74; data not shown; additional adjustments had no effect, except as indicated). Having ever had a multiple gestation before diagnosis also was associated with a decreased OR (0.83; 95% CI, 0.76–0.92; additionally adjusted for parity). The association with multiple gestations was consistent by age at diagnosis (<50 or 50+; OR, 0.68; 95% CI, 0.56–0.81 and OR, 0.63; 95% CI, 0.53–0.76, respectively), and by years from index pregnancy to diagnosis (<15, 15+; OR, 0.65; 95% CI, 0.52–0.81 and OR, 0.66; 95% CI, 0.56–0.77, respectively). Because the characteristics vary from those of singleton pregnancies, we excluded the 269 cases and 4,060 controls with multiple gestation pregnancies, leaving 22,489 cases and 220,546 controls for the remainder of analyses.

Cases and controls were similar with respect to infant gender (Table 2), and there was no clear pattern of offspring sex and breast cancer risk by the number of male and female pregnancies, with adjustment for parity (data not shown). Infants of cases were less likely to be born postterm (43–48 weeks gestation) than at term (38–<43 weeks), but there was no trend in risk with gestational length. There was a weak positive trend for birth weight and breast cancer risk; addition of a squared term for birth weight was not statistically significant (data not shown). Similar results were observed for size-for-gestational-age (8% increase per 1,000 × g; 95% CI, 5%–11%). Prepregnancy obesity (BMI 30+) and weight categories ≥140 lbs also were associated with modestly decreased ORs. Compared with gaining less than 25 pounds, weight gains of 31 to 39 pounds were associated with an increased OR, but risk did not increase with weight gain more than 40 lbs. Prepregnancy BMI was inversely correlated with pregnancy weight gain (r = −0.21; P < 0.0001). The results for prepregnancy obesity were similar with adjustment for weight gain (0.78; 95% CI, 0.65–0.92), whereas the OR for weight gain more than 40 lbs was attenuated (1.08; 95% CI, 0.92–1.26). There was little evidence of altered breast cancer risk by history of several pregnancy complications.

Reported levels of gestational diabetes, chronic and gestational hypertension, and preeclampsia increased during the study period (data not shown). The ORs for breast cancer in relation to diabetes by 10-year calendar time periods showed no strong or consistent pattern (data not shown). The ORs for preeclampsia over the study period were 0.87 (95% CI, 0.76–1.00) for 1980 to 1989; 1.15 (95% CI, 1.00–1.33) for 1990 to 1999; and 0.85 (95% CI, 0.60–1.20) for 2000 to 2009. The ORs for preeclampsia and breast cancer risk were not different when adjusted for BMI, or when assessed by gestational week at delivery (<34 weeks vs. 34+ weeks), infant birth weight (<2,500, 2,500–<4,000, and 4,000 g+), or whether the mother had diabetes (data not shown). The OR associated with preeclampsia seemed greatest among women with a low BMI (3.32; 95% CI, 1.13–9.79 for BMI <18.5; and 1.09, 95% CI, 0.83–1.44; 1.16, 95% CI, 0.78–1.74; and 0.88, 95% CI, 0.55–1.41 for BMIs of 18.5–<25, 25–<30, and 30+, respectively), although CIs were often wide and overlapping.

Results for the pregnancy characteristics were generally similar when examined by age at diagnosis as a proxy for pre- and postmenopausal breast cancer, with a few possible exceptions (Table 3). The OR for placental abnormality/insufficiency for diagnosis at age 50+ years was 2.01 (95% CI, 1.12–3.59) and for diagnosis at <50 years was 0.99 (95% CI, 0.77–1.28; P_interaction = 0.04). Pregnancies with a diagnosis of placental insufficiency were about 50% to 100% more likely to have a concurrent diagnosis of the other placental complications (i.e., preeclampsia, placenta accreta, placental abruption, placenta previa, and abnormal bleeding; data not shown). The lower risk in mothers of infants born postterm compared with term was present among women who were diagnosed 50+ years of age but not <50 years (P_interaction = 0.03).

The previously reported transient increase in breast cancer risk following pregnancy prompted us to examine associations separately by elapsed time between index pregnancy and cancer diagnosis (<15 years/15+ years; Table 4). The increasing trend in risk with increasing birth weight was present in those diagnosed 15+ years after the index pregnancy (OR, 1.08; 95% CI, 1.05–1.08) but not in those diagnosed within 15 years (OR, 1.01; 95% CI, 0.97–1.07; P_interaction = 0.05). The inverse association with prepregnancy obesity was slightly more pronounced in women diagnosed within 15 years of the index pregnancy, whereas the ORs for first trimester, late pregnancy, and postpartum bleeding were increased among those diagnosed later, although the interaction was only statistically significant for first trimester bleeding (P = 0.03). With further stratification by age at diagnosis (data not shown), the increased risks for first trimester bleeding (OR, 3.35; 95% CI, 1.48–7.55) and placental abnormality/insufficiency (OR, 2.24; 95% CI, 1.08–4.67) were particularly pronounced in women diagnosed 15+ years after the index pregnancy and at 50 years of age or older. In addition, an increased OR (3.02; 95% CI, 1.51–6.03) for very low birth weight (<1,500 g) was present in older cases who were diagnosed within 15 years of the index pregnancy. Results for cancer diagnosed within 5 years of delivery (N = 965 cases) were generally similar to associations for less than 15 years with a few exceptions (data not shown); the ORs for gestational diabetes (OR, 1.32; 95% CI, 0.99–1.77), chronic hypertension (OR, 1.31; 95% CI, 0.91–1.88), preeclampsia (OR, 1.11; 95% CI, 0.85–1.45), and placental insufficiency (OR, 1.27; 95% CI, 0.83–1.92) were more markedly increased, but for only placenta accreta (OR, 2.39; 95% CI, 1.29–4.40) did the CI exclude one.

The results for breast cancer risk were similar in analyses restricted to characteristics of a woman's first pregnancy (n = 14,246 cases and n = 140,449 controls), and when restricted to women who had no subsequent pregnancies on record after the index pregnancy and no known fetal deaths (10, 238 cases and 99,804 controls; data not shown). In addition, results were similar after excluding in situ disease (data not shown).

In this large study of data from linkage of state birth and cancer registries, results for multiple gestation pregnancies were consistent with a reduced risk of maternal breast cancer, and there was evidence of an increased risk with higher infant birth weight. Breast cancer risk was lower in women who were obese prepregnancy. The results for a history of placental abnormality or insufficiency and bleeding in the pregnancy and postpartum suggested an increased risk that was most pronounced among women who were >50 years of age at diagnosis.

A recent review of studies assessing the characteristics of pregnancy and breast cancer risk in the mother concluded that data for most factors were few and when available showed inconsistent results (18). This could be due to misclassification in exposure assessment, differences in populations studied or inadequate statistical power. In addition to our study's relatively large number of breast cancer cases, an important strength was the ability to use data collected in a generally similar fashion across many years in population-based registries, thereby avoiding the possibility of selection and recall biases.

Several (18), but not all (19, 20) studies have reported reduced breast cancer risk among women with a history of preeclampsia, although the association with in situ disease is unclear (21). Breast cancer risk may be decreased in relation to other markers of placental compromise including maternal floor infarctions and smaller placental size (6). However, we observed no consistently decreased risk of breast cancer associated with preeclampsia, and observed increased risks associated with selected placental abnormalities in apparent contradiction to an hypothesized relationship of reduced tumor occurrence among women with an antiangiogenic profile (6). Inconsistencies in results among studies may derive from the increase in preeclampsia incidence or diagnosis over time and possible emerging differences in its etiology. In our data, the OR for preeclampsia and breast cancer suggested an inverse relationship in the earliest calendar years that was not consistent over time. Women with preeclampsia due to overweight or obesity may remain heavy postpregnancy, and thereby have decreased risk of premenopausal breast cancer but increased postmenopausal risk associated with obesity (22), although adjusting for prepregnancy BMI did not change the risk estimate for preeclampsia. In women with very low prepregnancy BMI, the OR for breast cancer associated with preeclampsia was increased, whereas among women who were heavier prepregnancy no increase was observed suggesting that the preeclampsia-breast cancer association may vary by either prepregnancy weight or differential preeclampsia pathway.

Bleeding during the pregnancy and postpartum was associated with an increased breast cancer risk, with the association most pronounced in those diagnosed at 50+ years of age and 15+ years after the index pregnancy. A Swedish study reported that breast cancer incidence was approximately twice as high in women with a history of bleeding in late pregnancy (19). To the extent that bleeding indicates abnormal levels of pregnancy hormones, for which the placenta is a major source, our results related to bleeding and placental abnormalities are consistent with an important role of pregnancy hormones in breast cancer occurrence. Women who experience bleeding in the pregnancy may be more likely to deliver early, but delivering at an earlier gestational age was not associated with an increased risk in our data. We observed an increased risk of cancer within 5 years of delivery associated with placenta accreta; although this may have been a chance finding as it was not observed overall or for any other subgroup. However, it also suggests a role for abnormal placentation or vascularization potentially associated with tumor risk, although larger studies are required to examine this rare exposure. Diabetes, either established prepregnancy or gestational, was not associated with breast cancer in our data, except for a suggestion of a possible increased risk for breast cancer diagnosed within 5 years of delivery, which, if true, may be related to increased circulating insulin-like growth factor-I (IGF-I) levels, potentially relevant to breast cancer (23). A pooled analysis of 6 studies on gestational diabetes and breast cancer, however, found no association (24).

Findings on the association of gestational age or preterm delivery of offspring with breast cancer risk are inconsistent, although a few studies reported an increased risk with shorter gestational length (6). In a Norwegian study (25), breast cancer risk was increased among women who delivered their first birth prematurely (relative risk, 1.22; 95% CI, 0.97–1.53); risk decreased with increasing gestational length (based on 5,474 first births). Although there was no trend between gestational length and breast cancer risk in our data, risk was reduced among women who delivered postterm.

Having a multiple gestation pregnancy in our study was associated with an approximately 35% decreased OR for breast cancer. Several Scandinavian cohort studies (26–31), using data similar to ours, reported decreased risks ranging from 10% to 30% among women with multiple births. In some studies, risk reduction was most pronounced for twins in the last pregnancy (2, 26, 27), but not in others (29). Several other studies (32–36), however, have shown no association of multiple births with maternal breast cancer risk, and some have reported an increased risk (37, 38). Misclassification in exposure is unlikely to explain the differences in findings. Information on zygosity, chorionicity, and sex of twin pairs was unavailable in our study and others but may represent differences in maternal circulating hormones and other factors (39) and should be examined in future studies of breast cancer risk.

The inverse association between maternal prepregnancy size and breast cancer risk observed in our data is consistent with previous studies reporting a protective effect of weight on premenopausal but not postmenopausal breast cancer risk (22). Two Scandinavian registry-based studies evaluated the association of offspring birth weight and maternal breast cancer risk (20, 37). Birth weight in the last pregnancy was not associated with an increased risk (37), but a 500 g increase in birth weight in the first pregnancy was associated with an 11% increase (95% CI, 4%–18%) in risk after adjustment for gestational age in weeks and other possible confounders (19). The positive trend we observed with birth weight taking into account gestational age in pregnancy was more modest (8% increase per 1,000 g; 95% CI, 5%–11%) and only among women diagnosed 15+ years after the pregnancy. In contrast, in a subgroup analysis of cases diagnosed at older ages but within 15 years of the index pregnancy, we observed an increased risk associated with having a very low birth weight infant. Neither of the 2 other studies assessed risk associated with birth weights less than 3,000 g. The biologic mechanism explaining an increased breast cancer risk associated with a very low birth weight baby could be related to reasons for the fetus' small size that were independent of gestational age at delivery, such as fetal growth rate.

A possible limitation of our study was the accuracy of the data from the birth record. Although birth and hospital discharge data were each collected in a generally similar manner for all women according to standards set by the relevant agencies, there likely exist differences between hospitals and over time. Many validation studies report that birth weight, offspring sex, and multiple gestations are fairly accurately reported in birth records (40–43). Parity and demographic information are reliable compared with medical records (41) and Medicaid program data (41). There is high sensitivity for identifying low birth weight defined as <2,500 g (100% specificity) and preterm deliveries, or deliveries occurring before 37 weeks (93% specificity; ref. 41). Maternal prepregnancy weight information was supplied to the birth record from the women's medical charts, but in some instances may have been based on self-report, which may underestimate a women's true weight (44). Reporting of preeclampsia and diabetes (43, 45) is less reliable but improved when used in combination with hospital discharge data (45–47), such as we were able to do for much of the study period. Another limitation is that some exposure variables (maternal size, weight gain, and some pregnancy complications) were not available for many years of follow-up and when available were missing for some women. Restriction of analyses to years for which data existed in the registries resulted in decreased power. Otherwise, missing data may have biased risk estimates to the null if data availability was unrelated to later breast cancer diagnosis. We also lacked information on some recognized adult risk factors for breast cancer. However, the use of these population-based data minimized biases due to reporting or differential response, provided a large sample size, and allowed for greater generalizability of findings to the general population and to similar registry-linkage studies conducted in other regions.

Because controls were identified from birth records during the same years as the deliveries occurred for cases, it is possible that some controls may have moved out of state and subsequently been diagnosed with breast cancer elsewhere. If migration was unrelated to the pregnancy characteristics under study, this would potentially bias the ORs toward the null. In an earlier study in our state with similar data, restriction of analyses to subjects known to have renewed a state drivers' license within 5 years of diagnosis had little effect on results (13). Recent U.S. Census surveys report less than 3% of people who changed residences and moved out of state, and concern about health is rarely cited as the primary reason for moving (48).

Although previous data on multiple gestations are inconsistent, there seems to be enough suggestion of reduced risk that it should be explored further. Studies that include detailed information on zygosity or chorionicity of the pregnancy as well as offspring sex would be most informative. As there are few studies on placental abnormalities and bleeding and maternal breast cancer, the increased risk we observed also should be examined in additional studies with more detailed information. Because the associations for these complications were most pronounced in women who were 50 years or older at diagnosis, for which an early pregnancy would be protective, it would be useful to know if the observed associations are operating through inhibition of the protective effect of a first birth. This would have been difficult to assess in our study because of design, in which controls were matched to cases on maternal age and parity. Finally, if the association of placental abnormalities and bleeding are confirmed, exploration of the underlying biologic mechanisms is warranted.

No potential conflicts of interest were disclosed.

Conception and design: R. Troisi, B.A. Mueller

Development of methodology: R. Troisi, B.A. Mueller

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B.A. Mueller

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. Troisi, D.R. Doody, B.A. Mueller

Writing, review, and/or revision of the manuscript: R. Troisi, D.R. Doody, B.A. Mueller

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.R. Doody, B.A. Mueller

Study supervision: B.A. Mueller

The authors thank the Washington State Department of Health for data access, Mr. Bill O'Brien for programming assistance, and Dr. Robert N. Hoover for substantial contributions to the interpretation of data.

This research was funded by National Cancer Institute Contract #HHSN261201000725P (to R. Troisi, D.R. Doody, B.A. Mueller).

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.