Inherited factors contribute to an inverse association between preeclampsia and breast cancer

This study included two cohorts: (1) a Swedish nationwide cohort of pregnant women (nationwide cohort in short) and (2) a Swedish mammographic screening-based cohort (Karolinska Mammography project for risk prediction of breast cancer, KARMA) (Fig. 1).

Data on the nationwide cohort was retrieved from the Swedish Medical Birth Register, which contains data on more than 99% of all births [15], and includes all women who delivered their first child between 1973 and 2005 (N = 1,337,934). Pregnancy characteristics including height, pre-pregnancy weight, smoking status, and previous reproductive history were collected at the first antenatal visit (at approximately 8–12 weeks of gestation). Education level was collected from the Swedish Register of Education. Information on sisters of the women was obtained by linking the cohort to the Swedish Multi-Generation Register. Maternal age and diseases related to pregnancy are reported by clinicians on post-delivery hospital discharge. These diseases are registered according to the Swedish version of the International Classification of Diseases (ICD), and for preeclampsia is coded as follows; ICD-10 (1997 to the present): O14 and O15; ICD-9 (1987-1996): 642E, 642F and 642G; and ICD-8 (1969-1986): 63703, 63704, 63709, 63710 and 63799.

The KARMA cohort included 70,877 women attending mammography screening or clinical mammography at one of four hospitals in Sweden between 2011 and 2013 [16]. Apart from mammographic imaging and blood sample collection, the participants answered a web-based questionnaire covering demographic, anthropometric, reproductive, and lifestyle risk factors related to breast cancer (selected basic information in Table 1). Information about preeclampsia diagnosis was also sought in the questionnaire. We linked the KARMA cohort to the Swedish Multi-Generation Register to obtain information on sister relationships and subsequently identified sisters of the patients with preeclampsia. For this study, we only included women who delivered their first child after 1958 (considering the start of the cancer register), and completed the full questionnaire (N = 55,044).

Follow up of the nationwide cohort started from the date of birth of the first child (see above), and ended on the date of first breast cancer diagnosis, date of death, date of emigration or end of follow up (31 December 2011), whichever came first. Information on breast cancer diagnosis, death and emigration was obtained through cross-linking the cohort to the Swedish Cancer Register, Swedish Causes of Death Register and the Swedish Migration Register, using unique Swedish personal identification numbers [17]. Breast cancer diagnosis was based on ICD-7 code 170 in the cancer register. The Swedish Cancer register started from 1958 and is considered to have almost 100% completeness [18, 19]. Follow up of the KARMA cohort also started from the date of birth of the first child, and ended the same time as the nationwide cohort, except for an extension of the follow up until 28 February 2015. The study was approved by the Regional Ethical Review Board in Stockholm, Sweden.

Mammograms in the mediolateral oblique (MLO) position were obtained from four Swedish hospitals in women participating in the KARMA cohort during 2011–2013. The fully automated software STRATUS was used to measure area-based mammographic density (details of this method have been described elsewhere) [20]. STRATUS measures mammographic density regardless of vendor of mammography machine and thus ensures comparability of mammographic density at the population level. The percentage density was calculated by dividing the dense area by the total breast area in the mammogram. Women were excluded from the mammographic density analysis if they had any previous cancers, or any breast enlargement or reducing surgery, leaving 43,844 women available for this analysis.

Blood samples from a subset of 9263 women without breast cancer from the KARMA cohort were genotyped using a custom Illumina iSelect array (iCOGS), comprising 211,155 single nucleotide polymorphisms (SNPs) [21], or an Illumina Infinium OncoArray, comprising 499,170 SNPs [22]. Details of the array design, sample handling and quality control processes are described elsewhere [21, 22]. To assess genetic predisposition to breast cancer, we selected 171 genome-wide significant SNPs reported in a recent meta-analysis of breast cancer genome-wide association studies (GWAS) for constructing a PRS [22]. These SNPs were imputed using the 1000 Genomes Project March 2012 release as a reference [23] and passed quality control. For each individual, a weighted PRS was calculated using the following formula:

where β is the per-allele log odds ratio (OR) of breast cancer associated risk allele for SNP _k , x _k is the number of alleles for the same SNP (0, 1, 2), and _n is the total number of the disease SNPs included in the profile. The SNPs and corresponding log ORs (weights) used for the derivation of PRS are summarized in Additional file 1: Table S1. For analysis, women were categorized in the following percentiles of breast cancer risk based on PRS: 0–40%, 40–60%, 60–80%, 80–90% and 90–100%.

An age-adjusted incidence rate of breast cancer was calculated in both the nationwide and KARMA cohorts, taking the 1990 Swedish national census population as the standard population. Considering the few cases of breast cancer in the age category 70–80 years in the KARMA cohort (n = 5), the age-adjusted incidence rate was restricted to an age band of 20–70 years. Poisson regression models were used to calculate incidence rate ratios for breast cancer in patients with preeclampsia. In this analysis, preeclampsia was considered as a time varying exposure, in which the exposed person-time was counted from the time of preeclampsia diagnosis. The underlying time scale was attained age. We constructed two models to analyze the association between preeclampsia and breast cancer incidence: (1) a basic model (model 1) adjusted for calendar period (10-year categories) and (2) model 2: with additional adjustment for number of births (time varying covariate), age at first birth, weight status categories (based on World Health Organization (WHO) body mass index (BMI) cutoff points (underweight (<18.5), healthy weight (18.5–24.9), overweight (25.0–29.9), obese (≥ 30.0)), smoking status and education level. In the analysis of the KARMA cohort, model 2 was additionally adjusted for alcohol use, age at menarche, physical activity at age 18 years, body shape at age 18 years (detailed information on body shape categories has been described elsewhere and shown in Additional file 1: Figure S1) [24], and irregular menstrual cycles in adult life. We also conducted two additional analyses to further adjust for breast cancer PRS on the basis of model 2, and to separately evaluate the risk of estrogen receptor positive (ER+) and negative (ER-) breast cancer in the KARMA cohort.

To identify the genetic association between preeclampsia and breast cancer, we used logistic regression models to estimate the ORs of preeclampsia (as an outcome) among cancer-free sisters of the patients with breast cancer, compared to women in the nationwide cohort without history of breast cancer and without a sister with history of breast cancer, adjusting for number of births. We also calculated the OR of preeclampsia by percentiles of breast cancer PRS for women in the KARMA cohort who did not have breast cancer and were genotyped, adjusting for number of births and batch effect of genotyping. For both of the analyses, we additionally adjusted for age at first birth, weight status categories, smoking status and education level in model 2.

As mammographic density is widely considered to be an intermediate phenotype of breast cancer, we tested the association between percentage mammographic density and previous diagnosis of preeclampsia in cancer-free women in KARMA. For this analysis, linear regression models with robust “sandwich” standard errors for confidence intervals were used, to avoid assuming normally distributed error terms and homoscedastic variance of the outcomes. We only adjusted for age at mammogram (continuous) in model 1, and additionally for BMI categories, age at menarche, number of births, age at first birth, menopausal status at mammogram, irregular menstrual cycle, physical activity at age 18 years, body shape at age 18 years, education level, smoking status and alcohol consumption in model 2. In order to test the genetic association between preeclampsia and mammographic density, we selected the cancer-free women in KARMA who have a sister in the cohort (N = 3500) and investigated the differences of mammographic density in sisters of patients with preeclampsia, compared to women without a sister with preeclampsia.

Statistical analyses were performed using SAS (version 9.4; SAS Institute Inc, Cary, NC, USA) and Stata software (version 14.0; Stata Corporation, College Station, TX, USA), at a two-tailed alpha level of 0.05.