Examining the etiology of early-onset breast cancer in the Canadian Partnership for Tomorrow’s Health (CanPath)

Breast cancer incidence among younger women (under age 50) has increased over the past 25 years, yet little is known about the etiology among this age group. The objective of this study was to investigate relationships between modifiable and non-modifiable risk factors and early-onset breast cancer among three prospective Canadian cohorts. A matched case–control study was conducted using data from Alberta’s Tomorrow Project, BC Generations Project, and the Ontario Health Study. Participants diagnosed with breast cancer before age 50 were identified through provincial registries and matched to three control participants of similar age and follow-up. Conditional logistic regression was used to examine the association between factors and risk of early-onset breast cancer. In total, 609 cases and 1,827 controls were included. A body mass index ≥ 30 kg/m2 was associated with a lower risk of early-onset breast cancer (OR 0.65; 95% CI 0.47–0.90), while a waist circumference ≥ 88 cm was associated with an increased risk (OR 1.58; 95% CI 1.18–2.11). A reduced risk was found for women with ≥ 2 pregnancies (OR 0.76; 95% CI 0.59–0.99) and a first-degree family history of breast cancer was associated with an increased risk (OR 1.95; 95% CI 1.47–2.57). In this study, measures of adiposity, pregnancy history, and familial history of breast cancer are important risk factors for early-onset breast cancer. Evidence was insufficient to conclude if smoking, alcohol intake, fruit and vegetable consumption, and physical activity are meaningful risk factors. The results of this study could inform targeted primary and secondary prevention for early-onset breast cancer.


Introduction
Breast cancer is the most commonly diagnosed cancer globally among women with over two million new cases in 2018 [1]. Canada is among the top 25 countries worldwide with the highest rates of breast cancer [1]. A projected 1 3 27,400 breast cancer cases are expected to be diagnosed in Canada in 2020 [2]. Breast cancer incidence among younger women (under age 50) has increased in Canada since 2000 [3]. The Canadian Cancer Society (CCS) projected that in 2017, roughly 17% of new breast cancer cases will occur before age 50 [4]. Breast cancer diagnosed among younger women often occur outside age-restricted or high-risk screening programs, resulting in advanced stage at diagnosis and poorer survival rates [5,6]. The diagnosis of early-onset breast cancer also presents several unique challenges, including fertility concerns and reduced quality of life due to treatment-induced premature menopause [5,6]. While mutations in the BRCA1 and BRCA2 genes increase the risk of developing breast cancer, only a small number of breast cancers (~ 5 to 10%) diagnosed among young women are attributable to these mutations [7,8], suggesting other genetic, environmental or lifestyle factors may contribute to the development of early-onset breast cancer.
Several risk factors for breast cancer have been identified, with the majority of these risk factors applying to both pre and postmenopausal breast cancer [9]. While these risk factors for old and younger breast cancer cases are similar, there is evidence of early-onset breast cancer being biologically and etiologically distinct from breast cancer diagnosed among woman of older ages and further research is required [10]. Specifically, the relationship between anthropometric factors and early-onset breast cancer risk requires clarification, with previously reported inverse associations and a conflicted evidence base [11]. For example, current evidence suggests obesity is a risk factor for postmenopausal breast cancer [12,13], while being protective for premenopausal breast cancer [14]. Furthermore, the mechanisms explaining the underlying pathways in these associations remain unclear. However, present studies on early-onset breast cancer and associated risk factors have been limited or conflicting [10,15,16]. Studies have attempted to resolve previous conflicted evidence but have been impacted by smaller comparative sample sizes among younger women and varied definitions of "young women" across studies [10,15,16].
The objective of our study was to examine the impact of lifestyle and reproductive factors, as well as family and medical history on the risk of developing early-onset breast cancer (under the age of 50) using data from The Canadian Partnership for Tomorrow's Health (CanPath) [17]. The CanPath cohort is a national prospective cohort study that was developed to explore the relationships between various lifestyle, genetic, and environmental factors and outcomes of disease. The national study population includes over 330,000 Canadians who were recruited between the ages of 30 and 74 years from six regional cohorts. We combined three of the six cohorts involved in CanPath: Alberta's Tomorrow Project (ATP); British Columbia Generations Project (BCGP); and the Ontario Health Study (OHS) to investigate various risk factors for early-onset breast cancer.

Study population
In this study, we pooled baseline data from 2,437 participants from the ATP, OHS and BCGP regional cohorts. ATP was initiated in 2000 and enrolled approximately 55,000 participants between the ages of 35-69 at three different time points-2001 (phase 1), 2008 (phase 2), and 2011 (phase 3). Data from phases 2 and 3 from ATP were used in this analysis since the questionnaires from those phases were harmonized with the other cohorts in CanPath. OHS has recruited over 225,000 Ontario residents in the age range of 35-74 starting in 2009, while BCGP has enrolled approximately 30,000 British Columbia residents between the ages of 30 and 74 starting in 2009.

Study design
A nested matched case-control study design was used for this study. Cases and controls were selected from the same regional cohort in a 1:3 case-control ratio. Cases were defined as female participants who had an incident, primary breast cancer diagnosis under 50 years of age during the follow-up period. The incident cases were identified through data linkages done with the BC Cancer Registry, Alberta Cancer Registry and Cancer Care Ontario, respectively, using participants' Personal Health Identification Numbers. Using an incidence density sampling approach, controls were sampled without replacement and matched to cases on follow-up (± 6 months) and age (± 1 year). The follow-up period for each cohort was defined from cohort initiation to the most recent linkage with their respective provincial cancer registry, which was 2008-2018 for Alberta, 2009-2017 for British Columbia, and 2009-2017 for Ontario.

Data Collection-Questionnaires
Data on demographic, lifestyle, reproductive, and family and medical history factors under investigation as exposures in this study were collected through self-report questionnaires completed by participants within each cohort at baseline upon cohort entry, prior to diagnoses. The exposures were analyzed as categorical variables. Thresholds for exposure categories were defined prior to analysis and informed by previous literature. Where necessary, categories were combined to ensure sufficient case counts across categories.
The family and medical history exposures included participation in mammography screening (no, yes); family history of breast cancer and any cancer (no, yes); history of any diabetes (no, yes), including type 1, type 2, and gestational diabetes; and history of other chronic conditions (no, yes), including irritable bowel disease/syndrome, Crohn's disease, arthritis, high blood pressure, psoriasis, ulcerative colitis, liver cirrhosis, and systemic lupus erythematosus.

Statistical analysis
Participants from each provincial cohort were pooled to generate one combined cohort. All demographic, lifestyle, reproductive, and family and medical history exposures were described using means and standard deviations (SD) for numeric variables and frequency tables with proportions for categorical variables for the combined cohort. Multivariable conditional logistic regression models were used to evaluate the association between each exposure and the development of early-onset breast cancer. For each exposure, a multivariable model was built using a priori subject matter knowledge. We performed a review of the literature to culminate information regarding demographic, anthropometric, lifestyle, reproductive, family history, and medical history risk factors for breast cancer. For every exposure variable in our study, we generated a list of variables in our data that may potentially confound the association between the exposure of interest and early-onset breast cancer. We defined a confounding variable as a variable associated with the exposure of interest and early-onset breast cancer and not on the causal pathway between exposure and early-onset breast cancer. A detailed table describing the model building procedure and reasons for the inclusion/exclusion of variables for every multivariate model used to estimate the associations of interest is provided in Supplementary File 1. We also built multivariable models using a backwards deletion approach with a p value cut-off of 0.20. Variables included in each model following backwards deletion are provided in Supplemental File 2. In lieu of missing data, we performed complete case analyses.

Results
A total of 610 cases and 1,827 controls were eligible for analysis; however, 1 case was not matched to 3 controls and was excluded as a result. Therefore, the pooled analysis included 609 cases and 1,827 controls for a total sample size of 2,436 participants across these three regional Canadian cohorts of which 14.8% (n = 360) were from Alberta, 5.7% (n = 140) from British Columbia, and 79.5% (n = 1936) from Ontario ( Table 1). The mean age of all participants at baseline was 43.1 (SD = 5.1) years. The study population had a high annual average annual income (41.2% ≥ $100,000), were well-educated (44.5% had post-secondary education), and married/living with a partner (71.7%). Most of the study participants had at least one pregnancy (14.7% had one pregnancy, 63.6% had two or more), and the mean age of first pregnancy among these women was 26.1 years. In addition, these participants also reported using hormonal contraceptives (86.2%) and the mean duration of hormonal contraceptive use was 10.7 years. Only 11% of women in this study had a positive family history of breast cancer. Majority of women were absent of any diabetes (93.3%) and other chronic conditions (64.6%). Over one-third of the population was either classified as overweight (21.1%) or obese (17%) at baseline. With respect to lifestyle factors, most of study population consumed alcohol (89.6%), 17.8% were current smokers, 22.5% were past smokers, and 56.4% had never smoked. The frequency of missing data in each characteristic is also presented in Table 1.
The estimated adjusted associations between each exposure and the incidence of early-onset breast cancer in the combined cohort are presented in Table 2. There was evidence that early-onset breast cancer cases were less likely to have an average annual household income < $50,000 vs. ≥ $100,000 (OR 0.58; 95% CI 0.42-0.79), to be classified as overweight vs. underweight or normal BMI (OR 0.75; 95% CI 0.57-0.98), obese vs. underweight or normal BMI (OR 0.65; 95% CI 0.47-0.90), and to have had at

Discussion
To our knowledge, this pooled analysis is the first to investigate the etiology of early-onset breast cancer in a large Canadian population. For breast cancer under 50 years, we observed risk decreases of 25%, 35%, 42% and 24% in overweight BMI, obese BMI, low income, and multiple pregnancy categories, respectively. Conversely, risk increases of 58%, 41%, 39%, and 95% were observed for larger waist circumference, larger waist-to-hip ratio, lower educational attainment, and a positive family history of breast cancer categories, respectively. It is well established that weight gain and obesity increase breast risk in postmenopausal women [13]. Consistent positive associations have been reported between BMI, waist circumference, and waist-to-hip ratio and the risk of breast cancer among postmenopausal women [18,19] but results are inconsistent for younger women. Our study suggests that an overweight and obese BMI are associated with lower early-onset breast cancer risk. Similar results were seen in a pooled analysis of 19 prospective studies including women under 55 years of age [11]. Our results also suggest that high waist circumference and waist-to-hip ratio are predictors of breast cancer in younger women, independent of BMI, consistent with other prospective studies adjusting for BMI in premenopausal women [19][20][21][22][23]. The opposing effect of BMI and waist circumference on breast cancer risk in young women seen in our study and elsewhere requires further clarification. Possible differences in these associations could be related to differential roles of overall adiposity, captured by BMI, and central adiposity, captured by waist circumference, on metabolism and their contribution to breast cancer development among younger women. Both overall and central adiposity are associated with more anovulatory cycles and  lower estradiol levels in premenopausal women [24], which would be expected to reduce breast cancer risk. However, central adiposity is also an independent predictor of both hyperinsulinemia and levels of insulin-like growth factor 1 (IGF-1), which have been previously found to be related to premenopausal breast cancer risk [18,25]. These findings suggest that chronic inflammation and metabolic abnormalities induced by central adiposity are mechanistically important for higher breast cancer risk in young women, independent of sex hormones, which may explain the observed increase in risk for high waist circumference and waist-to-hip ratio in our study. These findings may also suggest that, in the absence of central obesity and its metabolic abnormalities, lower sex hormones are protective against breast cancer development in younger women, which could explain why young women with higher BMI are observed to Oral contraceptive duration-income, education, age at menarche, total number of pregnancies, fertility treatment, family history of breast cancer Fertility treatment-income, education, family history of breast cancer, history of chronic disease Physical activity-BMI, alcohol frequency in past 12 months, smoking status, income, education, total number of pregnancies, fruit and vegetable consumption, family history of breast cancer, history of chronic disease Fruit and vegetable consumption-BMI, alcohol frequency in past 12 months, smoking status, income, education, total number of pregnancies, physical activity, family history of breast cancer, history of chronic disease Family history of breast cancer-no adjustments Family history of any cancer-no adjustments History of chronic disease-BMI, waist circumference, alcohol frequency in past 12 months, smoking status, income, education, physical activity, fruit and vegetable consumption, family history of breast cancer BMI body mass index, CI confidence interval, METs metabolic equivalent tasks (defined as the caloric need per kilogram of body weight per hour of activity divided by the caloric need per kilogram of body weight per hour at rest), OR odds ratio a Includes occasional and daily smokers as current smokers be at lower breast cancer risk. Another possible explanation for the opposing effect of BMI and waist circumference on breast cancer risk in young women is estrogen/progesterone (ER/PR) tumor status. High BMI has been observed to be protective in ER/PR+ cancers but not in ER/PR-cancers, and higher waist circumference has been observed to increase risk of ER/PR-cancers but not ER/PR+ cancers [13]. The distribution of ER/PR+ and ER/PR− tumors in young women in this study may account for these opposing effects. Unfortunately, we did not have access to the tumor characteristic data to confirm whether the effect of BMI and waist circumference depended on receptor status or subtype. A positive family history is an important risk factor for breast cancer and the strength of association increases as age of diagnosis decreases. In a pooled analysis of data from 52 epidemiological studies, including 58,209 women with breast cancer and 101,986 controls, the risk of breast cancer was two times greater in women under 50 years of age with at least one affected first-degree relative, whereas a 1.5-fold increase in risk was observed in women older than 50 years [26]. We observed an approximate two-fold risk increase in women under age 50 years with a family history of breast cancer in the Canadian population. This cumulative evidence indicates a hereditary factor for breast cancer in young women. Several germline mutations elevate breast cancer risk in young women, including the BRCA1, BRCA2, TP53 (Li Fraumeni syndrome), and PTEN (Cowden's syndrome) mutations [27]. However, many young women with a strong family history of breast cancer do not carry these mutations and the genetic basis of breast cancer in young women requires further exploration [7,28,29]. The CCS and Canadian Task Force on Preventive Health Care (CTFPHC) recommend women with a strong family history consult with their primary care physician for personalized testing plans, which may include mammography screening at an earlier age, more frequent mammography screening, or use of an ultrasound or magnetic resonance imaging (MRI) [30,31]. The National Comprehensive Cancer Network (NCCN) guidelines recommend women at an increased risk (≥ 20% lifetime risk based on models largely dependent on family history) undergo annual MRI screening starting as early as age 25 and annual mammography screening as early as age 30 [32]. This study provides additional evidence to support that young women with a positive family history of breast cancer should continue to seek counseling from their primary care provider for personalized screening strategies for early detection of breast cancer before 50 years of age. Routine mammography screening is not recommended for women aged 40-49 years in Canada as trials have shown that the number needed to screen to prevent one death from breast cancer is 1,724 [33]; therefore, more efforts should be directed toward detection of high-risk young women based on familial and genetic factors.
The only reproductive risk factor that was statistically significant in this study was gravidity, in which a 24% reduction in breast cancer risk was observed for women with multiple pregnancies compared to nulliparous women. Recently, a 27% reduction in breast cancer risk was observed in a meta-analysis of 13 studies in women under the age 50 with multiple pregnancies compared to nulliparous women [34]. However, the age at first full term pregnancy influences the association between gravidity and early-onset breast cancer risk, which was not statistically significant in our models. In large prospective studies, it has been shown that women whose first full term pregnancy is in their thirties remained at an increased risk of breast cancer well into their fifties relative to nulliparous women [35][36][37]. Conversely, women whose first full term pregnancy is in their twenties were at an increased risk that peaks five years postpartum then declines to null 10-15 years postpartum and becomes protective around 20 years postpartum [35,37]. Several etiologic studies explored the underlying biological processes of this association. Russo et al. found that parous postmenopausal women without breast cancer had unique gene expression patterns including differential expression of apoptosis-related genes and others related to cell cycle and cell signaling, suggesting that pregnancy may induce a signature that provides long-term protection from developing breast cancer [38]. Another study showed that parity downregulates Wnt and Notch signaling, and suppresses progenitor cells, suggesting that this pathway could be a potential mechanism explaining the long-term protective effect of pregnancy [39]. The transient increase in the risk of breast cancer after giving birth is thought to be due to the mitogenic effect of high estrogen levels during pregnancy [40]. While our result indicates a protective effect from multiple pregnancies, no recommendations regarding risk reduction can be made given that the decision to have children is highly personal.
Most of the current evidence has found that higher socioeconomic status is linked to increased risk of breast cancer, specifically among high income countries in North America and Europe [41][42][43][44][45][46]. This association is mainly seen among postmenopausal breast cancer cases [42,47]. Our study found evidence that the odds of early-onset breast cancer were significantly lower for women with lower average annual household income. This finding is consistent with several studies that found women with higher household incomes had an increased risk of breast cancer compared to women in lower income households [48,49]. In contrast, our study found evidence of low education (high school degree or less) associated with increased risk of early-onset breast cancer. The majority of the findings on breast cancer and education suggest that higher education is associated with increased breast cancer risk in Western or high-income countries [46,50,51]. Correlations of higher education and breast cancer incidence in these studies were explained by factors such as age at first pregnancy, oral contraceptive (OC) use and hormone replacement therapy use. However, results from a Dutch study found respondents with low education had an increased risk of all cancers including breast cancer among women [52]. The study suggested that factors such as alcohol consumption, obesity and poor diet may explain, in part, the association between breast cancer and low education [52][53][54][55]. Another possibility for our findings of low education and increased breast cancer risk may be related to chronic inflammation and the biomarker C-reactive protein (CRP), a protein produced by the liver in response to systemic inflammation [56,57]. Several studies found elevated CRP levels among populations with lower education compared to higher levels of education attainment [56,58,59]. Relatedly, multiple systematic reviews have found chronic inflammation and CRP to be associated with increased breast cancer risk [56,58,60,61]. Our study provides additional evidence on the potential life-course pathways of the impacts of socioeconomic status and early-onset breast cancer.

Strengths and Limitations
Our study has numerous strengths worth highlighting. We pooled data from a national prospective study from three regional cohorts that were each linked to provincial cancer registries. Each cohort collected detailed health and lifestyle questionnaire data at baseline prior to a breast cancer diagnosis, ensuring that each risk factor temporally preceded the onset of disease. In addition, the prospective nature of the data ensured that the results of this study were not subject to recall or selection bias.
There are several limitations of this pooled analysis that must be acknowledged. First, data from the health and lifestyle data are from self-reported study participant questionnaires. While the questionnaires were deemed reliable and valid [62], data from self-reported questionnaires are inherently subject to response biases such as social desirability bias. Secondly, we only had data from one time point and assumed that these data represented average exposure during young adulthood. While differential misclassification of exposure by case status is unlikely since questionnaires were answered prior to the development of breast cancer, non-differential misclassification of exposures is likely. For binary exposures this would have biased effect estimates toward the null and either toward or away from the null for exposures with three or more categories. The sample size was also insufficient to explore whether or not the association between gravidity and breast cancer risk was modified by age of first pregnancy. There were also concerns regarding missing data for several exposures, particularly BMI, waist circumference, and waist-to-hip ratio. Although missingness in exposures was unlikely related to case status and therefore would not bias our estimates, exclusion of participants with incomplete exposure resulted in losses of precision and power. Finally, we did not include race/ethnicity data for participants and could not assess whether the associations between the various exposures and early-onset breast cancer differed by racial group.

Conclusion
Our study provides further insight on sociodemographic, lifestyle, and reproductive factors associated with the risk of breast cancer under 50 years in a large Canadian prospective cohort, which may have important implications for etiology and screening. We provide evidence consistent with other epidemiologic studies that central adiposity increases breast cancer risk in women under 50 years. Lifestyle behaviors, such as a healthy diet and physical activity that reduce central adiposity may be feasible primary prevention strategies to reduce breast cancer risk, in addition to other chronic diseases in younger women. We also provide further evidence that women under 50 years with a positive family history of breast cancer should speak with their primary care provider about options for breast cancer screening, which is in accordance to screening guidelines to high-risk women set by the Canadian Cancer Society and Canadian Task Force on Preventive Health Care. Future research is still needed to explore whether or not risk factors for breast cancer under age 50 differ by molecular subtype (luminal A, luminal B, HER2 enriched, and triple-negative breast cancer). Detailed analyses are also needed to assess differences in risk factors by race and socioeconomic status, since our study only had data on income and education, but not occupation and other measures of social inequality. Etiologic research, such as exome-sequencing studies, is warranted to explore the mutational impact of lifestyle behaviors on the genome and yield discoveries regarding the underlying biological processes of breast cancer development in younger women. Such studies may provide mechanistic insights on the opposing effect of high BMI and waist circumference on breast cancer risk, as well as how pregnancy may lead to a transient increase in risk, followed by a long-term protective effect.