Advancements in medicine and technology have led to an increasing number of cancer survivors. The development of a second primary cancer is one of the most severe sequelae of a cancer diagnosis, particularly for cancers that lack an effective screening tool as with ovarian cancer. Breast and ovarian cancer are major causes of morbidity and mortality in women; in the U.S., breast cancer has the highest incidence in women and ovarian cancer is the most fatal of gynecological cancers. Further, these two cancers have been found to co-occur. Along with possible treatment effects of the first cancer, shared risk factors, shared genetics, and interactions between these two have been hypothesized to contribute to their co-occurrence. Research on shared risk factors for second cancers is lacking and being able to identify potentially modifiable factors associated with second primary cancer could improve clinical recommendations for cancer survivors. Therefore, this dissertation examined risk factors for the development of double primary breast and ovarian cancer (DPBOC) in three parts 1) a comprehensive review of the literature to identify studies assessing risk factors for DPBOC, 2) a case-control study assessing the association between three potentially-modifiable risk factors (oral contraceptive (OC) use, parity, and breastfeeding), and risk of second primary ovarian cancer following breast cancer (BR-OV), second primary breast cancer following ovarian cancer (OV-BR), single primary ovarian cancer (OV), and single primary breast cancer (BR), and 3) a cohort study assessing OC use, parity, and breastfeeding and risk of BR-OV, OV, and BR. The comprehensive review identified few studies assessing epidemiologic risk factors for the development of DPBOC and most of the findings were not statistically significant. The majority of studies focused on treatment of breast cancer and risk of second primary ovarian cancer. While most of the findings on chemotherapy, radiotherapy, and Tamoxifen were heterogeneous and lacked statistical significance, hormone therapy for breast cancer may be associated with an increased risk of second primary ovarian cancer. The majority of studies on genetic risk factors for DPBOC looked at BRCA1/2 mutations or a crude measure of family history. Both BRCA1/2 and family history were consistently associated with risk of DPBOC, but studies varied on the extent of this risk due to differences in study design, exposure and outcome definition, and statistical power. No studies were identified examining DNA methylation and risk of DPBOC. The case-control study used data from the three clinic-based sites of the Breast Cancer Family Registry (BCFR) which consisted of women from breast and ovarian cancer families. We observed an inverse association with both OC use (OR=0.38, 95% CI: 0.22, 0.60) and breastfeeding (OR=0.52, 95% CI: 0.31, 0.87) and risk of DPBOC, but a positive association with parity (≥2 full-term pregnancies: OR=5.78, 95% CI: 2.82, 14.58), regardless of diagnosis order (BR-OV or OV-BR). We found similar associations for our OV and BR outcomes as well. When we examined differences between high and average risk women (using BRCA1/2 mutation status and predicted lifetime risk of breast or ovarian cancer), the inverse association with OC use only remained in women at average risk while the inverse association with breastfeeding only remained in women at high risk. As the positive association with parity and all of our outcomes disagreed with our hypothesis we conducted several sensitivity analyses to explore this finding. Survivor bias may have influenced our results as we observed differences in our findings between cases diagnosed ≤2 or ≤5 years before the baseline interview (pseudo-incident) and cases diagnosed >2 or >5 years before the baseline interview (prevalent). Specifically, the inverse association with OC use and all of our outcomes, and the positive association with parity and all of our outcomes were attenuated in the pseudo-incident group. To address concerns of selection and information bias in our case-control study, we conducted a cohort study using data from The Breast Cancer Prospective Family Study Cohort (ProF-SC). In contrast to our case-control findings, we observed a suggestive positive association between OC use and risk of BR-OV (HR=1.62, 95% CI: 0.91, 2.90) which became stronger in women at high risk, and an inverse association between having two or more full-term pregnancies compared to nulliparous and risk of BR-OV (HR=0.47, 95% CI: 0.22, 0.97) which did not vary by underlying risk of breast and ovarian cancer. However, our BR-OV results may have similarly been influenced by survivor bias as we observed differences in our results between our pseudo-incident and prevalent BR-OV cases; the association between OC use and BR-OV only remained in the prevalent cases. In summary, the results of this dissertation highlight the methodological challenges in the study of second primary cancers and the importance of considering survivor bias in a cohort of cancer survivors being followed for second cancers. Further, our results are suggestive of a discordant effect of OC use on first primary versus second primary ovarian cancer which should be explored in future studies.

Ovarian cancer is the fifth most common cause of cancer death in females and the most lethal gynecologic cancer. Globally, an estimated 240,000 people are diagnosed with ovarian cancer each year, with 22,530 new cases in the United States in 2019. Parity, oral contraceptive use, and lactation are protective, while early menarche, late menopause, and nulliparity have opposite effects. The “incessant ovulation” theory has thus emerged, in which a higher number of ovulations may be a cause of epithelial ovarian cancer (EOC). However, the mechanisms of this theory are unknown; one possibility is that the chance of acquiring a cancer-initiating pathogenic variant increases with each ovulatory cycle because of a microenvironment that promotes DNA damage. In this dissertation, we aimed to leverage genetic epidemiologic data to test this potential mechanism by evaluating the presence of gene-environment interaction between DNA repair capacity (measured through the presence of pathogenic variants in DNA repair genes) and lifetime ovulatory years (LOY). In the first aim of this dissertation, we conducted a systematic review and meta-analysis, following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, to formally evaluate the strength of evidence and to generate summary point estimates for the association between LOY and EOC. We then executed two analytic aims to evaluate if the presence of pathogenic variants in DNA repair genes exacerbated the increase in ovarian cancer risk associated with LOY. In Aim 2, we evaluated interaction on the additive scale in the United Kingdom (UK) Biobank through use of a novel DNA repair capacity score developed in this dissertation, measured by quantifying the number of pathogenic variants present per individual from a list of 163 DNA repair genes, using whole exome sequencing (WES) data. In Aim 3, we evaluated the presence of interaction between pathogenic BRCA1/2 status and LOY in the Breast Cancer Family Registry (BCFR), a cohort enriched for familial risk. In both empirical aims, we assessed the presence of interaction on the additive scale using the relative excess risk due to interaction (RERI) formula. We compared results across the two empirical aims. We found the relationship between lifetime ovulations and ovarian cancer risk to be consistent and replicable in the published literature. In pooled estimates from 22 published studies, a one-year increase in LOYs was associated with a 4% (3-6%) increased risk of ovarian cancer and those with a high number of ovulations (compared to low LOYs) had a 2.15-fold (95% CI 1.82, 2.54) increased risk of ovarian cancer. We also confirmed the positive association between increasing LOYs and ovarian cancer risk in the UK Biobank and the BCFR cohorts. Although interaction on the additive scale was not detected, there were strong positive associations between pathogenic variants in DNA repair genes and ovarian cancer risk. In the UK Biobank, the presence of at least one pathogenic variant in a DNA repair gene was associated with a significant 27% increased risk of epithelial ovarian cancer (EOC) (95% CI 5-55%). Among women at high risk of ovarian cancer due to family history of breast and/or ovarian cancer, there was a strong relationship between BRCA1/2 pathogenic variants and ovarian cancer, regardless of the number of ovulations experienced. The association between LOY and ovarian cancer was found to be consistent and replicable, despite differences in study design, covariates, and measurement. We also detected robust evidence that increasing lifetime ovulations and pathogenic DNA repair variants were associated with ovarian cancer risk. Such variants were exceedingly rare in both cohorts, which limited power to detect interaction in an already rare cancer. Despite such associations, there was no evidence of synergy between LOY and impaired DNA repair capacity, but rather, high LOY and impaired DNA repair capacity may be independent risk factors of ovarian cancer. Each exposure may describe a separate class of women at increased risk of ovarian cancer that should be targeted for future prevention and screening strategies.