{"paper_id":"bb72899e-e7bc-48d6-a984-6b63e6cfb7cf","body_text":"Women with an inherited germline pathogenic variant in  BRCA1  or  BRCA2  ( BRCA  “carriers”) have a greatly increased risk of developing ovarian cancer, with an estimated average lifetime risk of between 45% and 50% for  BRCA 1 variant carriers and 17% and 21% for  BRCA 2 carriers, compared with less than 3% for the general population ( 1 – 3 ). Despite this greatly increased risk, the incomplete penetrance suggests that other factors are also important. Factors associated with an increased risk of developing ovarian cancer in the general population include endometriosis, body mass index (BMI) and smoking, while protective factors include parity, oral contraceptive pill (OCP) use and tubal ligation ( 4 ).\nDespite a growing body of research, there is little evidence about whether or how epidemiologic risk factors vary between  BRCA  carriers and noncarriers. Most studies have used a case–control design to evaluate risk factors among  BRCA  variant carriers, and have focused mainly on reproductive or hormonal risk factors, with little consideration of modifiable lifestyle factors ( 5 ). We have previously reported that the association between smoking and poor survival after a diagnosis of ovarian cancer was stronger among  BRCA  carriers than women without a pathogenic  BRCA  variant (“noncarriers”; ref.  6 ).  BRCA  variant carriers most commonly develop high-grade serous carcinoma (HGSC; ref.  7 ) which is not associated with smoking in the general population. Conversely, mucinous ovarian carcinoma, which is associated with smoking, rarely occurs in  BRCA  carriers ( 8 ). Only one small study ( n  = 109) has evaluated the relation between smoking and ovarian cancer risk in  BRCA  carriers; they reported that former smokers had a 69% increased risk of developing ovarian cancer compared with never smokers, with a significant dose–response relationship between smoking duration and risk ( 9 ).\nChronic inflammation, which has been associated with the development of many cancers, including ovarian cancer ( 10 – 12 ), is associated with increased DNA damage ( 10 ). Pathogenic  BRCA  variants cause homologous recombination deficiency and thus impaired DNA repair ( 13 ). Therefore, it is possible that  BRCA  carriers may be more susceptible to the adverse effects of inflammatory risk factors such as smoking.\nOur aim was to evaluate whether established risk factors for ovarian cancer differ between  BRCA  variant carriers and noncarriers. In particular, we hypothesized that inflammatory risk factors, particularly smoking, that are weakly associated with ovarian cancer risk overall, may have a stronger association among  BRCA  carriers than noncarriers because inflammation can lead to DNA damage ( 10 ). We also investigated differences by histotype to ensure any differences were not due to histologic differences between carriers and noncarriers. A better understanding of variation in risk factors by  BRCA  variant status may lead to improved targeted prevention strategies for this high risk group.\n\nWe pooled data from three case–control and two case-only studies participating in the Ovarian Cancer Association Consortium (OCAC;  Supplementary Table S1 ; ref.  14 ). Two studies were from Australia [Australian Ovarian Cancer Study (AUS; ref.  15 ), and Ovarian cancer Prognosis And Lifestyle study (OPL; ref.  16 )], two from the United States [the Mayo Clinic Case-Only Ovarian Cancer Study (MAC) and Mayo Clinic Ovarian Cancer Case Control Study (MAY; refs.  17 ,  18 ), combined as MAY for these analyses], and one from the United Kingdom [UK Studies of Epidemiology and Risk Factors in Cancer Heredity Ovarian Cancer Study (SEA; ref.  19 )]. Cases were identified both prospectively (AOCS, MAC, MAY, OPL, SEA) and retrospectively (SEA). Data on exposures, confounders, and clinical information were provided to OCAC by the participating sites, and centrally harmonized. Cases had pathogenic germline  BRCA  testing either as part of clinical care or through their participation in these studies.  BRCA  status was not available for controls, so we employed a case-only approach to provide direct estimates of the gene–environment interaction.\nThese five studies included 8,027 women diagnosed with invasive epithelial ovarian cancer between 1991 and 2019. We excluded 1,430 women from SEA diagnosed between 2011 and 2019 as  BRCA  status was not available for any of this group; this left 6,597 eligible for inclusion in this analysis. We additionally excluded women missing information on their germline  BRCA  status ( n  = 1,434, 22%) or who had a mutation of uncertain significance (as defined by individual studies;  n  = 237) leaving 4,926 women for analysis. The most common reason for missing  BRCA  status ( n  = 382, 23% of those missing) was that women had low-grade serous or mucinous cancers that are rarely associated with  BRCA  mutation. All studies had ethics approval from their relevant Institutional Review Board and all study participants provided informed consent.\nWe considered the following epidemiologic risk factors: age at menarche (years); full-term pregnancy (ever, never); parity (0, 1, 2, 3, 4þ); breastfeeding (yes, no—parous women only); OCP use (ever, never); duration of OCP use (<5, 5–10, 10þ years); history of endometriosis (yes, no); tubal ligation (yes, no); menopausal hormone therapy (MHT) use (ever, never—postmenopausal women only); height (cm); BMI (<25, 25–29, ≥30 kg/m 2 ); physical inactivity [active, inactive ( 20 )]; smoking (never, former, current); alcohol intake (0, <10, 10þ g/day) and NSAID use [regular use (at least once per week), non-regular use]. We also considered the potential confounding effects of age and level of education (high school or less, some university, university graduate, graduate degree or above).\nData availability for exposures and covariates varied by OCAC site. Two studies (AUS, OPAL) collected information on all exposures and covariates; MAY did not collect information on breastfeeding, alcohol intake, NSAID or aspirin use; SEA did not collect information on endometriosis, inactivity, alcohol intake, NSAID or aspirin use. Among studies with data, the proportion of missing data ranged up from 4% for full-term pregnancy to 44% for inactivity and MHT use; we therefore performed multiple imputation by chained equations using the  mice  package in R statistical software ( 21 ). Further detail is in the  Supplementary Materials and Methods . We conducted analyses using both the original and imputed data, and the findings were consistent. Therefore, all results presented are based on imputed data unless otherwise specified to allow inclusion of studies that did not have information on all variables.\nWe employed a case-only analysis approach to evaluate differences in risk factors for ovarian cancer by  BRCA  status ( 22 ). We used the  lme4  package in R statistical software to generate generalized linear mixed models including a random effect for OCAC study to estimate interaction risk ratio (IRR) and 95% confidence intervals (CI), with  BRCA  (carrier vs. noncarrier) as the response variable ( 22 – 24 ). In brief, when the risk factor and outcome variable (here  BRCA  carrier) are independent, the IRR estimates the ratio between the relative risks for two different outcomes, but does not estimate the magnitude of the individual relative risks. An IRR of 1 would indicate the association between an exposure and ovarian cancer is similar for  BRCA  carriers and noncarriers; less than 1 suggests that the relative risk is lower (i.e., a weaker positive or stronger inverse association) among  BRCA  carriers; greater than 1 indicates a higher relative risk (i.e., a stronger positive or weaker inverse association) for  BRCA  carriers. We initially calculated IRRs for all  BRCA  carriers versus noncarriers and then separately for  BRCA 1 and  BRCA 2.\nIt is important to note that the IRR assumes independence between  BRCA  status and the risk factors. While  BRCA  testing is becoming more common, it was rarely performed during the time periods of the included studies, especially among the general population. Furthermore, given the majority of known risk factors relate to the period prior to menopause and the average age of study participants was 59, it is highly unlikely many women would have been tested at an early enough age to modify these factors.\nWe used directed acyclic graphs to identify confounders a  priori , and all models were adjusted for age at diagnosis. We considered the effects of education, OCP use, parity, inactivity, BMI, smoking, alcohol intake, NSAID/aspirin use as potential confounders of other associations; however, these were only included in the final models if their inclusion altered the beta coefficients for the exposure of interest by at least 10% ( 25 ). We also considered adjusting for year of diagnosis as the rate of  BRCA  positivity was higher in more recent studies, likely in part due to changes in clinical testing; however, estimates did not differ by greater than 10% so this was not included in the final models. As  BRCA  carriers are more likely to develop HGSC, we performed sensitivity analyses restricted to; (i) HGSC and (ii) endometrioid/clear cell carcinomas (END/CCC). We did not consider parity (per pregnancy) or differentiate between  BRCA1  and  BRCA2  in these analyses due to small numbers. We calculated Woolf statistic of homogeneity to assess differences in estimates between  BRCA 1 and  BRCA2  carriers and to compare differences between associations for HGSC and END/CCC ( 26 ).\nTo assess the effects of excluding women missing  BRCA  status from the analysis, we performed a sensitivity analysis by comparing results from complete cases models with and without weighting for the probability of inclusion in the model. Further detail is in the  Supplementary Materials and Methods .\nAnalyses were conducted using SAS version 9.4 (SAS Institute) and R statistical software (R Foundation for Statistical Computing).\nData described in the article cannot be made publicly available due to privacy and ethical limitations imposed by the original studies in which these data were collected; but can be shared upon approval of a data request form by the OCAC Data Access Coordinating Committee and with appropriate human subjects approval and data transfer agreements.\n\nDemographic, clinical, and lifestyle characteristics varied by  BRCA  status ( Table 1 ).  BRCA  carriers were generally younger, had a higher proportion of HGSC (83% vs. 62%) and were more likely to have used the OCP. The proportion of  BRCA  carriers with a family history of breast or ovarian cancer was more than double that for noncarriers (50% vs. 20%).\nFigure 1  presents the case-only comparisons between  BRCA  carriers ( BRCA1  or  BRCA2 ) and noncarriers for all included risk factors. Further disaggregation of results for  BRCA1  and  BRCA2  carriers is shown in  Supplementary Table S2 . IRRs were above 1.0 for known protective factors including ever being pregnant (IRR = 1.29, 95% CI; 1.00–1.67) and ever using the OCP (1.30, 95% CI; 1.07–1.60), suggesting the protective effects of pregnancy and OCP may be reduced in carriers compared with noncarriers. Similar differences were seen for tubal ligation and breastfeeding for  BRCA1  carriers. In contrast, the IRRs were below 1.0 for known risk factors including endometriosis (0.75, 95% CI; 0.54–1.01) and MHT use (0.66, 95% CI; 0.51–0.85), indicating that the risk associated with endometriosis and MHT use is reduced in  BRCA  carriers. This suggests that known associations with hormonal and reproductive factors were generally weaker for  BRCA  carriers than for noncarriers. None of the differences between  BRCA1  and  BRCA2  carriers was statistically significant, with P values from the Woolf statistic all greater than 0.1.\nResults restricted to the HGSC and END/CCC subtypes are shown in  Supplementary Table S3 . The IRR for ever being pregnant was 1.05 (0.77–1.42) for HGSC and 2.65 (0.91–7.76) for END/CCC, suggesting that among  BRCA  carriers the protective effects of pregnancy may be similar to noncarriers for HGSC but weaker for END/CCC. The opposite pattern was seen for OCP use where the IRRs were 1.47 (1.17–1.84) for ever use and 1.08 (1.00–1.17) per 5 years of use for HGSC suggesting a weaker association among  BRCA  carriers, compared with 0.81 (0.37–1.75) and 0.90 (0.61–1.33), respectively, for END/CCC.\nThe IRRs for endometriosis and MHT use were also lower for END/CCC [0.36 (0.06–2.08) and 0.36 (0.13–0.98), respectively] than for HGSC [0.87 (0.59–1.28)] and [0.69 (0.54–0.90), respectively]. This suggests the attenuation of these associations with cancer risk among  BRCA  carriers is greater for END/CCC cancers than for HGSC. However, the CIs were all wide and the differences between the histotypes were not statistically significant (all  P  values >0.1).\nThe IRRs for the lifestyle risk factors were generally close to 1.0, suggesting that there was no significant heterogeneity in the associations for smoking, BMI, inactivity, alcohol intake, NSAID use and aspirin use by  BRCA  status ( Supplementary Table S2 ). The pattern was similar for  BRCA1  and  BRCA2  and when we restricted to HGSC and END/CCC subtypes ( Supplementary Table S3 ).\n\nWe used a case–case analysis to investigate differences and similarities in risk factors for ovarian cancer between women with and without a pathogenic  BRCA  variant. In the general population, it is well known that many hormonal and reproductive factors are associated with the risk of developing ovarian cancer. Established protective factors include parity, OCP use, breastfeeding, and tubal ligation, while endometriosis and MHT use have been associated with an increased risk ( 27 ). Our results suggest that these known associations are generally weaker for those with a pathogenic  BRCA  variant than for those without, that is, the protective factors may be relatively less beneficial while risk factors are less harmful. The differences for tubal ligation and breastfeeding appeared to be limited to  BRCA1  carriers; however, the differences between  BRCA1  and  BRCA2  were not statistically significant. The weaker association we saw for OCP is consistent with a previous case-only study, which found that OCP use had a weaker protective effect in  BRCA  carriers although increasing parity had a greater protective effect; however, those estimates were not adjusted for potential confounders so it is difficult to draw conclusions ( 28 ).\nWhile the IRRs suggested that associations for reproductive/hormonal factors are generally weaker for  BRCA  carriers, it is still likely that these factors affect risk in carriers as previous studies restricted to  BRCA  carriers have shown associations. OCP use, tubal ligation, and breastfeeding have been found to be protective for both  BRCA 1 and  BRCA 2 carriers, whilst parity has been shown to be protective for  BRCA 1 but not  BRCA 2 ( 29 – 33 ). No association has been found between MHT use and ovarian cancer risk among  BRCA  carriers ( 34 ).\nGenerally, lifestyle risk factors are not as strongly associated with the risk of ovarian cancer as reproductive/hormonal factors. Increasing height, BMI and inactivity levels are associated with an increased risk of ovarian cancer, while regular NSAID use may be protective ( 27 ). No association has been found for alcohol while smoking has been shown to have differential associations by histotype ( 27 ). We found no evidence to suggest differences between  BRCA  carriers and noncarriers for any of the lifestyle factors included. This suggests that maintaining a healthy weight, being physically active and regular NSAID use are likely as protective for  BRCA  carriers as in the general population. One cohort study ( N  = 3,275) of  BRCA  carriers suggested an increased risk of ovarian cancer for those with greater than 4.3 total pack-years ( 9 ); however, we saw no evidence of a difference in the association for smoking by  BRCA  status.\nIt has been established in the general population that there are differences in risk factors by histologic subtype. In the general population, parity appears to be more protective for endometrioid and clear cell than for serous cancers, while OCP use is more protective for clear cell than for the endometrioid and serous subtypes ( 35 ). In addition, MHT use increased the risk for all histological subtypes except mucinous ( 35 ).Our results suggest that there is generally greater attenuation of risk for END/CCC (which tend to be more strongly associated with hormonal exposures in the general population) among  BRCA  carriers when compared with noncarriers. However, the small number of  BRCA  carriers with END/CCC meant that our results were imprecise and should be interpreted with caution.\nA strength of our study is that we used centrally harmonized data from five studies from Australia, the United Kingdom, and United States, allowing us to consistently adjust for confounders within the analyses. We used a case-only design to provide direct estimates of the gene–environment interaction, and investigated associations both overall and by histologic subtype to account for differences in histotype distribution by  BRCA  status. Despite pooling data from studies within OCAC, we were limited by the small sample size with only 371  BRCA1  carriers and 266  BRCA2  carriers which reduced the power of the study and the precision of the estimates.\nThe small sample size was in part due to the unavailability of  BRCA  status for the 1,430 women diagnosed between 2011 and 2019 in the SEA study and missing information for a further 1,434 women across all studies. Exclusion of the SEA women is unlikely to have introduced bias as they were excluded systematically based on their date of diagnosis and not due to their clinical characteristics. The remaining women excluded were more likely to have low-grade serous or mucinous cancers which are not associated with  BRCA  status and so not indicated for  BRCA  testing. Results from the sensitivity analysis showed no notable difference when including weighting for the probability of inclusion in the complete cases analysis, suggesting minimal impacts from possible selection bias.\nA further limitation was the retrospective self-reporting of pre-diagnosis hormonal/reproductive and lifestyle factors which potentially introduced recall error. However, recall error is unlikely to differ by  BRCA  status, and so would likely bias our findings toward the null, resulting in an underestimate of the differences. We did not have  BRCA  status for cancer-free controls so we were unable to estimate the magnitude of the risk in both carriers and noncarriers.\nIn summary, it appears that the associations for established hormonal and reproductive risk factors for ovarian cancer are generally weaker in  BRCA  carriers than in the general population while the associations for lifestyle factors are generally similar. Current recommendations for reducing risk, such as maintaining a healthy weight, abstaining from smoking, and being physically active, will also benefit  BRCA  carriers. Further research is needed to provide a better understanding of the extent to which ovarian cancer risk can be modified by lifestyle factors in  BRCA  variant carriers.","source_license":"public-domain-us","license_restricted":false}