A framework for assessing interactions for risk stratification models: the example of ovarian cancer

Generally, risk stratification models for cancer use effect estimates from risk/protective factor analyses that have not assessed potential interactions between these exposures. We have developed a 4-criterion framework for assessing interactions that includes statistical, qualitative, biological, and practical approaches. We present the application of this framework in an ovarian cancer setting because this is an important step in developing more accurate risk stratification models. Using data from 9 case-control studies in the Ovarian Cancer Association Consortium, we conducted a comprehensive analysis of interactions among 15 unequivocal risk and protective factors for ovarian cancer (including 14 non-genetic factors and a 36-variant polygenic score) with age and menopausal status. Pairwise interactions Received: March 23, 2023. Revised: June 08, 2023. Accepted: June 30, 2023 VC The Author(s) 2023. Published by Oxford University Press. All rights reserved. For permissions, please email: [email protected] JNCI: Journal of the National Cancer Institute , 2023, 115(11), 1420–1426 https://doi.org/10.1093/jnci/djad137 Advance Access Publication Date: July 12, 2023 Brief Communications between the risk/protective factors were also assessed. We found that menopausal status modifies the association among endometriosis, first-degree family history of ovarian cancer, breastfeeding, and depot-medroxyprogesterone acetate use and disease risk, highlighting the importance of understanding multiplicative interactions when developing risk prediction models. The development of risk stratification approaches to identify individuals who would most benefit from primary prevention strategies has become increasingly important. Risk stratification models use the effect estimates for the risk/protective factors considered to be unequivocal in their association with the disease under study. Generally, the effect estimates come from analyses in which multiplicative relationships were assumed among risk and protective factors. Using invasive epithelial ovarian cancer (ovarian cancer), we offer a strategy for the initial steps needed to develop accurate risk stratification models, including a 4-crite- rion framework for assessing whether potential interactions should be included. Interaction analyses are notoriously under- powered, so using this framework ensures that important differ- ences that may indicate departures from multiplicativity are not missed. • Criterion A (statistical approach): A likelihood ratio test com- paring a logistic model with the interaction term vs the same model without the interaction term (a 2-sided P < .05 for inter- action was considered statistically significant was used here, but other statistical approaches could be used); • Criterion B (qualitative approach): Comparing the consis- tency and magnitude of the odds ratios (ORs) of a factor across the levels of the other factor (visualization from strati- fied analysis); • Criterion C (biological approach): Considering biological plau- sibility; and • Criterion D (practical approach): Assessing the prevalence of the risk/protective factors to determine whether an interaction would have a meaningful impact on the risk stratification model. Ovarian cancer is an ideal example for refining risk stratification approaches because primary prevention strategies are available for w o m e na tb o t ha v e r a g ea n dh i g hr i s k ,i n c l u d i n gr i s k - r e d u c i n gs a l - pingo-oophorectomy, opportunistic salpingectomy, tubal ligation, and possibly hormonal contraceptives ( 1-4). Unequivocal ovarian cancer risk and protective factors include 14 non-genetic factors (4- 14) and a 36-variant polygenic score for ovarian cancer (15)( 1 5f a c - tors are shown in Supplementary Table 1 , available online). Importantly for ovarian and many other cancers affecting women, the effects of age and menopausal status on the risk/protective fac- tors must first be disentangled to determine whether one, both, or neither modifies the associations (16). We applied the framework to questionnaire data from 9 Ovarian Cancer Association Consortium case-control studies from Australia ( 17), Germany (18), and the United States ( 19-25). Institutional review board approval was obtained by the original studies, and all participants had provided written informed con- sent. To determine whether there was an age interaction, a men- opausal status interaction, or both, the initial ovarian cancer and risk and protective factor analyses were conducted among partic- ipants in the following strata (Table 1): • Stratum 1: Younger than 45 years of age and premenopausal • Stratum 2: Aged 45 to 54 years and premenopausal • Stratum 3: Aged 45 to 54 years and postmenopausal • Stratum 4: Aged 55 to 64 years and postmenopausal • Stratum 5: Aged 65 to 84 years and postmenopausal We found differences in the associations between the risk/pro- tective factors for ovarian cancer by menopausal status but not by age (particularly informed by comparing results between strata 2 and 3; Supplementary Table 2, A-D, available online) based on the 4-criterion interaction evaluation framework described earlier. Menopausal status appeared to modify the associations between ovarian cancer risk and endometriosis, first-degree fam- ily history of ovarian cancer, breastfeeding, and depot-medroxy- progesterone acetate use ( Table 2). For example, a self-reported history of endometriosis was associated with a greater increase in risk of ovarian cancer among premenopausal women than among postmenopausal participants ( P ¼ .04 for interaction; cri- terion A). Moreover, although no standardized definitions exist on how different the 2 stratum-specific associations should be for a factor to be an effect modifier, it is widely accepted that an OR less than 1.5 is considered a small effect size, while an OR between 1.5-2.0 is considered medium ( 26). Thus, the magni- tude of the difference in the endometriosis association between premenopausal (OR ¼ 1.94) and postmenopausal (OR ¼ 1.33) women is qualitatively meaningful (criterion B). Further, the endometriosis-menopausal status interaction is biologically plausible (criterion C) because during the premenopausal period, endometriosis is active (ovulatory proinflammatory and prolifer- ative processes) ( 27-30), whereas endometriosis is generally qui- escent in the postmenopausal period ( 31). Finally, endometriosis is estimated to have a prevalence of up to 10% in the general pop- ulation (32); thus, it is sufficiently common to warrant fitting sep- arate risk stratification models for pre- and postmenopausal women to be able to incorporate different effect estimates for endometriosis (criterion D). Given that 4 risk and protective factors suggest an interaction with menopausal status based on our framework, including one that met all 4 criteria, we further evaluated pairwise interactions between the risk and protective factors separately for pre- and postmenopausal women. Ultimately, our application of the framework led to the decision that there were no meaningful interactions among the 14 environmental factors or the polygenic score within the pre- or postmenopausal groups. As an example, among premenopausal women, the pairwise interaction between family history and parity was statistically significant ( P ¼ .022 for interaction; criterion A; Supplementary Table 2, O , available online). Parity also appeared to be more protective among women with a family history of ovarian cancer (OR ¼ 0.25 for 3 þ parity compared with nulliparity) vs women without a family history (OR ¼ 0.52) (criterion B); this interaction may also be biologically plausible (criterion C). Elevated progesterone levels during preg- nancy may clear genetically abnormal cells in the Fallopian tube fimbriae (33), which may preferentially benefit genetically driven ovarian cancers (34). Although this potential pairwise interaction may be useful for individual-level precision prevention, it would have minor impacts on ovarian cancer risk stratification because of the low proportion of people with a positive family history of ovarian cancer [approximately 2% ( 35)] as well as the low abso- lute risk of ovarian cancer among premenopausal women ( 36) (criterion D). Thus, we concluded that it is not necessary to M. T. Phung et al. | 1421 Table 1. Characteristics of participants with (cases) and without (controls) ovarian cancer included in the analysis, by age and menopausal status group Premenopausal women younger than 45 years Premenopausal women aged 45-54 years Postmenopausal women aged 45-54 years Postmenopausal women aged 55-64 years Postmenopausal women aged 65-84 years Case partici- pants Control partici- pants Case partici- pants Control partici- pants Case partici- pants Control partici- pants Case partici- pants Control partici- pantss Case partici- pants Control partici- pants (n ¼ 965) (n ¼ 2111) (n ¼ 1269) (n ¼ 2109) (n ¼ 903) (n ¼ 1214) (n ¼ 2493) (n ¼ 3502) (n ¼ 2226) (n ¼ 3148) Ovarian Cancer Association Consortium study, No. (%) AUS 2001-2005Australia 114 (11.8) 266 (12.6) 183 (14.4) 226 (10.7) 134 (14.8) 146 (12.0) 479 (19.2) 454 (13.0) 435 (19.5) 390 (12.4) DOV 2002-2009Washington, USA 116 (12.0) 182 (8.6) 209 (16.5) 311 (14.7) 99 (11.0) 122 (10.0) 425 (17.0) 660 (18.8) 224 (10.1) 414 (13.2) GER 1993-1998Germany 26 (2.7) 90 (4.3) 25 (2.0) 66 (3.1) 15 (1.7) 53 (4.4) 72 (2.9) 175 (5.0) 42 (1.9) 125 (4.0) HAW1993-2008Hawaii, USA 105 (10.9) 246 (11.7) 89 (7.0) 174 (8.3) 111 (12.3) 127 (10.5) 175 (7.0) 240 (6.9) 203 (9.1) 290 (9.2) HOP 2003-2009Western Pennsylvania, northeast Ohio, western New York, USA 64 (6.6) 176 (8.3) 120 (9.5) 354 (16.8) 34 (3.8) 125 (10.3) 208 (8.3) 489 (14.0) 252 (11.3) 534 (17.0) NEC 1992-2008New Hampshire, eastern Massachusetts, USA 235 (24.4) 496 (23.5) 249 (19.6) 354 (16.8) 187 (20.7) 214 (17.6) 422 (16.9) 542 (15.5) 347 (15.6) 452 (14.4) NJO 2002-2009New Jersey, USA 22 (2.3) 19 (0.9) 50 (3.9) 43 (2.0) 20 (2.2) 21 (1.7) 77 (3.1) 154 (4.4) 45 (2.0) 205 (6.5) UCI 1994-2005Southern California, USA 41 (4.2) 132 (6.3) 74 (5.8) 99 (4.7) 33 (3.7) 74 (6.1) 101 (4.1) 150 (4.3) 117 (5.3) 140 (4.4) USC 1993-2010Los Angeles, California, USA 242 (25.1) 504 (23.9) 270 (21.3) 482 (22.9) 270 (29.9) 332 (27.3) 534 (21.4) 638 (18.2) 561 (25.2) 598 (19.0) Age at diagnosis for cases/reference age for controls, year Mean (SD) 38.3 (5.28) 36.9 (5.92) 48.9 (2.63) 48.7 (2.63) 51.4 (2.31) 51.4 (2.34) 59.6 (2.77) 59.5 (2.78) 70.8 (4.46) 70.9 (4.49) Median (Min, Max) 40.0 (20.0, 44.0) 38.0 (18.0, 44.0) 49.0 (45.0, 54.0) 49.0 (45.0, 54.0) 52.0 (45.0, 54.0) 52.0 (45.0, 54.0) 60.0 (55.0, 64.0) 59.0 (55 .0, 64.0) 70.0 (65.0, 84.0) 70.0 (65.0, 84.0) Race/ethnicity, No. (%) Asian 111 (11.5) 148 (7.0) 108 (8.5) 121 (5.7) 69 (7.6) 61 (5.0) 100 (4.0) 99 (2.8) 130 (5.8) 158 (5.0) Black 30 (3.1) 54 (2.6) 24 (1.9) 45 (2.1) 28 (3.1) 24 (2.0) 57 (2.3) 53 (1.5) 35 (1.6) 52 (1.7) Hispanic White 67 (6.9) 135 (6.4) 49 (3.9) 92 (4.4) 59 (6.5) 67 (5.5) 120 (4.8) 110 (3.1) 70 (3.1) 56 (1.8) Non-Hispanic White 691 (71.6) 1603 (75.9) 1044 (82.3) 1749 (82.9) 688 (76.2) 1005 (82.8) 2121 (85.1) 3116 (89.0) 1927 (86.6) 2778 (88.2) Other a 62 (6.4) 157 (7.4) 41 (3.2) 97 (4.6) 54 (6.0) 55 (4.5) 84 (3.4) 119 (3.4) 58 (2.6) 99 (3.1) Missing 4 (0.4) 14 (0.7) 3 (0.2) 5 (0.2) 5 (0.6) 2 (0.2) 11 (0.4) 5 (0.1) 6 (0.3) 5 (0.2) Education level, No. (%) Less than high school 55 (5.7) 95 (4.5) 87 (6.9) 105 (5.0) 86 (9.5) 89 (7.3) 352 (14.1) 348 (9.9) 467 (21.0) 439 (13.9) High school 205 (21.2) 389 (18.4) 261 (20.6) 374 (17.7) 172 (19.0) 251 (20.7) 585 (23.5) 805 (23.0) 634 (28.5) 932 (29.6) Some college 297 (30.8) 605 (28.7) 376 (29.6) 644 (30.5) 275 (30.5) 374 (30.8) 742 (29.8) 1021 (29.2) 593 (26.6) 869 (27.6) College graduate or above 396 (41.0) 959 (45.4) 529 (41.7) 935 (44.3) 346 (38.3) 461 (38.0) 742 (29.8) 1237 (35.3) 441 (19.8) 810 (25.7) Missing 12 (1.2) 63 (3.0) 16 (1.3) 51 (2.4) 24 (2.7) 39 (3.2) 72 (2.9) 91 (2.6) 91 (4.1) 98 (3.1) a Other includes mixed race and those that do not belong in one of the specified racial/ethnic groups. AUS¼ Australian Ovarian Cancer Study; DOV ¼ Diseases of the Ovary and their Evaluation; GER¼ German Ovarian Cancer Study; HAW ¼ Hawaii Ovarian Cancer Case-Control Study; HOP ¼ Hormones and Ovarian Cancer Prediction; NEC ¼ New England Case Control Study; NJO ¼ New Jersey Ovarian Cancer Study; SD ¼ Standard deviation; USA ¼ United States of America; UCI ¼ University California Irvine Ovarian Study; USC ¼ Study of Lifestyle and Women’s Health. 1422 | JNCI: Journal of the National Cancer Institute , 2023, Vol. 115, No. 11 include an interaction term for family history and parity in a risk stratification model. Our proposed framework has some level of subjectivity. The risk associations for 3 of the 4 risk factors that drove our conclu- sion that associations differ by menopausal status were not stat- istically significantly different in the 2 strata (criterion A) but met the other 3 criteria used for evaluation. Some investigators, how- ever, may want to prioritize statistical significance (either using the interaction test presented here or using the Bayes false- positive probability) over the other 3 criteria and only use criteria B through D to decide against there being an interaction. Operationally, we decided that criterion A or B must be met before criteria C and D are considered. When criteria conflict with each other, however, we considered all criteria to inform our decision-making process (see the examples earlier). Another example is the age-parity interaction among postmenopausal women. The interaction was statistically significant ( P ¼ .009 for interaction; criterion A) and the prevalence of ever having given birth [85% (37)] is sufficient for this potential interaction to have a meaningful impact on risk stratification (criterion D). There was no pattern in the odds ratios for parity across the age groups (Supplementary Table 2, C, available online; criterion B), suggest- ing that this is a chance finding. We therefore determined, based on applying our framework, that this was not an interaction that should be incorporated into a risk stratification model. In conclusion, the application of our 4-criterion interaction evaluation framework ( Supplementary Tables 2, A-F , available online) demonstrates that menopausal status modifies the asso- ciation of at least one ovarian cancer risk/protective factor and the disease risk, supporting the use of separate models by meno- pausal status in risk stratification. The menopausal status–risk factors interactions are likely not influenced by histotype because the distributions are similar between pre- and postme- nopausal women aged 45 to 54 years ( Supplementary Table 3 , available online). The finding of no age–risk factor interactions could in part be due to the differences in histotype distributions across age groups. Interaction analyses stratified by histotype, however, would not be meaningful because of the small sample size of the rare histotypes. Additional research in prospective cohorts is needed to estimate absolute risk incorporating interac- tions to assess their impact on risk stratification. To develop meaningful risk stratification models, it is critical first to comprehensively assess interactions using statistical, qualitative, biological, and practical approaches (criteria A-D). Many published cancer risk stratification models either do not consider interactions or are based solely on P values (criterion A) to assess interactions ( 38-44). This approach has limitations because P values vary according to sample size, and there are issues related to multiple comparison. As such, we propose a framework that co-emphasizes the statistical (criterion A) and qualitative (criterion B) approaches and also includes the biologi- cal approach (criterion C) and practical approach (criterion D). Comprehensive interaction analysis for risk stratification can most effectively be done within consortia with large sample sizes. Continued collaboration in the field is necessary, and using the data fully must be a priority to move closer to realizing the goals of precision cancer prevention. Data availability The data generated in this study are not publicly available because of limitations imposed by the original studies in which these data were collected. The corresponding author will Table 2. Associations between risk/protective factors and ovarian cancer that differed by menopausal status among women aged 45-54 years Risk/protective factor All women aged 45-54 years (pre- and postmenopausal combined) Premenopausal women aged 45-54 years Postmenopausal women aged 45-54 years Case participants,a No. Control participants,a No. ORb (95% CI) Case participants,a No. Control participants,a No. ORb (95% CI) Case participants,a No. Control participants,a No. ORb (95% CI) P- interactionc Interaction criteria metd Breastfeeding Never 1212 1278 1.0 693 763 1.0 519 515 1.0 <12 months 551 978 0.76 (0.64 to 0.89) 327 595 0.78 (0.62 to 0.98) 224 383 0.71 (0.55 to 0.92) /C21 12 months 397 998 0.59 (0.49 to 0.72) 240 706 0.53 (0.42 to 0.68) 157 292 0.69 (0.51 to 0.94) .064 (b), (c), (d) Depot-medroxyprogesterone acetate use No 1886 2793 1.0 1109 1793 1.0 777 1000 1.0 Yes 21 76 0.61 (0.36 to 1.02) 10 52 0.51 (0.26 to 1.00) 11 24 0.80 (0.38 to 1.69) .35 (b), (c), (d) First-degree family history of ovarian cancer No 1610 2548 1.0 943 1633 1.0 667 915 1.0 Yes 119 87 2.15 (1.56 to 2.97) 69 48 2.43 (1.58 to 3.73) 50 39 1.83 (1.15 to 2.91) .39 (b), (c), (d) Endometriosis No 1889 3058 1.0 1128 1981 1.0 761 1077 1.0 Yes 269 255 1.60 (1.32 to 1.95) 135 123 1.94 (1.47 to 2.57) 134 132 1.33 (1.00 to 1.76) .041 (a), (b), (c), (d) a Numbers may not sum to total because of missing values. CI¼ confidence interval; OR ¼ odds ratio. b Pooled estimates from logistic regression models in the 50 imputed datasets, adjusted for age at diagnosis for cases/reference age for controls (45-49 years vs 50-54 years), race/ethnicity, education level, and Ovarian Cancer Association Consortium study. c P value for interaction between risk or protective factor and menopausal status using the likelihood ratio test. d Criteria to assess interactions: (a) P < .05 for interaction; (b) odds ratios of a factor across the levels of the other factor are consistent, and the differences in magnitude are large; (c) the interaction is biologically plausible; and (d) the prevalence of the risk factors is large enough so that the interaction would have a meaningful impact on the risk stratification model. M. T. Phung et al. | 1423 facilitate access through existing data request processes for the Ovarian Cancer Association Consortium. Author contributions Minh Tung Phung, PhD, MPH (Conceptualization; Formal analy- sis; Methodology; Writing—original draft; Writing—review & edit- ing), Malcolm C. Pike, PhD (Funding acquisition; Writing—review & editing), Bhramar Mukherjee, PhD (Writing—review & editing), Rafael Meza, PhD (Writing—review & editing), Gillian E. Hanley, PhD (Writing—review & editing), Kathleen R. Cho, MD (Writing— review & editing), Andrew Berchuck, MD (Funding acquisition; Writing—review & editing), Argyrios Ziogas, PhD (Funding acquis- ition; Writing—review & editing), Nur Zeinomar, PhD (Funding acquisition; Writing—review & editing), Anna H. Wu, PhD (Funding acquisition; Writing—review & editing), Penelope M. Webb, PhD (Funding acquisition; Writing—review & editing), Linda J. Titus, PhD (Funding acquisition; Writing—review & edit- ing), Kathryn L. Terry, ScD (Funding acquisition; Writing—review & editing), Bo Qin, PhD (Funding acquisition; Writing—review & editing), Celeste Leigh Pearce, PhD, MPH (Conceptualization; Funding acquisition; Methodology; Resources; Supervision; Writing—original draft; Writing—review & editing), Paul D. P. Pharoah, PhD (Funding acquisition; Writing—review & editing), Francesmary Modugno, PhD, MPH (Funding acquisition; Writing—review & editing), Allan Jensen, PhD (Funding acquisi- tion; Writing—review & editing), Holly R. Harris, ScD, MPH (Funding acquisition; Writing—review & editing), Marc T. Goodman, PhD (Funding acquisition; Writing—review & editing), Renee T. Fortner, PhD (Funding acquisition; Writing—review & editing), Jennifer Anne Doherty, MS, PhD (Funding acquisition; Writing—review & editing), Daniel W. Cramer, MD, ScD (Funding acquisition; Writing—review & editing), Jenny Chang-Claude, PhD (Funding acquisition; Writing—review & editing), Michael E. Carney, MD (Funding acquisition; Writing—review & editing), Elisa V. Bandera, MD, PhD (Funding acquisition; Writing—review & editing), Hoda Anton-Culver, PhD (Funding acquisition; Writing—review & editing), Karen McLean, MD, PhD (Writing— review & editing), Alice W. Lee, PhD, MPH (Writing—review & editing), Kirsten B. Moysich, MS, PhD (Funding acquisition; Writing—review & editing), Britton Trabert, PhD (Conceptualization; Funding acquisition; Methodology; Supervision; Writing—original draft; Writing—review & editing). Funding The Ovarian Cancer Association Consortium is supported by a grant from the Ovarian Cancer Research Fund thanks to dona- tions by the family and friends of Kathryn Sladek Smith (PPD/ RPCI.07). The scientific development and funding for this project were in part supported by the US National Cancer Institute GAME-ON (Genetic Associations and Mechanisms in Oncology) Post-GWAS Initiative (U19-CA148112). This study used data gen- erated by the Wellcome Trust Case Control consortium, which was funded by the Wellcome Trust under award No. 076113. The

published here are in part based on data generated by The Cancer Genome Atlas Pilot Project established by the National Cancer Institute and National Human Genome Research Institute (dbGap accession No. phs000178.v8.p7). The Ovarian Cancer Association Consortium OncoArray geno- typing project was funded through grants from the US National Institutes of Health (CA1X01HG007491-01, U19-CA148112, R01- CA149429, and R01-CA058598; Canadian Institutes of Health Research (MOP-86727 and the Ovarian Cancer Research Fund (A.B.). The COGS project was funded through a European Commission’s Seventh Framework Programme grant (agreement No. 223175 - HEALTH-F2-2009-223175). Funding for individual studies: AUS: The Australian Ovarian Cancer Study was supported by the US Army Medical Research and Materiel Command (DAMD17-01-1-0729); National Health & Medical Research Council of Australia (199600, 400413, and 400281); Cancer Councils of New South Wales, Victoria, Queensland, South Australia, and Tasmania; and Cancer Foundation of Western Australia (Multi-State Applications 191, 211, and 182). The Australian Ovarian Cancer Study gratefully acknowledges additional support from Ovarian Cancer Australia and the Peter MacCallum Foundation; DOV: National Institutes of Health R01-CA112523 and R01-CA87538; GER: German Federal Ministry of Education and Research, Programme of Clinical Biomedical Research (01 GB 9401), and the German Cancer Research Center (DKFZ); HAW: US National Institutes of Health (R01-CA58598, N01-CN-55424, and N01-PC-67001); HOP: University of Pittsburgh School of Medicine Dean’s Faculty Advancement Award (F. Modugno), US Department of Defense (DAMD17-02-1-0669), and National Cancer Institute (K07- CA080668, R01-CA95023, P50-CA159981, MO1-RR000056, R01- CA126841); NEC: R01-CA54419 and P50-CA105009 and US Department of Defense W81XWH-10-1-02802; NJO: National Cancer Institute (NIH-K07 CA095666, R01-CA83918, NIH-K22- CA138563, and P30-CA072720) and the Rutgers Cancer Institute of New Jersey; UCI: Australian Ovarian Cancer Study R01- CA058860 and the Lon V. Smith Foundation grant LVS-39420; USC: P01CA17054, P30CA14089, R01CA61132, N01PC67010, R03CA113148, R03CA115195, N01CN025403, and California Cancer Research Program (00-01389 V-20170, 2II0200). Funding for individuals: B.T.: Cancer Center Support Grant, National Cancer Institute, P30CA040214; P.M.W.: National Health & Medical Research Council of Australia GNT1173346; C.L.P.: the National Institutes of Health/National Cancer Institute Support Grant P30 CA046592. M.C.P. was supported in part through the National Institutes of Health/National Cancer Institute Support Grant P30 CA008748 (P.I. S.M. Vickers) to Memorial Sloan Kettering Cancer Center. Conflicts of interest P.M.W. has received a speaker’s fee and funding from AstraZeneca (2017-2020) for an unrelated study of ovarian can- cer. E.V.B. served in Pfizer’s Advisory Board to enhance participa- tion of people from minority groups in clinical trials. Other authors do not have any conflicts of interest.

We are grateful to the family and friends of Kathryn Sladek Smith for their generous support of the Ovarian Cancer Association Consortium through their donations to the Ovarian Cancer Research Fund. The OncoArray and COGS genotyping projects would not have been possible without the contributions of the following: Per Hall (COGS); Douglas F. Easton, Kyriaki Michailidou, Manjeet K. Bolla, Qin Wang (BCAC); Marjorie J. Riggan (OCAC); Rosalind A. Eeles, Douglas F. Easton, Ali Amin Al Olama, Zsofia Kote-Jarai, and Sara Benlloch (PRACTICAL); Georgia Chenevix-Trench, Antonis Antoniou, Lesley McGuffog, Fergus Couch, and Ken Offit (CIMBA); Joe Dennis, Jonathan P. 1424 | JNCI: Journal of the National Cancer Institute , 2023, Vol. 115, No. 11 Tyrer, Siddhartha Kar, Alison M. Dunning, Andrew Lee, and Ed Dicks; Craig Luccarini and the staff of the Centre for Genetic Epidemiology Laboratory; Javier Benitez, Anna Gonzalez-Neira, and the staff of the CNIO genotyping unit; Jacques Simard and Daniel C. Tessier; Francois Bacot, Daniel Vincent, Sylvie LaBoissie` re, Frederic Robidoux, and the staff of McGill University and G /C19enome Qu /C19ebec Innovation Centre; Stig E. Bojesen, Sune F. Nielsen, Borge G. Nordestgaard, and the staff of the Copenhagen DNA laboratory; and Julie M. Cunningham, Sharon A. Windebank, Christopher A. Hilker, Jeffrey Meyer, and the staff of Mayo Clinic Genotyping Core Facility. We pay special tribute to the contribution of Brian Henderson to the GAME-ON consortium until he sadly passed away on June 20, 2015; to Olga M. Sinilnikova for her contribution to CIMBA and for her part in the initiation and coordination of GEMO until she sadly passed away on June 30, 2014, and to Catherine M. Phelan for her contribution to OCAC and coordination of the OncoArray until she passed away on September 22, 2017. We thank the study participants, doctors, nurses, clinical and scientific collaborators, health care providers, and health information sources who have contributed to the many studies contributing to this manuscript.

for individual studies: AUS: The AOCS also acknowledges the cooperation of the participating institutions in Australia and the contribution of the study nurses, research assistants, and all clinical and scientific collaborators. The com- plete AOCS Study Group can be found at www.aocstudy.org.W e would like to thank all the women who participated in this research program; GER: The German Ovarian Cancer Study (GER) thank Ursula Eilber for competent technical assistance; NJO:W e thank Drs Sara Olson, Lisa Paddock, Lorna Rodriguez, and all par- ticipants and research staff at the Rutgers Cancer Institute of New Jersey, the New Jersey State Cancer Registry, and Memorial Sloan Kettering Cancer Center. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Earlier results were presented at the American Association for Cancer Research annual meeting in June 2020 as an e-poster.

1. Hanley GE, Pearce CL, Talhouk A, et al. Outcomes from opportun- istic salpingectomy for ovarian cancer prevention. JAMA Netw Open. 2022;5(2):e2147343. doi: 10.1001/jamanetworkopen.2021. 47343. 2. Menon U, Karpinskyj C, Gentry-Maharaj A. Ovarian cancer pre- vention and screening. Obstet Gynecol . 2018;131(5):909-927. doi:10.1097/AOG.0000000000002580. 3. Beral V, Doll R, Hermon C, Peto R, Reeves G; Collaborative Group on Epidemiological Studies of Ovarian Cancer. Ovarian cancer and oral contraceptives: collaborative reanalysis of data from 45 epidemiological studies including 23,257 women with ovar- ian cancer and 87,303 controls. Lancet. 2008;371(9609):303-314. doi:10.1016/S0140-6736(08)60167-1. 4. Pearce CL, Rossing MA, Lee AW, et al.; Ovarian Cancer Association Consortium. Combined and interactive effects of environmental and GWAS-identified risk factors in ovarian can- cer. Cancer Epidemiol Biomarkers Prev . 2013;22(5):880-890. doi:10.1158/1055-9965.EPI-12-1030-T. 5. Wentzensen N, Poole EM, Trabert B, et al. Ovarian cancer risk factors by histologic subtype: an analysis from the Ovarian Cancer Cohort Consortium. J Clin Oncol . 2016;34(24):2888-2898. doi:10.1200/JCO.2016.66.8178. 6. Lee AW, Rosenzweig S, Wiensch A, et al.; Australian Ovarian Cancer Study Group. Expanding our understanding of ovarian cancer risk: the role of incomplete pregnancies.J Natl Cancer Inst. 2021;113(3):301-308. doi:10.1093/jnci/djaa099. 7. Babic A, Sasamoto N, Rosner BA, et al. Association between breastfeeding and ovarian cancer risk. JAMA Oncol . 2020;6(6):e200421. doi:10.1001/jamaoncol.2020.0421. 8. Phung MT, Lee AW, Wu AH, et al.; Ovarian Cancer Association Consortium. Depot-medroxyprogesterone acetate use is associ- ated with decreased risk of ovarian cancer: the mounting evidence of a protective role of progestins.Cancer Epidemiol Biomarkers Prev. 2021;30(5):927-935. doi:10.1158/1055-9965.EPI-20-1355. 9. Dixon-Suen SC, Nagle CM, Thrift AP, et al.; Ovarian Cancer Association Consortium. Adult height is associated with increased risk of ovarian cancer: a Mendelian randomisation study. Br J Cancer . 2018;118(8):1123-1129. doi: 10.1038/s41416- 018-0011-3 10. Day FR, Thompson DJ, Helgason H, et al.; LifeLines Cohort Study; InterAct Consortium; kConFab/AOCS Investigators; Endometrial Cancer Association Consortium; Ovarian Cancer Association Consortium; PRACTICAL consortium. Genomic analyses identify hundreds of variants associated with age at menarche and support a role for puberty timing in cancer risk. Nat Genet. 2017;49(6):834-841. doi:10.1038/ng.3841. 11. Wu AH, Pearce CL, Lee AW, et al. Timing of births and oral con- traceptive use influences ovarian cancer risk. Int J Cancer . 2017;141(12):2392-2399. doi:10.1002/ijc.30910. 12. Lee AW, Ness RB, Roman LD, et al.; Ovarian Cancer Association Consortium. Association between menopausal estrogen-only therapy and ovarian carcinoma risk. Obstet Gynecol . 2016;127(5):828-836. doi:10.1097/AOG.0000000000001387. 13. Lee AW, Wu AH, Wiensch A, et al.; Ovarian Cancer Association Consortium. Estrogen plus progestin hormone therapy and ovar- ian cancer: a complicated relationship explored. Epidemiology. 2020;31(3):402-408. doi:10.1097/EDE.0000000000001175. 14. Olsen CM, Nagle CM, Whiteman DC, et al.; Ovarian Cancer Association Consortium. Obesity and risk of ovarian cancer sub- types: evidence from the Ovarian Cancer Association Consortium. Endocr Relat Cancer. 2013;20(2):251-262. doi:10.1530/ ERC-12-0395. 15. Lee A, Yang X, Tyrer J, et al. A comprehensive epithelial tubo- ovarian cancer risk prediction model incorporating genetic and epidemiological risk factors. J Med Genet . 2020;59(7):632-643. doi:10.1136/jmedgenet-2021-107904. 16. Trentham-Dietz A, Sprague BL, Hampton JM, et al. Modification of breast cancer risk according to age and menopausal status: a combined analysis of five population-based case-control stud- ies. Breast Cancer Res Treat . 2014;145(1):165-175. doi: 10.1007/ s10549-014-2905-y. 17. Merritt MA, Green AC, Nagle CM, Webb PM; ACSO Cancer, AOCS Group. Talcum powder, chronic pelvic inflammation and NSAIDs in relation to risk of epithelial ovarian cancer. Int J Cancer. 2008;122(1):170-176. doi:10.1002/ijc.23017. 18. Royar J, Becher H, Chang-Claude J. Low-dose oral contracep- tives: protective effect on ovarian cancer risk. Int J Cancer . 2001;95(6):370-374. doi: 10.1002/1097- 0215(20011120)95:63.0.co;2-t. 19. Bodelon C, Cushing-Haugen KL, Wicklund KG, Doherty JA, Rossing MA. Sun exposure and risk of epithelial ovarian cancer. Cancer Causes Control . 2012;23(12):1985-1994. doi: 10.1007/ s10552-012-0076-x. 20. Lurie G, Terry KL, Wilkens LR, et al. Pooled analysis of the asso- ciation of PTGS2 rs5275 polymorphism and NSAID use with M. T. Phung et al. | 1425 invasive ovarian carcinoma risk. Cancer Causes Control . 2010;21(10):1731-1741. doi:10.1007/s10552-010-9602-x. 21. Ness RB, Dodge RC, Edwards RP, Baker JA, Moysich KB. Contraception methods, beyond oral contraceptives and tubal ligation, and risk of ovarian cancer. Ann Epidemiol . 2011;21(3):188-196. doi:10.1016/j.annepidem.2010.10.002. 22. Terry KL, De Vivo I, Titus-Ernstoff L, Shih MC, Cramer DW. Androgen receptor cytosine, adenine, guanine repeats, and hap- lotypes in relation to ovarian cancer risk. Cancer Res . 2005;65(13):5974-5981. doi:10.1158/0008-5472.CAN-04-3885. 23. Bandera EV, King M, Chandran U, Paddock LE, Rodriguez- Rodriguez L, Olson SH. Phytoestrogen consumption from foods and supplements and epithelial ovarian cancer risk: a population-based case control study. BMC Womens Health . 2011;11:40. doi:10.1186/1472-6874-11-40 24. Ziogas A, Gildea M, Cohen P, et al. Cancer risk estimates for fam- ily members of a population-based family registry for breast and ovarian cancer. Cancer Epidemiol Biomarkers Prev . 2000;9(1):103-111. 25. Wu AH, Pearce CL, Tseng CC, Pike MC. African Americans and Hispanics remain at lower risk of ovarian cancer than non- Hispanic Whites after considering nongenetic risk factors and oophorectomy rates. Cancer Epidemiol Biomarkers Prev . 2015;24(7):1094-1100. doi:10.1158/1055-9965.EPI-15-0023. 26. Sullivan GM, Feinn R. Using effect size-or why the P value is not enough. JG r a dM e dE d u c. 2012;4(3):279-282. doi:10.4300/JGME-D- 12-00156.1. 27. Salehi F, Dunfield L, Phillips KP, Krewski D, Vanderhyden BC. Risk factors for ovarian cancer: an overview with emphasis on hormonal factors. J Toxicol Environ Health B Crit Rev . 2008;11(3- 4):301-321. doi:10.1080/10937400701876095. 28. Vercellini P, Somigliana E, Vigan /C18o P, Abbiati A, Barbara G, Crosignani PG. Endometriosis: current therapies and new phar- macological developments. Drugs. 2009;69(6):649-675. doi:10.2165/00003495-200969060-00002. 29. Macci /C18o A, Madeddu C. Inflammation and ovarian cancer. Cytokine. 2012;58(2):133-147. doi:10.1016/j.cyto.2012.01.015. 3 0 . D u f f yD M ,K oC ,J oM ,B r a n n s t r o mM ,C u r r yT E .O v u l a t i o n :p a r a l - lels with inflammatory processes. Endocr Rev. 2019;40(2):369-416. doi:10.1210/er.2018-00075. 31. Tan DA, Almaria MJG. Postmenopausal endometriosis: drawing a clearer clinical picture. Climacteric. 2018;21(3):249-255. doi:10.1080/13697137.2018.1450855. 32. Eskenazi B, Warner ML. Epidemiology of endometriosis. Obstet Gynecol Clin North Am . 1997;24(2):235-258. doi: 10.1016/s0889- 8545(05)70302-8. 33. Rodriguez GC, Kauderer J, Hunn J, et al. Phase II trial of chemo- preventive effects of levonorgestrel on ovarian and fallopian tube epithelium in women at high risk for ovarian cancer: an NRG Oncology Group/GOG study. Cancer Prev Res (Phila) . 2019;12(6):401-412. doi:10.1158/1940-6207.CAPR-18-0383. 34. Milne RL, Osorio A, Ram /C19on y Cajal T, et al. Parity and the risk of breast and ovarian cancer in BRCA1 and BRCA2 mutation car- riers. Breast Cancer Res Treat . 2010;119(1):221-232. doi: 10.1007/ s10549-009-0394-1. 35. Ramsey SD, Yoon P, Moonesinghe R, Khoury MJ. Population- based study of the prevalence of family history of cancer: impli- cations for cancer screening and prevention. Genet Med . 2006;8(9):571-575. doi:10.1097/01.gim.0000237867.34011.12. 36. Roett MA, Evans P. Ovarian cancer: an overview. Am Fam Physician. 2009;80(6):609-616. 37. Martinez GM, Daniels K, Febo-Vazquez I. Fertility of men and women aged 15-44 in the United States: national survey of fam- ily growth, 2011-2015. Natl Health Stat Rep. 2018;(113):1-17. 38. Colditz GA, Atwood KA, Emmons K, et al. Harvard report on cancer prevention volume 4: Harvard Cancer Risk Index. Risk Index Working Group, Harvard Center for Cancer Prevention. Cancer Causes Control . 2000;11(6):477-488. doi: 10.1023/ a:1008984432272 . 39. Rosner BA, Colditz GA, Webb PM, Hankinson SE. Mathematical models of ovarian cancer incidence. Epidemiology. 2005;16(4):508-515. doi:10.1097/01.ede.0000164557.81694.63. 40. Vitonis AF, Titus-Ernstoff L, Cramer DW. Assessing ovarian cancer risk when considering elective oophorectomy at the time of hys- terectomy. Obstet Gynecol . 2011;117(5):1042-1050. doi: 10.1097/ AOG.0b013e318212fcb7. 41. Pfeiffer RM, Park Y, Kreimer AR, et al. Risk prediction for breast, endometrial, and ovarian cancer in white women aged 50 y or older: derivation and validation from population-based cohort studies. PLoS Med . 2013;10(7):e1001492. doi: 10.1371/jour- nal.pmed.1001492. 42. Pearce CL, Stram DO, Ness RB, et al. Population distribution of lifetime risk of ovarian cancer in the United States. Cancer Epidemiol Biomarkers Prev . 2015;24(4):671-676. doi: 10.1158/1055- 9965.EPI-14-1128. 43. Clyde MA, Palmieri Weber R, Iversen ES, et al.; on behalf of the Ovarian Cancer Association Consortium. Risk prediction for epi- thelial ovarian cancer in 11 United States-based case-control studies: incorporation of epidemiologic risk factors and 17 con- firmed genetic loci. Am J Epidemiol . 2016;184(8):579-589. doi:10.1093/aje/kww091. 44. Lee A, Yang X, Tyrer J, et al. Comprehensive epithelial tubo- ovarian cancer risk prediction model incorporating genetic and epidemiological risk factors. J Med Genet . 2022;59(7):632-643. doi:10.1136/jmedgenet-2021-107904. 1426 | JNCI: Journal of the National Cancer Institute , 2023, Vol. 115, No. 11