Timing of births and oral contraceptive use influences ovarian cancer risk.

There were 1,632 cases and 2,340 controls available for analysis. The race/ethnicity, age, education and income distributions of these cases and controls are shown in Supporting Information Table 1 . Similarly, the relative risks for the well-established risk factors of family history of ovarian cancer, history of endometriosis, talc use and BMI are available in Supporting Information Table 2 . Table 1 shows the relative risks for age at first birth and age at last birth fitted separately and together, adjusted for the ‘basic factor set’ and number of births (continuous) and duration of COC use (continuous at age <35 years and continuous at age ≥35 years—see below). When fitted separately, age at first birth and age at last birth were both highly significantly related to ovarian cancer risk. However, when fitted together, only age at first birth remained significant. Compared to a first birth under age 25, a first birth after age 35 was associated with a 47% reduced risk of ovarian cancer (RR = 0.53, 95% CI 0.32–0.88). Similar results were obtained when we restricted the analysis to cases and controls who had only one birth, and when we restricted the analysis to cases and controls who had two births (data not shown). The relative risks associated with increasing numbers of births adjusted for the ‘basic factor set’ and duration of COC use (continuous at age <35 years and continuous at age ≥35 years—see below) are shown in Table 2 . The relative risk associated with a single birth was 31% lower than that associated with nulliparity, much lower than one would expect based on a linear model (on a logistic scale) of reduction in risk with each birth (RR = 0.87). However, when fitted together with the effect of age at first birth, this markedly larger effect of a single birth was eliminated completely. This was due to the much greater average age at first birth among women who only had one birth: 46% of controls who only had one birth had the birth at age 30 or later, while only 14% of controls who had more than one birth had the first birth at age 30 or later. The relative risks associated with the duration of use of COCs are shown in Table 3 . Each 5 years of use was associated with a 28% lower risk of ovarian cancer ( p < 0.001). There was a 34% reduction in risk associated with 5 years of COC use before age 35, while 5 years of use after age 35 was only associated with a 14% lower risk ( p for difference in effect, 0.019). There was little difference in the reduction in risk from a fixed duration of COC use before age 25 (0.62, 95% CI 0.52–0.80) and such use at ages 25–34 (0.71, 95% CI 0.60–0.83). The decreased protective effect seen with use after age 35 was consistently seen in all estrogen and progestin ‘potency’ formulations (data not shown). Table 4 shows the relative risks for numbers of births and age at first birth by histotype. Although the relative risks show the same relationships with each histotype, the effects are greater for endometrioid and clear cell ovarian cancers, and the effect of age at first birth is weakest and not statistically significant for HGSOC. Table 4 also shows the relative risks for duration of use of COCs overall and by age at use by histotype. The relative risks show approximately the same relationships with each tumor type, with the possible exception of clear cell ovarian cancers for which COC use in both age groups was equally protective. The relative risks associated with numbers of births, age at first birth and duration of COC use by age at diagnosis is shown in Table 5 . Overall there was a 13% reduction in ovarian cancer risk per birth. The reduction in risk per birth was much stronger at younger ages at diagnosis: a 27% reduction per birth at diagnosis ages <45 years compared to an 8% reduction at diagnosis ages ≥65 years. A similar trend was observed for age at first birth with a 16% reduction at diagnosis ages <45 years compared to a 9% reduction at ages ≥65 years ( Table 5 ). No clear effect of age at diagnosis was seen with the reductions in risk with COC use.

We have previously published details of the methods used in the four studies. 1 , 12 These studies were approved by the University of Southern California (USC) Institutional Review Board, and written informed consent was obtained from each case and control before her interview. For all studies, newly diagnosed histologically confirmed ovarian cancer cases were identified from the Cancer Surveillance Program (CSP) of the University of Southern California; the CSP is the Surveillance, Epidemiology and End Results (SEER) Registry for LAC. Eligible patients were female residents of LAC of self-reported non-Hispanic white, Hispanic, or African-American race/ethnicity. Cases were eligible for inclusion in the study if they were between 18 and 74 years of age at diagnosis and up to age 79 for cases diagnosed between 2003 and 2008. A total of 3,370 patients met the study criteria. Overall, 15.7% of patients declined to be interviewed, 16.9% had died or were too ill to be interviewed, and 11.4% could not be located or had moved out of LAC. We were thus able to carry out in-person interviews with 1,886, a 63.2% participation rate of the patients approached. The response rate was higher for patients diagnosed with localized cancer (69%) compared with those with more advanced stage (61%). Response rates were highest for those diagnosed under age 60 (70%), intermediate for those ages 60–69 (59%) and lowest for those ages 70+ (47%). In this analysis, we excluded 185 patients who had a previous cancer (other than non-melanoma skin cancer) or had prior bilateral oophorectomy, and a further 69 cases with missing or inconsistent data on one or more of the variables included in the analysis presented here. The final analysis was based on 1,632 cases. Controls were female residents of LAC with at least one intact ovary identified using a well-tested control selection algorithm. 1 Briefly, we initially defined a specified sequence of houses to be visited in the neighborhoods where index cases lived at the time of diagnosis. We then sought to interview the first eligible resident in the sequence. If the first eligible control subject refused to participate, the second eligible one in the sequence was asked, and so on. Letters were left when no one was home, and follow-up was by mail and telephone. Controls were individually matched to cases on neighborhood, race/ethnicity and year of birth (±5 years). In one study, selection of controls for cases >65 years of age was augmented, if necessary, by using lists of female residents of LAC provided by the Health Care Financing Administration, matched to the case on zip code, race/ethnicity and year of birth closest to the case’s year of birth. We excluded potential controls who had ever been diagnosed with a cancer (ignoring non-melanoma skin cancer). Overall, 70% of the controls interviewed were the first identified control. In-person interviews were conducted using a standardized questionnaire that included the use of a life calendar. The questionnaire covered events up to 12 months before a case’s diagnosis date and a similar reference date for the controls. The demographic, lifestyle and medical history variables considered in this analysis include age, age at menarche, education, SES (socio-economic status based on census tract of residence), 13 tubal ligation, self-reported physician-diagnosed endometriosis, first-degree family history of ovarian cancer, talc use, current body mass index (BMI, kg m −2 ), COC use, gravidity, parity, self-reported physician-diagnosed history of infertility not attributable to problems of the partner and tumor histology (from the SEER registry). We used stratified multivariate logistic regression, using the statistical package program STATA 13 (StataCorp, College Station, TX). At the termination of each study, some cases had not been matched to a control and there were some controls whose cases had to be excluded after they completed the interview, because they were ineligible for the current analysis (e.g., not ovarian cancer or did not live in LAC at the time of diagnosis). Although the studies were designed as matched case–control studies, in this report we have used all interviewed cases and controls by adopting a stratified multivariate logistic regression analysis approach with joint stratification for the four studies, three race/ethnicity groups and age groups (<30, 5-year age groups through age 79). The following risk factors were adjusted for in all (stratified) analyses: first-degree family history of ovarian cancer (mother or sister), history of endometriosis, talc use, tubal ligation and history of infertility; all of which were entered in the logistic regression analysis as categorical binary (Yes/No) variables. Age at menarche (continuous variable) and BMI (continuous variable) were also adjusted for in all analyses. In addition, education (<high school graduate, high school graduate, some further training, college graduate); and five levels of SES were also adjusted for as categorical variables in all analyses. Age and type of menopause was also adjusted for in all analyses as a categorical variable with categories: premenopausal, hysterectomy (before menopause) at age <40 years and ≥40 years, natural menopause (age at last natural menstrual period—LNMP) at age <45 years, 45–, 50– and 55+. For a woman who started menopausal hormone therapy (HT) while she was still having natural periods and did not have any natural periods after stopping this episode of HT, age at menopause was taken to be the age at which the subject began this episode of HT. If a woman had a natural menstrual period at least 3 months after stopping HT, this episode of use was not considered in defining her age at menopause and not considered to be an episode of HT. If a woman had no natural menstrual periods after stopping COCs (which were started before menopause), her age at menopause was taken as the age at which she last took a COC. The justification for this approach is given in Pike et al. 14 Each episode of HT use was categorized as estrogen alone (ET), sequential estrogen-progestin (sEPT), continuous combined estrogen-progestin (cEPT) or other. ET was defined as use of estrogen alone for ≥20 days per ‘month’; sEPT as use of estrogen for ≥20 days per month and number of progestin days ≤60% of number of estrogen days; and cEPT as use of estrogen for ≥20 days per month and number of progestin days the same or greater than number of days of estrogen. The above factors, taken together, are termed ‘the basic factor set’ and are adjusted for in all analyses. Number of births (defined as a pregnancy of >26 weeks, counting from date of last menses) was analyzed categorically (0, 1, 2, 3 and 4+) as well as a continuous variable; age at first birth was analyzed categorically (<20 years, 20–, 25–, 30– and 35+) as well as a continuous variable with these categories scored as 1, 2, 3, 4 and 5; and age at last birth was analyzed categorically (<25 years, 25–29, 30–34, 35–39 and 40+) as well as a continuous variable with these categories scored as 1, 2, 3, 4 and 5. Number of incomplete pregnancies (pregnancy of ≤26 weeks) was analyzed as a continuous variable. Duration of COC use was analyzed categorically (no use, use for <5 years, 5–, 10– and 15+ years) as well as a continuous variable. We also analyzed the ages at which COCs were used (duration of use at ages 35 μg or if the dose of mestranol was >70 μg. The progestin component was considered high if the dose was equivalent to 0.30 mg of d,l -norgestrel or higher. This schema was the one used in the two previous papers addressing the issue of COC dose. 1 , 15 We also had six women who reported use of a progestin-only oral contraceptive and these women were included as COC users. The histologic type and grade of the ovarian cancers were obtained from the LAC SEER registry. The ovarian cancers were categorized into high (poor/undifferentiated) grade serous ovarian cancer (HGSOC), low grade serous (LGSOC), endometrioid, clear cell, mucinous or not known (including serous of unknown grade). Odds ratios and corresponding 95% confidence intervals (CIs) were calculated as estimates of the relative risks (RRs), and are presented as RRs. All statistical significance values ( p values) quoted are two-sided.

Parity and COC use are well-established protective factors for ovarian cancer, but as we move toward a model of personalized prevention it is increasingly important to gain a more in-depth understanding of these and other lifestyle factors associated with risk. In the analysis presented here, the timing of both parity and COC use influence the protection against ovarian cancer afforded by these exposures; a later age at first birth and an earlier age at COC use provide greater benefit. The reduction in risk we observed with increasing parity, a relative risk of 0.87 per birth, is in good agreement with the results of previous studies. 3 – 9 We found an overall reduction in relative risk of ovarian cancer of 0.87 per 5 years later age at first birth. One possible explanation for this protective effect of a late first birth that we observed is that the underlying reason for a late age is related to a woman’s underlying ovarian cancer risk. We examined the characteristics, including a history of infertility, endometriosis and SES, of the women who had late births and could not find a compelling explanation. If the underlying mechanism of the protective effect of a birth is to reduce cell proliferation one would expect that age at the birth would have no effect. 16 If a birth results in a reduction of a stem cell pool, an early age at birth should provide greater protection, as is seen in breast cancer. 16 None of these explanations seems to fit with our finding. Previous studies on the effect of age at first and last births showed associations with late age at both first and last births. 3 – 9 We also found a reduction in risk with increasing age at last birth—relative risk of 0.90 per 5 years later. Five of the studies shown in Table 6 also reported on the association of ovarian cancer risk with age at last birth. 4 – 8 In three of them, 4 , 5 , 8 the reduced risk with age at last birth was clearly greater than the reduction in risk with first birth. We found that the effect of age at last birth was eliminated after adjustment for age at first birth ( Table 1 ), but Whiteman et al. 8 found the reverse. Considering the results of all the studies of birth timing shown in Table 6 together with the results we found, we can conclude that a late birth has a clear additional protective effect against ovarian cancer over and above the protective effect of number of births. If a birth results in a reduction of precursor lesions, a late age at birth would provide greater protection, as is seen in endometrial cancer. 18 We found that the protective effects of parity and late age at first birth diminished with increasing age at diagnosis ( Table 5 ). This reduction in protection with increasing age at diagnosis has been consistently found in previous studies, although these studies expressed this in terms of increased time since first/last birth. Age at diagnosis and time since first/last birth are, of course, positively correlated, and in our data the results were much clearer when expressed in terms of age at birth and age at diagnosis. The attenuation over time is readily explained by models based on ovulatory years; the years pregnant as a proportion of ovulatory years plus time since menopause is reduced with increasing age, 16 but as Adami et al. 3 pointed out, the effect of age at births cannot be readily explained on the basis of models of incessant ovulation or ovulatory years. The important conclusion to be drawn from the results, when all studies are considered together, is that parity and age at births are both important protective factors and that the protection extends over 30 or more years. We found, as has been repeatedly found before, 10 a clear protective effect of COC use against ovarian cancer. The surprising aspect of our finding is that earlier age at use of COCs is associated with greater protection. We investigated whether this was driven by temporal changes in the doses of COCs, with early age at use potentially being associated with higher dose COCs, but this was not the case. This finding has important implications with respect to risk reduction given that using COCs at a young age as a preventive strategy is appealing given that the absolute risk of breast cancer at a young age is low. Thus, the transient increased risk of breast cancer associated with COC use 11 would be less of a concern. It is important to note, however, that in The Collaborative Group on Epidemiological Studies of Ovarian Cancer major overview of epidemiological results on ovarian cancer and COC use, 10 no ““index of timing of use … had any material effec”” on the reduction in risk of ovarian cancer with COC use, so our finding of an attenuation of protective effect with COC use after around age 35 should be regarded as tentative and needs further investigation in other data sets. Overall, it appears that the age at births has an effect on the degree of protection afforded by parity. Achieving a better understanding of the effect of births on endometriotic lesions, as the site of origin of endometrioid and clear cell ovarian cancers, and on the fallopian tube as the major site of HGSOC are needed to provide the information needed to develop prevention strategies based on this finding.