Preterm delivery is associated with an increased risk of epithelial ovarian cancer among parous women.

Ovarian cancer is the most lethal gynecological malignancy. The incidence in the Nordic countries is among the highest in the world with 9.5 cases per 100,000 women-years. 1 The etiology is complex and only partly understood. Epithelial ovarian cancer is not one distinct disease, but rather several histologically and clinically distinct subtypes: serous (70-75%), mucinous (5-10%), clear cell (10%) and endometrioid (10%). The carcinogenesis of epithelial ovarian cancer has been described by a dualistic model, 2 with two pathways involving different precursor lesions. Mucinous, clear cell, endometrioid and low-grade serous tumors are thought to develop from transformation of implant cysts on the ovary and have a better prognosis, while high-grade serous tumors are thought to develop from premalignant lesions in the fallopian tube and have a poorer prognosis. 3 The risk of epithelial ovarian cancer is lower in parous women, with a protective effect across all epithelial subtypes. 4 – 9 Results are conflicting for other pregnancy-related factors such as pre- and postterm delivery, 10 – 12 maternal age at first and last birth, 4 , 7 , 13 – 17 birth weight, 10 , 14 preeclampsia 14 , 18 and multiple pregnancies. 14 , 19 , 20 In this large population-based Nordic study, we evaluated the risk of epithelial ovarian cancer and its specific histologic subtypes in parous women in relation to pregnancy and birth characteristics, particularly pregnancy length and maternal age at birth.

The study included 10,957 cases and 107,864 controls with a median age at ovarian cancer diagnosis/matching of 52 years (range 19-85 years) ( Table 1 ). Due to the later establishment of the Finnish birth registry, there were fewer cases from Finland and mean age at diagnosis was lower ( Table 2 ). Analysis by histological subtype was possible for 7,971 cases ( Table 2 ). Increasing number of births was inversely associated with the risk of epithelial ovarian cancer ( Table 3 ), with an additional risk reduction for each subsequent pregnancy. Preterm delivery in a woman’s last pregnancy was associated with an increased risk of ovarian cancer - the shorter the pregnancy, the stronger the association [pregnancy length ≤30 versus 39-41 weeks: adjusted OR 1.33 (95%CI 1.06-1.67)]. The OR was 0.98 (95%CI 0.97-0.99) for each added gestational week ( Figure 1 ). The results remained when we adjusted for age at first or last birth (i.e. OR for each added gestational week: 0.98 (95%CI 0.97-0.99) when adjusted for age at last birth in addition to number of births). The OR for preterm delivery overall, defined as <37 weeks of gestation, versus 37-42 weeks, was 1.16 (95%CI 1.07-1.26). Further adjustment for smoking had no impact on results (i.e., OR <37 pregnancy-weeks, versus 37-42 weeks: 1.16 (95%CI 1.07-1.26). Postterm pregnancies (≥42 weeks) compared with 39-41 weeks were not associated with risk ( Table 3 ). The OR for preterm delivery in the woman’s last pregnancy did not vary by number of births ( sTable 2 ). Results were similar for pregnancy length in the woman’s first pregnancy (data not shown). As described earlier, women who were diagnosed with ovarian cancer within six months after giving birth were excluded to minimize the possibility that associations were related to preterm delivery due to the cancer. Results were also similar if women diagnosed with cancer within one year after pregnancy were excluded. There were no associations between ovarian cancer and preeclampsia, multiple births, or offspring birth length, neither overall nor by subtype ( Table 3 ). Results did not change when the association of offspring birth length and weight and preeclampsia and ovarian cancer were adjusted for smoking. Older age at first and last birth and shorter time since first and last birth were all inversely associated with ovarian cancer risk ( Table 3 ). First birth at age 30-39, versus before 25 years, was associated with decreased risk [OR 0.76 (95%CI 0.70-0.83)]; similar results were seen with age at last birth [30-39 versus <25 years: OR 0.76 (95%CI 0.71-0.82)]. In addition, shorter time since first and last birth were associated with decreased risk ( Table 3 ). The OR for age at first and last birth and time since first and last birth did not vary by number of births ( sTable 2 ). Since age at first and last birth is the same among women with only one pregnancy, we also performed a sensitivity analysis where 1-para cases were excluded, which did not alter the results (i.e. per year increase in age at last birth: all cases: OR 0.98 (95% CI 0.97-0.98); >1 births: OR 0.98 (95% CI 0.97-0.98). Among cases with known histology, 5250 (66%) were serous, 1284 (16%) mucinous, 998 (12%) endometrioid, and 439 (6%) clear cell carcinomas ( Table 4 ). The inverse association observed with increasing number of births was strongest for the clear cell subtype [per birth: OR 0.74, 95%CI 0.67-0.81) p-het 40 years versus <25 years OR 0.39 (95%CI 0.24-0,65; p-het <0.001). Smoking was associated with mucinous ovarian cancer (OR 2.37 (95%CI 1.78-3.15); p-het 0.003). In addition, low offspring birth weight in the last pregnancy was associated with mucinous ovarian cancer [adjusted for pregnancy length and parity: 1500-2499g versus 2500-4500g OR 1.47 (95%CI 1.07-2.01) also when additionally adjusted for smoking: OR 1.46 (95%CI: 1.06-2.00)]. In a sensitivity analysis in which we restricted the mucinous group to cases diagnosed since 2005 (n=459), the results were similar. Higher offspring birth weight was associated with endometrioid ovarian cancer. [>4500g versus 2500-4500: OR 1.62 (95%CI 1.15-2.29), p - het 0.02] When stratifying on the mother’s age at diagnosis (<50, 50-59, and ≥60 years, sTable 3 ) 31 , 32 , each additional birth was associated with a 15% risk reduction for ovarian cancer in the premenopausal group, but only a 7% risk reduction in the oldest age category (≥60 years [ P-het 0.007]. The risk associations were similar among the age groups when stratified by the mother’s year of birth ( sTable 4 ).

In the largest study to date on epithelial ovarian cancer among parous women, we found that preterm delivery was associated with increased ovarian cancer risk. Higher age at both first and last birth were inversely associated with ovarian cancer risk, and the associations persisted regardless of a woman’s total number of births. While abortion (induced or spontaneous) does not seem to be associated with risk of ovarian cancer, 33 few previous studies have examined the associations with preterm delivery. Mucci et al. reported that preterm deliveries were associated with an increased risk of ovarian cancer in a cohort with 1017 cases (including a subset of the 3,673 Swedish cases in our study from year 1973-2001) with a mean age of 43 years at diagnosis. 10 The same research group also showed an increased risk among women with postterm delivery. 11 However, only 31 of 316 ovarian cancer patients in the study had a postterm delivery, and mean age at diagnosis was equally low. Preterm delivery in any pregnancy was associated with increased risk in an Australian case-control study of 1203 cases: 12 in that study only 74 cases experienced preterm delivery. We could examine the associations with preterm delivery in greater detail, categorized in intervals of 2-3 weeks and found that additional weeks of preterm delivery were associated with an increased risk. Preterm delivery (n=707 cases) was associated with increased risk in both younger (<50 years) and older (≥60 years) women at diagnosis. Mucci et al. 10 found the strongest associations with pregnancy length among women with one birth, while Jordan et al. 12 found this association among multiparous women. In our data, the increased risk associated with short length of a woman’s last pregnancy was present independent of number of births. Unlike previous studies, we could analyze the association between pregnancy length and ovarian cancer subtypes and found the same trend for all. Most previous studies have reported reduced ovarian cancer risk with higher age at first or last birth. 4 , 7 , 13 , 15 – 17 Four studies mutually adjusted for age at first and last birth with conflicting results. 7 , 15 – 17 Two studies, a pooled case-control study of 620 cases by Whiteman et al. 15 and a registry-based, cohort study of 1271 cases by Albrektsen et al. 7 (including a subset of the 4,223 Norwegian cases in our study from year 1967-1991), found that only age at last birth remained statistically significant after adjustment. In contrast, in a recently published pooled case-control study of 1632 cases, Wu et al. 17 found that only age at first birth remained significant after adjustment. Neither age at first or last birth was related materially to ovarian cancer risk when mutually adjusted in a smaller case-control study (n=563 cases). 16 Risk reductions from pregnancies are most pronounced during the first decade after last birth, 7 , 15 , 16 as our findings also indicate. An increasing number of births had a stronger effect on risk in younger than older women, which is also consistent with the difference in risk by time (younger women are more likely to have a shorter time since last birth). The association between number of births and histologic subtype differ, with the strongest inverse associations observed for clear cell and endometrioid subtypes 6 , 8 , 9 and weaker associations for serous and mucinous carcinomas. 6 , 8 , 9 , 34 Our results on only parous women are in line with studies comparing nulliparous to parous women, including the subtype specific results from the Ovarian Cancer Cohort Consortium. 6 Smoking was associated with an increased risk of mucinous carcinomas, as reported earlier. 35 , 36 For other risk factors, results by subtype were comparable. We had a higher proportion of the mucinous subtype than expected, especially before 2005. However, when we restricted the analysis to mucinous cases diagnosed from 2005 or applied stricter coding in the selection of mucinous tumors, the results remained unchanged. Any misclassification with gastrointestinal tumors in our study would be random with respect to the pregnancy factors we analyzed, resulting in bias towards the null. The mechanism underlying the reduced risks associated with pregnancies is not known, although there are several hypotheses. Based on our findings we favor “the cell clearance hypothesis”, whereby a full-term pregnancy provides protection through clearance of precancerous cells from the ovarian epithelium/fallopian tubes, mediated by placental or ovarian hormones 13 . Since hormone levels that mediate cell clearance increase during the last trimester of pregnancy, 37 our finding of increased risk with shorter pregnancy length is consistent with the cell clearance hypothesis. Moreover, since both increasing time since pregnancy and older age at pregnancy increase the risk of accumulating premalignant cells, shorter time since last pregnancy, and pregnancies at older ages (a more recent cell clearance), are associated with lower risk and these results are also consistent with the cell clearance hypothesis. Some preclinical data support cell clearance mediated by progesterone, which rises during pregnancy. 37 Administration of progesterone to a human ovarian carcinoma cell line increased the expression of TP53 tumor suppressor gene and induced apoptosis. 38 Further, a study of sheep ovarian cancer cells indicated that progesterone is involved in the restoration or apoptosis of ovarian surface epithelial cells containing damaged DNA after ovulation. 39 Synthetic progestin induces apoptosis in ovarian cancer cell lines, 40 and treatment with progestin induces apoptosis in the ovarian epithelium of macaques 41 , 42 and chickens. 43 Further, oral contraceptives with high-progestin potency seem to provide greater protection against ovarian cancer than low-progestin formulations. 44 However, the effects of synthetic progestin might differ from the effects of endogenous progesterone in pregnancy. The earlier most accepted hypothesis is the incessant ovulation hypothesis, 45 proposing that the disruption and repair of ovarian epithelium can cause mutations in ovarian epithelial cells, resulting in a higher risk of cancer. However, the risk reduction occurring with nine months of pregnancy is stronger than the risk reduction associated with nine months of missed ovulatory cycles due to other causes. 46 The association we found for increased risk with shorter length of pregnancy could have several causes, such as fewer ovulatory cycles, lack of hormone exposure in the last trimester, or shorter exposure to pregnancy hormones. However, the inverse association seen with increasing age at pregnancy cannot be explained only by reduction in ovulatory cycles, since pregnancy at any age would then be expected to have the same effect on risk. Thus, pregnancies likely provide protection through other mechanisms (such as cell clearance), rather than solely reduction in ovulatory cycles. This large population-based study has a number of notable strengths: 1) almost complete information on reproductive history collected from birth registries avoided recall bias; 2) cancer diagnoses were collected in a similar and standardized manner from nationwide registries in four Nordic countries; 1 and 3) the study’s large size (larger than most previous studies) provided a sufficient number of cases for disaggregation by subtype and facilitated more precise risk estimates. Because the analysis was restricted to parous women (pregnancy characteristics were the exposures of interest), the cancer cases were relatively young with a mean age at diagnosis of 52 years. However, our findings are consistent with cohort studies that included primarily sporadic ovarian cancers in postmenopausal women. 5 , 6 A study weakness was lack of information on possible confounders, most notably use of oral contraceptives. Oral contraceptives were introduced in the Nordic countries during the 1960s; use has since increased and is most frequent among younger women. 47 , 48 This suggests that women born before 1940 are less likely to have used oral contraceptives. Thus, since our results did not differ by birth year of the mother, they are unlikely to be affected by oral contraceptive use. In addition, adjusting for oral contraceptive use in previous studies had no major impact on estimates of parity and number of births 5 , 6 , 15 or age at birth. 17 We also lacked information on oophorectomy/salpingectomy or hysterectomy, and thus could not exclude controls who had undergone gynecological surgery. However, since the prevalence of oophorectomy/hysterectomy in the Nordic countries is low (200 per 100.000 women), this is unlikely to have had a major effect on our risk estimates. 49 Moreover, we lacked information on other potential confounders such as age at menarche, body mass index, and history of endometriosis. Finally, the influence of pregnancy loss (prior to week 22) could not be investigated. In this large population based case-control study including over 10,000 epithelial ovarian cancer cases, we found that preterm delivery was associated with an increased risk of ovarian cancer. Older age at both first and last birth was associated with a lower risk of ovarian cancer. Our findings are in line with the cell clearance hypothesis, which might help us to better understand this fatal disease. Learning how pregnancy removes precancerous cells could possibly provide a clinical opportunity for prevention of ovarian cancer.

This case-control study was undertaken as a Nordic collaboration, using linked data from two nationwide population-based registries: Medical birth registries (MBR) were founded in 1973 (Denmark), 1987 (Finland), 1967 (Norway), and 1973 (Sweden). They contain information from antenatal, obstetric, and neonatal medical records and are based on mandatory reporting of all births. 21 – 25 The attending physician/midwife completes a standardized form shortly after birth, documenting maternal health and pregnancy characteristics prior to, during, and at birth, and selected fetal outcomes. Cancer registries, founded in 1943 (Denmark), 1953 (Finland and Norway), and 1958 (Sweden), are based on compulsory reporting of all incident cancer cases, including date of diagnosis and tumor histology. Case reporting is essentially complete. 1 , 26 – 28 Medical birth registries (MBR) were founded in 1973 (Denmark), 1987 (Finland), 1967 (Norway), and 1973 (Sweden). They contain information from antenatal, obstetric, and neonatal medical records and are based on mandatory reporting of all births. 21 – 25 The attending physician/midwife completes a standardized form shortly after birth, documenting maternal health and pregnancy characteristics prior to, during, and at birth, and selected fetal outcomes. Cancer registries, founded in 1943 (Denmark), 1953 (Finland and Norway), and 1958 (Sweden), are based on compulsory reporting of all incident cancer cases, including date of diagnosis and tumor histology. Case reporting is essentially complete. 1 , 26 – 28 Record linkage across these registries is possible through the personal identification number assigned to each citizen and permanent resident in the Nordic countries at birth or upon immigration. We included all parous women registered in the MBRs at childbirth ( Table 1 ), who had a subsequent diagnosis of epithelial ovarian cancer recorded in a cancer registry during 1973-2011 in Denmark (n=2494), 1987-2012 in Finland (n=567), 1967-2013 in Norway (n=4223), and 1973-2013 in Sweden (n=3673) ( Table 2 ). Cases were free of other cancers at time of inclusion. Information on every pregnancy for each woman was included in the linkage to the MBR; however, since many women had had pregnancies prior to the start of the MBR, data was most complete for the most recent pregnancy before the case’s date of diagnosis. Ovarian cancer was defined by ICD-10/ICD-O-3 code C56.9, and by ICD-7 code 175.0 in Denmark before 1978 and in Sweden before 1993, and to identify epithelial ovarian cancer we used ICD-O-2/ICD-O-3 code 8010-8231, 8246-8576, 9014-9015 and 9110 and WHO/HS/CAN/C24.1 codes 096, 196 and 146 according to the International Agency for Research on Cancer’s 29 definition ( sTable 1 ). For each case, we sampled up to ten female controls that were alive and cancer-free at the time of the cases’ diagnosis and who were registered in a MBR with a prior pregnancy lasting longer than 22 weeks. Controls were matched by country and the case´s year of birth. We examined the following exposures: Number of births at the time of matching; age at first and last birth; time since first and last birth; preeclampsia and multiple pregnancy (in any pregnancy). Pregnancy length in completed weeks and birth length and weight of the offspring were ascertained from the cases’ and controls’ most recent pregnancy before the case’s date of diagnosis. Many of the cases’ first pregnancies occurred before the MBRs started, thus, the most recent pregnancy provided the most complete data. Information on smoking habits during (any) pregnancy was available in Denmark 1991-2010, Finland 1987-2012, Norway 1998-2013, and Sweden 1982-2013. We used conditional logistic regression (conditioned on birth year of the case and country) to estimate the impact of pregnancy-related factors. Odds ratios (ORs) with 95% confidence intervals (CIs) were computed for epithelial ovarian cancer by histologic subtype. In analyses of pregnancy length and birth length/weight of offspring, women who were diagnosed with ovarian cancer within six months after giving birth were excluded (92 cases, 702 controls), to minimize the possibility that associations were related to preterm delivery due to the cancer. We performed analysis adjusted for number of births. We also performed analyses stratified by number of births. Offspring length/weight were adjusted for pregnancy length (as a continuous variable), as a measure of fetal growth. We stratified analyses by birth year (<1940, 1940-1959, and ⩾1960) to evaluate differences over time and during time periods reflecting changing use of oral contraceptives, and by three age categories, <50 years (as a proxy for premenopausal status), 50-59 years and ⩾60 years. Since coding of mucinous ovarian cancers has changed over time, we conducted a sensitivity analysis evaluating cases diagnosed since 2005. When different patterns of associations were observed, we tested for heterogeneity ( p-het ) of associations over strata in stratified analyses, using a likelihood ratio test. R version 3.3.2 was used for all analyses. 30 We obtained approval from Ethics Committees in Norway and Sweden, from the Data Protection Agency in Denmark, and from the National Institute for Health and Welfare in Finland after consulting its Data Protection Authority.