Association Between Breastfeeding and Ovarian Cancer Risk

Key Points

Question  Is breastfeeding associated with risk of ovarian cancer overall and by histotype?

Findings  In this pooled analysis including 9973 women with ovarian cancer and 13 843 controls from 13 case-control studies, breastfeeding was associated with a 24% reduced risk of invasive epithelial ovarian cancer. Longer breastfeeding duration and shorter time since last breastfeeding episode were associated with a further decrease in risk.

Meaning  This large study with extensive information on breastfeeding provides epidemiological evidence that breastfeeding, a potentially modifiable factor, may confer significant reduction in ovarian cancer risk, including high-grade serous, the deadliest subtype.

Importance  Breastfeeding has been associated with a reduced risk of epithelial ovarian cancer in multiple studies, but others showed no association. Whether risk reduction extends beyond that provided by pregnancy alone or differs by histotype is unclear. Furthermore, the observed associations between duration and timing of breastfeeding with ovarian cancer risk have been inconsistent.

Objective  To determine the association between breastfeeding (ie, ever/never, duration, timing) and ovarian cancer risk overall and by histotype.

Design, Setting, and Participants  A pooled analysis of parous women with ovarian cancer and controls from 13 case-control studies participating in the Ovarian Cancer Association Consortium was performed. Odds ratios (ORs) and 95% CIs of the overall association were calculated using multivariable logistic regression and polytomous logistic regression for histotype-specific associations. All data were collected from individual sites from November 1989 to December 2009, and analysis took place from September 2017 to July 2019.

Exposures  Data on breastfeeding history, including duration per child breastfed, age at first and last breastfeeding, and years since last breastfeeding were collected by questionnaire or interview and was harmonized across studies.

Main Outcomes and Measures  Diagnosis of epithelial ovarian cancer.

Results  A total of 9973 women with ovarian cancer (mean [SD] age, 57.4 [11.1] years) and 13 843 controls (mean [SD] age, 56.4 [11.7] years) were included. Breastfeeding was associated with a 24% lower risk of invasive ovarian cancer (odds ratio [OR], 0.76; 95% CI, 0.71-0.80). Independent of parity, ever having breastfed was associated with reduction in risk of all invasive ovarian cancers, particularly high-grade serous and endometrioid cancers. For a single breastfeeding episode, mean breastfeeding duration of 1 to 3 months was associated with 18% lower risk (OR, 0.82; 95% CI, 0.76-0.88), and breastfeeding for 12 or more months was associated with a 34% lower risk (OR, 0.66; 95% CI, 0.58-0.75). More recent breastfeeding was associated with a reduction in risk (OR, 0.56; 95% CI, 0.47-0.66 for <10 years) that persisted for decades (OR, 0.83; 95% CI, 0.77-0.90 for ≥30 years; P for trend = .02).

Conclusions and Relevance  Breastfeeding is associated with a significant decrease in risk of ovarian cancer overall and for the high-grade serous subtype, the most lethal type of ovarian cancer. The findings suggest that breastfeeding is a potentially modifiable factor that may lower risk of ovarian cancer independent of pregnancy alone.

Ovarian cancer survival remains poor with 5-year survival less than 50%, mostly owing to late detection.¹ Prevention is crucial for reducing mortality from this disease. With few modifiable risk factors beyond oral contraceptive (OC) use,^1,2 identifying additional modifiable factors is needed to tailor prevention strategies.

Numerous studies have investigated the association between breastfeeding and ovarian cancer risk, with some showing a significant decrease in risk and others showing no association,^3-31 leading the World Cancer Research Fund International to describe evidence of the association as limited.² Furthermore, relationships between ovarian cancer and breastfeeding patterns, including number of breastfeeding episodes, breastfeeding duration, and timing of breastfeeding, have been inconsistent. Prior studies have had insufficient sample size to evaluate heterogeneity by disease histotype. More than 90% of ovarian cancers are epithelial in origin,¹ with the major histotypes exhibiting varied risk profiles²⁹ and likely having distinct etiologic pathways. Therefore, we evaluated associations between breastfeeding and epithelial ovarian cancer risk overall and by histotype using data from 13 case-control studies participating in the Ovarian Cancer Association Consortium.

The Ovarian Cancer Association Consortium was established in 2005 to promote collaborative research on epidemiologic factors associated with ovarian cancer. Our analysis includes data from 13 Ovarian Cancer Association Consortium studies^32-44 with information on breastfeeding history (eTable 1 in Supplement). All studies are population-based and used in-person interviews for data collection except the Australian Ovarian Cancer Study³⁷ and the German Ovarian Cancer Study,⁴¹ which used self-administered questionnaire for data collection. All studies frequency matched by age categories. The Hawaii Ovarian Cancer Case-Control Study,³⁴ Family Registry for Ovarian Cancer and Genetic Epidemiology of Ovarian Cancer study,³⁶ and University of Southern California Study of Lifestyle and Women’s Health⁴³ additionally matched for race. The New Jersey Ovarian Cancer Study³² did not have any matching variables. All participants provided informed consent, and participating institutions obtained approval from relevant ethics committees. We excluded women who were nulliparous (n = 6309), missing parity information (n = 233), missing breastfeeding status (n = 480), and had nonepithelial tumors (n = 115). All data were collected from individual sites from November 1989 to December 2009.

Detailed information on breastfeeding history for each pregnancy lasting 6 months or longer, including ever breastfeeding (yes/no), breastfeeding duration (months), and age at each pregnancy was self-reported. Ever breastfeeding was defined as any breastfeeding, regardless of duration. A single breastfeeding episode refers to breastfeeding offspring from a given pregnancy (including twins and multiples). Total duration of breastfeeding was calculated by summing durations of individual breastfeeding episodes. If duration of any breastfeeding episode was unknown, total duration was considered unknown (n = 51). Mean breastfeeding duration per episode was obtained by dividing total duration by number of breastfeeding episodes. Timing of supplementation by formula, milk, or other foods was assessed in 5 studies (Connecticut Ovarian Cancer Study,³⁸ Diseases of the Ovary and their Evaluation,⁴⁰ Hawaii Ovarian Cancer Case-Control Study,³⁴ North Carolina Ovarian Cancer Study,⁴⁴ and University of Southern California Study of Lifestyle and Women’s Health⁴³), although specific wording varied by sites. To define the duration of exclusive breastfeeding consistently across studies, we calculated the duration as the time between birth and initial supplementation based on these questions. Analyses of ages at first and last breastfeeding episode were restricted to girls and women 15 years or older at first or last breastfeeding episode. Other relevant variables include age at diagnosis/interview, decade of participant’s birth, self-reported race, attained education, total duration of OC use, parity, history of endometriosis, tubal ligation, menopausal status, young adult body mass index (calculated as weight in kilograms divided by height in meters squared), and family history of ovarian cancer.

We used unconditional logistic regression to estimate the odds ratios (ORs) and 95% CIs for associations between breastfeeding and ovarian cancer risk for each site. We used indicator variables to account for missing data (race/ethnicity, n = 881; OC use, n = 264). In the main multivariate model, we adjusted for age (continuous), race/ethnicity (white, black, Asian, other/unknown), OC use (never, <1 year, 1 to <5 years, 5 to <10 years, ≥10 years, unknown), parity (continuous), and decade of birth (to account for possible changes in breastfeeding practices over time). We also considered adjustments for additional ovarian cancer risk factors including education (>high school, high school, some college, college, unknown), tubal ligation (no, yes, unknown), endometriosis (no, yes, unknown), body mass index at young adulthood (<20, 20 to <25, 25 to <30, ≥30, unknown), and family history of ovarian cancer (no, yes, unknown). However, these factors did not change the association between breastfeeding and ovarian cancer risk more than 10% and were not included in final models. We used random-effects meta-analyses of invasive cancers and borderline tumors and likelihood ratio tests to assess heterogeneity of the associations between study sites. Because we observed no significant heterogeneity, we used the pooled data set for all analyses adjusted for study site. We performed polytomous logistic regression to evaluate the associations between breastfeeding and cancer risk by invasive (low-grade serous, high-grade serous, mucinous, endometrioid, clear cell, other) and borderline (serous, mucinous) histotypes and likelihood ratio tests to evaluate differences in the associations by histotype. Among invasive serous tumors, low grade was defined as grade of 1 and high grade as grade of 2 or higher.

Because parity and breastfeeding are highly correlated, we created variables to evaluate each possible combination of parity and breastfeeding (eg, 1 liveborn with no breastfeeding, 1 liveborn with breastfeeding) to evaluate the independent association of breastfeeding with outcomes. We performed stratified analyses by age (≤50 years, >50 years), body mass index at young adulthood (<25, 25 to <30, ≥30), history of endometriosis, family history of ovarian cancer, and race/ethnicity. To evaluate interactions, we created crossproduct terms involving breastfeeding and the potential effect modifier and calculated likelihood-ratio statistics comparing models with and without the interaction terms. We tested for trend using Wald tests of continuous variables (months of breastfeeding, number of breastfeeding episodes, ages at first and last breastfeeding episodes, time since last breastfeeding episode) among women who reported ever having breastfed. Additionally, we conducted a sensitivity analysis using residuals to examine the association of breastfeeding patterns independent of parity. We computed residuals from a linear regression model with parity as the independent variable and number of breastfeeding episodes/total duration of breastfeeding as dependent variables. Residuals of breastfeeding episodes or total breastfeeding duration were next grouped into quintiles and used as the exposure of interest in multivariate models. To address potential confounding by parity, we restricted the analysis to women with exactly 2 births and examined whether breastfeeding both vs 1 child was associated with differential risk, adjusting for total duration of breastfeeding.

When examining the associations between ages at first and last breastfeeding episodes and ovarian cancer risk, we additionally adjusted for total breastfeeding duration and ages at first or last birth, respectively. We examined whether ages at first and last breastfeeding were independently associated with risk by simultaneously including both variables in the model. We additionally adjusted for age at last birth when examining the association between time since last breastfeeding and ovarian cancer risk, and we performed the duration and timing analyses restricted to high-grade serous histotype.

We used SAS version 9.4 (SAS Institute) for logistic regression analyses. For polytomous logistic regression, we used Stata/IC, version 15.1 (StataCorp). All P values are 2-sided, with .05 as the significance threshold. Analysis was performed from September 2017 to July 2019.

A total of 9973 women with ovarian cancer and 13 843 controls were included in the analysis. Among parous controls, the prevalence of ever having breastfed ranged from 41% to 93% across studies, and mean breastfeeding duration per episode ranged from 3.4 to 8.7 months (eTable 1 in the Supplement). The mean (SD) age was 57.4 (11.1) years for women with ovarian cancer and 56.4 (11.7) years for controls, and 21 103 participants (89%) identified as white (eTable 2 in the Supplement). Compared with controls, those with ovarian cancer were older, more likely to be postmenopausal, primiparous, have never used OCs, have a history of endometriosis, and have a family history of ovarian cancer.

After multivariable adjustment, ever having breastfed was inversely associated with risks of both invasive (OR, 0.76; 95% CI, 0.71-0.80) and borderline (OR, 0.72; 95% CI, 0.64-0.81) tumors compared with never breastfeeding (Table 1). Associations were consistent across study sites for both invasive and borderline tumors (Figure). Among invasive tumors, the association was statistically significant for high-grade serous (OR, 0.75; 95% CI, 0.70-0.81), endometrioid (OR, 0.73; 95% CI, 0.64-0.84), and clear cell tumors (OR, 0.78; 95% CI, 0.64-0.96). We observed similar, although not statistically significant, associations for low-grade serous and no significant inverse association for mucinous tumors. For borderline tumors, there was a statistically significant inverse association for both mucinous (OR, 0.68; 95% CI, 0.57-0.82) and serous tumors (OR, 0.77; 95% CI, 0.66-0.89). In the 5 studies^{34,38,40,43,44} with data on exclusive breastfeeding, we observed a similar inverse association with ovarian cancer risk for women who breastfed exclusively for at least 3 months (OR, 0.81; 95% CI, 0.72-0.92) and breastfed but not exclusively for 3 months (OR, 0.70; 95% CI, 0.62-0.79) compared with women who never breastfed.

To examine the association between breastfeeding and ovarian cancer risk within parity strata, we created a crossclassified variable of parity and breastfeeding (eFigure in the Supplement). Among primiparous women, ever breastfeeding was associated with 14% lower risk of all invasive disease (OR, 0.86; 95% CI, 0.75-0.99) and 16% lower risk of high-grade serous tumors (OR, 0.84; 95% CI, 0.71-0.99) compared with never breastfeeding. For multiparous women, we observed a similar inverse association for both all invasive and high-grade serous tumors.

In stratified analyses, we observed no statistically significant effect modification by age, body mass index at young adulthood, history of endometriosis, and family history of ovarian cancer (eTable 3 in the Supplement). Furthermore, we observed significant interaction by race/ethnicity (P for interaction = .01), with the strongest association in white women (OR, 0.73; 95% CI, 0.69-0.79) and no significant association among black women (OR, 0.92; 95% CI, 0.66-1.29) or Asian women (OR, 0.81; 95% CI, 0.58-1.12).

Compared with women who never breastfed, longer mean duration of breastfeeding per episode was inversely associated with invasive ovarian cancer risk (Table 2). Mean breastfeeding duration of less than 3 months per child was associated with 18% lower risk (OR, 0.82; 95% CI, 0.76-0.88) while mean breastfeeding duration of more than 12 months per child was associated with 34% lower risk (OR, 0.66; 95% CI, 0.58-0.75; P for trend = .001). Similar associations were observed in analyses restricted to high-grade serous tumors (Table 2) and in regression analyses of residuals controlled for parity (data not shown). In addition, we stratified mean breastfeeding duration per episode by parity (1, 2, and ≥3 live births) and observed associations similar in magnitude (Table 2). Similarly, total breastfeeding duration was significantly associated with reduced ovarian cancer risk (P for trend <.001 among women who breastfed; data not shown).

Among women who breastfed, older ages at first (OR, 0.90; 95% CI, 0.85-0.96, per 5 years of age increase at first breastfeeding; P for trend = .001) and last breastfeeding episode (OR, 0.94; 95% CI, 0.89-0.99, per 5 years of age increase at last breastfeeding; P for trend = .02) were associated with lower risk of invasive ovarian cancer (eTable 4 in the Supplement). When adjusted for ages at first and last birth together, both associations were attenuated. Breastfeeding was associated with a substantial reduction in ovarian cancer risk for decades after the last breastfeeding episode, although the magnitude of risk reduction attenuated over time. Compared with women who never breastfed, time since last breastfeeding episode of less than 10 years was associated with 44% lower risk (OR, 0.56; 95% CI, 0.47-0.66), while time since last breastfeeding episode of 30 years or less was associated with 17% lower risk (OR, 0.83; 95% CI, 0.77-0.90; P for trend = .02). Associations for age at breastfeeding and timing of breastfeeding were similar for high-grade serous tumors (eTable 4 in the Supplement) and when restricted to women with 2 births (eTable 5 in the Supplement).

In this largest study of breastfeeding and ovarian cancer risk to date and to our knowledge, including 9973 parous women with ovarian cancer from 13 studies, a history of breastfeeding was associated with 24% decrease in invasive cancer risk and 28% decrease in borderline tumor risk, with consistent risk reduction across study sites. Furthermore, mean breastfeeding duration per episode was inversely associated with risk, with 18% reduction in risk for women who breastfed less than 3 months per live birth and 34% reduction for women who breastfed 12 months or longer per live birth. Similar associations were observed for high-grade serous tumors, the deadliest ovarian cancer histotype.

Consistent with some but not all prior studies,^{3-15,17,20,22,24,26,27,29,30,44-55} we observed an inverse association between mean breastfeeding duration per episode and risk of invasive and borderline ovarian tumors. Statistically significant risk reduction associated with a mean breastfeeding duration of less than 3 months per episode suggests even a short duration of breastfeeding is beneficial. We did not observe a statistically significant difference in risk associated with nonexclusive breastfeeding compared with exclusive breastfeeding; however, sample size was limited and exclusive breastfeeding definitions varied between studies.⁵⁶ Associations between breastfeeding and ovarian cancer risk were stronger with older ages at first and last breastfeeding episodes, consistent with previous reports.^57-59 However, these associations were attenuated after mutual adjustment for ages at first and last breastfeeding. While risk reduction was greatest for women who had breastfed within the past 10 years, the association persisted more than 30 years, a pattern also seen with time since last OC use and last birth.⁵⁹

Our findings indicate that breastfeeding is associated with lower risk of high-grade serous cancers, the most common and fatal subtype. While previous studies^4,5,29,46,47 reported inconsistent findings for serous histotypes, most did not separate low- from high-grade serous tumors, which are thought to have separate etiologic pathways.^29,60 The Million Women Study⁵ separately analyzed low- and high-grade serous tumors and reported a lower risk of both histotypes with longer total and mean duration per episode of breastfeeding. However, the associations did not reach statistical significance. Similar to ovarian cancer overall, we observed a reduced risk of high-grade serous tumors with as little as 3 months’ mean breastfeeding duration per episode and greater risk reduction with longer breastfeeding duration, older ages at first and last breastfeeding episodes, and recent (<10 years) breastfeeding. The inverse association with time since last breastfeeding persisted for more than 30 years. While we observed similar associations with breastfeeding patterns and endometrioid ovarian cancers, fewer cases of other invasive histotypes limited our ability to evaluate them with the same level of detail.

Biological mechanisms through which breastfeeding could reduce ovarian cancer risk are not well understood. To date, the leading hypothesis has been that ovulation suppression during breastfeeding inhibits epithelial cell division and proliferation, thereby reducing the opportunity to initiate or promote carcinogenesis. This may especially be pertinent in the first few months post partum when immune function and tumor surveillance mechanisms remain suppressed.⁶¹ However, we observed a stronger inverse association with longer breastfeeding duration, suggesting anovulation cannot entirely explain the association because ovulation typically returns once solids are introduced. Several lines of evidence suggest that breastfeeding may also be associated with long-term modulation of inflammatory, immune, or metabolic pathways, which could influence ovarian cancer risk.^62,63

Important strengths of this study include detailed data on breastfeeding patterns, including information on duration and timing of each breastfeeding episode. We included a large number of women with ovarian cancer and controls from established studies with geographic diversity and extensive data on demographic, reproductive, and lifestyle variables. The large sample size enabled us to examine breastfeeding patterns in detail and specifically to assess high-grade serous tumors, the most common and fatal histotype. In addition, our pooled analysis allowed us to harmonize variables for consistency across the 13 studies, adjust for a single set of confounders, evaluate histotype-specific associations, and assess potential effect modification in stratified analyses. Importantly, we were also able to disentangle the association between breastfeeding and ovarian cancer risk from that of parity.

Our study has some limitations, including the potential for differential self-reporting of breastfeeding information by disease status, which could distort the magnitude of associations. Our results may also be influenced by selection bias as controls participating in these studies may differ from women with ovarian cancer by factors influencing breastfeeding or ovarian cancer risk, including unknown factors that could not be accounted for in the analysis. Despite adjustment for potential confounders, unmeasured confounding may persist but is unlikely to substantively alter our findings given our robust associations. Validation in prospective cohorts is needed to address these concerns. Finally, because our study population included predominantly white women, we could not sufficiently evaluate details of breastfeeding patterns in black women, Asian women, and less common racial and ethnic groups.²⁰ The association between breastfeeding and ovarian cancer needs to be investigated in large populations of other races and ethnicities.

In conclusion, breastfeeding is associated with a significant decrease in ovarian cancer risk overall and for high-grade serous tumors, the most lethal subtype. The World Health Organization recommends exclusive breastfeeding for at least 6 months and continued breastfeeding with complementary foods for 2 or more years.^2,64 Our results support these recommendations, while noting that breastfeeding fewer than 3 months per child is still associated with significant ovarian cancer risk reduction.

Corresponding Author: Naoko Sasamoto, MD, MPH, Obstetrics and Gynecology Epidemiology Center, Brigham and Women’s Hospital, 221 Longwood Ave, Boston, MA 02115 (nsasamoto@bwh.harvard.edu).

Accepted for Publication: February 6, 2020.

Published Online: April 2, 2020. doi:10.1001/jamaoncol.2020.0421

Author Contributions: Dr Terry had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Drs Babic and Sasamoto contributed equally as co–first authors, and

Drs Modugno and Terry contributed equally as co–senior authors to this work.

Concept and design: Babic, Sasamoto, Moysich, Ness, McGuire, Webb, Modugno, Terry.

Acquisition, analysis, or interpretation of data: Babic, Sasamoto, Rosner, Tworoger, Jordan, Risch, Harris, Rossing, Doherty, Fortner, Chang-Claude, Goodman, Thompson, Moysich, Ness, Kjaer, Jensen, Schildkraut, Titus, Cramer, Bandera, Qin, Sieh, Sutphen, Pearce, Wu, Pike, Webb, Modugno, Terry.

Drafting of the manuscript: Babic, Sasamoto, Jordan, Pike, Modugno, Terry.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Babic, Sasamoto, Rosner, Tworoger, Jordan, Risch, Harris, Schildkraut, Pike, Modugno.

Obtained funding: Risch, Rossing, Doherty, Goodman, Ness, Kjaer, Schildkraut, Titus, Sutphen, Modugno, Terry.

Administrative, technical, or material support: Risch, Harris, Rossing, Fortner, Thompson, Jensen, Titus, Cramer, Sieh, Sutphen, Pearce, Wu, Webb, Modugno.

Supervision: Ness, Webb, Modugno, Terry.

Conflict of Interest Disclosures: Drs Babic, Sasamoto, Rosner, Tworoger, Cramer, Qin, Schildkraut, Titus, Pearce, Pike, Terry, and Modugno report grants from the National Institutes of Health during the conduct of the study. Dr Jordan reports grants from the National Health and Medical Research Council of Australia during the conduct of the study. Dr Fortner reports grants from the Federal Ministry of Education and Research of Germany during the conduct of the study. Dr Kjaer reports grants from the Danish Cancer Society and The Mermaid I Project during the conduct of the study. Drs Pearce, Webb, and Modugno report grants from Congressionally Directed Medical Research Programs and National Institutes of Health during the conduct of the study. No other disclosures were reported.

Funding/Support: The scientific development and funding for this project were in part supported by the US National Cancer Institute GAME-ON Post-GWAS Initiative (grant U19-CA148112).

Role of the Funder/Sponsor: The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: We thank all of the members of the Ovarian Cancer Association Consortium for their support of this collaborative network and 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. We also thank the funders for the individual studies analyzed in this article: The Australian Ovarian Cancer Study Group was supported by the Army Medical Research and Materiel Command, National Health and Medical Research Council of Australia, Cancer Councils of New South Wales, Victoria, Queensland, South Australia, and Tasmania, and Cancer Foundation of Western Australia. The Connecticut Ovarian Cancer Study, the Diseases of the Ovary and their Evaluation study, Hawaii Ovarian Cancer Case-Control Study, and Family Registry for Ovarian Cancer and Genetic Epidemiology of Ovarian Cancer study were supported by the National Institutes of Health. The German Ovarian Cancer Study was supported by the German Federal Ministry of Education and Research, Programme of Clinical Biomedical Research and the German Cancer Research Center. The Hormones and Ovarian Cancer Prediction Study was supported by the University of Pittsburgh School of Medicine Dean’s Faculty Advancement Award (Dr Modugno), US Department of Defense, and National Institutes of Health. The Malignant Ovarian Cancer Study was supported by the National Institutes of Health, Danish Cancer Society; and the Mermaid I Project. The North Carolina Ovarian Cancer Study was supported by the National Cancer Institute and US Department of Defense. The New England Case Control Study was supported by the National Institutes of Health and US Department of Defense. The New Jersey Ovarian Cancer Study was supported by the National Institutes of Health and the Rutgers Cancer Institute of New Jersey. The Southern Ontario Ovarian Cancer Study was supported by the National Health Research and Development Program, Canada. The University of Southern California Study of Lifestyle and Women’s Health study was supported by the National Institutes of Health and California Cancer Research Program.