Urinary PGE-M Levels and Risk of Ovarian Cancer Free

Chronic inflammation may contribute to the etiology of epithelial ovarian cancer. There is increasing evidence that regular use of NSAID is associated with a modestly lower risk of ovarian cancer; however, the underlying biologic mechanisms remain poorly understood (1–7). Aspirin and nonaspirin NSAIDs downregulate the prostaglandin synthesis pathway via inhibition of the cyclooxygenase (COX) enzymes (8–11). COX1 and COX2 are overexpressed in ovarian tumor tissue relative to normal tissue (12–15), and greater COX1 and COX2 expression have been associated with poorer prognosis (15–19), suggesting that NSAIDs may influence risk through the prostaglandin pathway.

COX1 and COX2 promote the conversion of arachidonic acid into bioactive prostaglandins, the most abundant of which is prostaglandin E2 (PGE2; refs. 20, 21). While endogenous levels of circulating PGE2 cannot be reliably measured in humans (22, 23), prior studies of prostaglandins and cancer have measured urinary 11 alpha-hydroxy,9,15-dioxo-2,3,4,5-tetranor-prostane-1,20-dioic acid (PGE-M), the primary urinary metabolite of PGE2, to approximate systemic levels of prostaglandins (22, 24). In these studies, PGE-M was associated with an increased risk of colorectal cancer (25, 26), colorectal adenoma (27), gastric cancer (28, 29), small-cell lung cancer (30), pancreatic cancer (31, 32), and postmenopausal breast cancer (33, 34).

Here, we evaluated the association between prediagnosis urinary PGE-M levels and ovarian cancer risk in a case–control study nested in three prospective cohort studies, the Nurses' Health Study (NHS), NHSII, and the Shanghai Women's Health Study (SWHS). We hypothesized that higher PGE-M levels would be associated with increased ovarian cancer risk.

The NHS was established in 1976 with the enrollment of 121,700 female registered nurses, ages 30–55. The NHSII was established in 1989 with the enrollment of 116,429 female registered nurses, ages 25–42. NHS/NHSII participants completed a questionnaire on lifestyle factors, medication use, and disease status at the time of enrollment, and provided updated lifestyle and health information biennially, by mailed questionnaire. Cases of epithelial ovarian cancer were identified on return of biennial questionnaires or via linkage to the National Death Index. An expert gynecologic pathologist confirmed cases by medical record review and, when records were not available, cases were confirmed by linkage to cancer registries. Tumor behavior and histopathology characteristics were abstracted for confirmed cases, and tumor histopathology was confirmed by slide review for all cases with available tumor blocks.

A subset of women in NHS/NHSII provided a urine sample (described in refs. 35, 36). In brief, 18,743 NHS participants ages 53–80 sent spot urine specimens by overnight mail (with an icepack) between 2000 and 2002 (93% first morning urine), where it was aliquoted without a preservative and stored in liquid nitrogen freezers at ≤−130°C. Similarly, 29,611 NHSII participants ages 32–54 years provided urine specimens between 1996 and 1999 (80% first morning urine). Of these, 18,521 premenopausal women provided a urine sample 7–9 days prior to the anticipated start of their next menstrual cycle (luteal phase). The other 11,090 participants provided an untimed specimen. All participants completed a biospecimen collection questionnaire asking about time of urine collection, whether it was a first morning urine, medication use, and body weight, among other characteristics. NHS/NHSII study protocols were approved by the Institutional Review Boards of the Brigham and Women's Hospital and Harvard T.H. Chan School of Public Health (Boston, MA), and those of participating registries as required. Research was conducted in accordance with the Belmont report. Participant return of self-administered questionnaires was accepted as informed consent.

The SWHS conducted baseline interviews from 1996 to 2000, capturing data from 74,942 Chinese women, ages 40–70, living in urban communities in Shanghai. Women were approached by a trained interviewer and, after providing informed consent, completed a self-administered questionnaire and in-person interview to collect data on lifestyle factors, medication use, and disease outcomes. In-person follow-up surveys have been conducted every 2–6 years to obtain information on lifestyle factors and disease outcomes. Ovarian cancer cases were identified with a combination of record linkage to the Shanghai Cancer Registry or Shanghai Vital Statistics Unit and in-person follow-up surveys. Diagnoses were confirmed by medical record review and histologic type was assigned using International Classification of Diseases for Oncology codes.

Between 1997 and 2000, 65,754 SWHS women provided an untimed spot urine sample, as described previously (37). In brief, each participant answered questions related to urine collection and provided a urine sample using a sterilized 100 mL cup with 125 mg of ascorbic acid. The sample was transported to the laboratory, with an ice pack, within 6 hours of collection and stored at ≤−70°C.

The institutional review boards of all relevant institutions in the United States and the P.R. China approved the SWHS study protocol, and subjects provided written informed consent. Research was conducted in accordance with the Belmont report.

Two controls were matched to each case using incidence density sampling, within cohort, on year of birth (±1 year for NHS/NHSII; ±2 years for SWHS), date (±1 month), and time (±2 hours for NHS/NHSII; morning vs. afternoon for SWHS) of collection, and menopausal status (premenopausal, postmenopausal, and unknown). NHS/NHSII additionally matched on menopausal status at diagnosis (premenopausal, postmenopausal, and unknown), hormone therapy use at collection (yes/no), and luteal day (NHSII women only; ±1 day); NHS cases diagnosed before 2004 were not matched on time of day or hormone therapy. SWHS was also matched on antibiotic use at the time of urine collection. Covariate data were assessed from questionnaires or interviews at or near the time of sample collection, including parity, oral contraceptive use, intrauterine device (IUD) use, tubal ligation, hysterectomy, family history of ovarian cancer, smoking, weight and height [for calculation of body mass index (BMI) in kg/m²], and use of anti-inflammatory drugs.

The Eicosanoid Core Laboratory (principal investigator, Ginger Milne) at Vanderbilt University (Nashville, TN) measured PGE-M levels in the NHS/NHSII samples using a LC/MS method that has been described previously (27, 30, 33, 38). Briefly, PGE-M in each 0.5 mL urine specimen was stabilized by conversion to the O-methyloxime derivative and purified by C18 solid phase extraction with subsequent addition of the O-methyloxime derivatized deuterium-labeled internal standard (custom synthesis). Liquid chromatography was performed on an Acquity BEH C18 column (2.0 × 50 mm, 1.7 μm particle, Waters Corporation) connected to a Waters Acquity I-Class UPLC system and delivered to a Waters Xevo TQ-S Micro Triple Quadrupole Mass Spectrometer (Waters Corporation). The lower limit of detection of PGE-M was 0.05 ng/mL, substantially lower than levels typically detected in human urine. Results from a pilot study of 40 NHS participants with urine collected 2 years apart, reported good within-person stability of PGE-M (intraclass correlation = 0.62).

For SWHS samples, levels of PGE-M were measured at the Shanghai Institutes for Biological Sciences (principal investigator, Huiyong Yin) using a similar LC/MS method. Briefly, d6PGEM internal standard (2 ng, Cayman Chemical) was added to 0.75 mL urine and acidified to pH 3 with HCl; endogenous PGE-M was then converted to the Omethyloxime derivative by treatment with methoxyamine HCl. The methoximated PGE-M was extracted, applied to a C18 SepPak, and eluted with ethyl acetate. Liquid chromatography was performed on a Phenomenex KinetexC18 column (2.6 μm, 2.1 mm × 50.0 mm) attached to a CTCHTS autosampler and Shimadzu LC10 A VP system. The lower limit of detection of PGE-M was 0.04 ng/mL, substantially lower than levels typically detected in human urine.

Case–control pairs were assayed in the same batch, and quality control (QC) samples were included to assess analytic error. Laboratory personnel were blinded to case, control, and QC status. The intra-assay coefficients of variation (CV) were <4% and the interassay CVs were <9% in our NHS/NHSII samples and among prior PGE-M measures of SWHS samples at the Yin laboratory (33). Urinary creatinine was measured using a test kit from Enzo Life Sciences, Inc. to standardize levels of PGE-M to account for variation in urine concentrations (ng PGE-M/mg creatinine). To account for differences between laboratories, we recalibrated all participant PGE-M measures to a common standard by reassaying a batch of 30 samples (10 samples from each cohort) at the Milne laboratory (39).

We log-transformed ng PGE-M/mg creatinine and identified statistical outliers using the generalized extreme studentized deviate many-outlier procedure (40). One outlier with high PGE-M levels was detected in the SWHS dataset and excluded. We estimated ORs and 95% confidence intervals (CI) using conditional logistic regression. To avoid linearity assumptions we evaluated PGE-M in quartiles with cutoff points determined using the distribution of PGE-M values among controls, and we tested for trends using study-specific quartile medians. We first meta-analyzed the results from NHS/NHSII and SWHS and tested for heterogeneity between studies. In the absence of heterogeneity (P > 0.20 for all estimates), we pooled data from all three studies and used overall quartile medians in tests for trend in pooled analyses. Models were adjusted for matching factors (except antibiotic use) via conditional logistic regression of matched pairs, and multivariable models were further adjusted for parity (nulliparous, 1, 2, 3, and >3), oral contraceptive use (never, <1 year, 1–5 years, and ≥5 years), IUD use (ever and never), tubal ligation (yes and no), hysterectomy (yes and no), family history of ovarian cancer (yes and no), smoking (current, past, and never), and BMI (continuous). Pooled analyses included an interaction term between cohort and IUD use, as prior work suggests different associations for this exposure between the cohorts (41, 42).

We considered different associations for type II (high-grade serous and poorly differentiated) versus type I (low-grade serous, endometrioid, clear cell, and mucinous) ovarian cancers. Serous ovarian carcinomas with unknown grade were classified as type II ovarian cancers, while cases without information on tumor histology were excluded from histotype analyses. To calculate the associations, we used polytomous logistic regression and adjusted for matching factors, and to test for heterogeneity we used a likelihood ratio test comparing the polytomous model with a model with constant associations for type II and type I ovarian cancers. We evaluated multiplicative effect modification by BMI, aspirin use, and smoking using unconditional logistic regression adjusted for matching factors and covariates by including cross-product terms between each level of these categorical variables and PGE-M in our models and conducting a likelihood ratio test. Per the World Health Organization, the definition of overweight differs by race/ethnicity, so SWHS women (Chinese) were classified as overweight if BMI ≥ 24 kg/m², and NHS/NHSII (primarily non-Hispanic white) were classified as overweight if BMI ≥ 25 kg/m² (43). Aspirin use was defined as aspirin, nonaspirin NSAID, or acetaminophen use in the last 72 hours (NHS), aspirin use in the last 72 hours (NHSII), aspirin or nonaspirin NSAID use in the last 7 days (SWHS), or regular use of aspirin (NHS/NHSII: ≥2 times per week; SWHS: ≥3 times per week) over the past 2 years. Smoking status was dichotomized as ever versus never, given the low number of current smokers.

We used SAS 9.4 by SAS Institute Inc. for all analyses. Statistical tests assumed two-sided P values with α = 0.05.

We included 304 (123 NHS, 71 NHSII, and 110 SWHS) cases of epithelial ovarian cancer and 600 matched controls with urinary PGE-M measures. The control populations differed across cohorts with respect to PGE-M levels, age, menopausal status, hysterectomy, oral contraceptive use, IUD use, tubal ligation, and regular use of aspirin-based medications (Table 1). Smoking prevalence was low in all three studies. Overall, parity and oral contraceptive use were less common among ovarian cancer cases compared with controls, while family history of ovarian cancer was more common. Among controls, regular aspirin use was associated with lower urinary PGE-M, while smoking was associated with higher urinary PGE-M levels (Supplementary Table S1).

There was no evidence of an association across PGE-M quartiles and risk of ovarian cancer in study-specific multivariable models using study-specific quartile cutoff points (top vs. bottom quartile, NHS OR = 0.59, 95% CI = 0.30–1.15, P_trend = 0.10; NHSII OR = 0.93, 95% CI = 0.31–2.83, P_trend = 0.93; SWHS OR = 1.18, 95% CI = 0.51–2.73, P_trend = 0.48; Supplementary Table S2), or when considering common quartile cutoff points (top vs. bottom quartile, NHS OR = 0.62, 95% CI = 0.32–1.20, P_trend = 0.12; NHSII OR = 0.76, 95% CI = 0.21–2.72, P_trend = 0.97; SWHS OR = 1.10, 95% CI = 0.42–2.88, P_trend = 0.71; Table 2). No significant heterogeneity was observed (P > 0.40 for all estimates) so the cohorts were pooled (top vs. bottom quartile, OR = 0.80; 95% CI, 0.51–1.27; P_trend = 0.37; Table 2). In the combined study population, there was no heterogeneity by tumor histology (P = 0.53; Table 3).

The maximum time from urine collection to ovarian cancer diagnosis was 17.8 years with a median time of 5.7 years. As a sensitivity analysis, we removed cases diagnosed within 1 year of urine collection from our analyses (Supplementary Table S3). Results were similar to the main analysis (top vs. bottom quartile OR, 0.86; 95% CI, 0.54–1.39; P_trend = 0.65). We also stratified by median time from collection to case diagnosis (Supplementary Table S3) and observed no statistically significant evidence of an association between urinary PGE-M and ovarian cancer for those with a case diagnosis in the 5.7 years immediately following biospecimen collection (top vs. bottom quartile OR, 1.10; 95% CI, 0.56–2.15; P_trend = 0.92), or after the first 5.7 years (top vs. bottom quartile OR, 0.64; 95% CI, 0.32–1.28; P_trend = 0.25).

We hypothesized a priori that the association between prediagnosis urinary PGE-M and ovarian cancer risk would differ by levels of inflammatory exposures, including aspirin use and BMI. Results did not differ by aspirin use (P_interaction = 0.59; Table 4). For example, among women who reported recent or regular use of aspirin, the OR comparing the top versus bottom quartile was 0.80 (95%CI, 0.36–1.77; P_trend = 0.46) and among women who did not report aspirin use, the OR was 0.89 (95%CI, 0.50–1.61; P_trend = 0.77). Results also did not differ when stratifying by BMI (P_interaction = 0.82). Among normal weight women, the OR comparing the top versus bottom quartile was 0.65 (95%CI, 0.34–1.24; P_trend = 0.29) and among overweight or obese women, the OR was 1.02 (95%CI, 0.52–1.99; P_trend = 0.85). Results stratified by smoking status were not statistically different (P = 0.14) but trended in opposite directions. Among never smokers, the OR comparing the top versus bottom quartile was 1.13 (95%CI, 0.65–1.96; P_trend = 0.55) and among ever smokers, the OR was 0.46 (95%CI, 0.20–1.08; P_trend = 0.06).

A prior SWHS study of urinary PGE-M and breast cancer that was conducted among women with a very low smoking prevalence observed a dose-response association among postmenopausal women with BMI < 25 kg/m² (33), so we also conducted a post hoc analysis restricted to normal weight women who had never smoked (Supplementary Table S4). There was evidence of a positive association between PGE-M and ovarian cancer risk when comparing the top and bottom PGE-M quartiles in an unadjusted model (OR, 2.40; 95% CI, 0.92–6.29; P_trend = 0.04), but this finding was attenuated by adjustment for ovarian cancer risk factors (OR, 1.60; 95% CI, 0.50–5.14; P_trend = 0.29).

Given that aspirin and nonaspirin NSAIDs can affect PGE-M levels, we decided to further consider the influence of aspirin use on the association between PGE-M levels and risk of ovarian cancer, and evaluate the potential for effect modification by PGE-M levels in the association between aspirin and risk of ovarian cancer. First, we conducted an additional post hoc sensitivity analysis restricting our study population to those who had not reported aspirin (SWHS/NHS/NHSII), nonaspirin NSAID (SWHS/NHS), or acetaminophen (NHS) use in the past 72 hours (NHS/NHSII) or past week (SWHS). Results from this analysis were very similar to the main analysis. For example, the OR comparing the top versus bottom quartile was 0.84 (95%CI, 0.48–1.46; P_trend = 0.96; Supplementary Table S5), while the comparable OR from the main analysis was 0.80 (95%CI, 0.51–1.27; P_trend = 0.37; Table 2). Second, we conducted an exploratory analysis among women in NHS/NHSII with both PGE-M and aspirin data to consider whether PGE-M levels may alter the association between aspirin use and ovarian cancer risk (Supplementary Table S6). Considering standard dose (325 mg) aspirin, among those with low PGE-M levels, we observed a 1.44-fold (95%CI, 0.62–3.38) increased odds of ovarian cancer among current regular users of standard dose aspirin compared with never regular users, and among those with high PGE-M levels the comparable OR was 0.86 (95%CI, 0.25–2.91). For low dose (≤100 mg) aspirin, among those with low PGE-M levels, we observed lower odds of ovarian cancer among current regular users of low dose aspirin compared with never regular users (OR, 0.73; 95% CI, 0.36–1.47), and among those with high PGE-M levels the comparable odds ratio was 1.14 (95%CI, 0.51–2.58). There was no statistically significant evidence of heterogeneity by PGE-M level for standard dose (P_{heterogeneity} = 0.36) or low dose (P_{heterogeneity} = 0.89) aspirin, but these findings merit further exploration in a larger study.

In this large, prospective nested case–control study including primarily non-Hispanic white and Chinese women, we observed that prediagnosis urinary PGE-M levels were not significantly associated with risk of ovarian cancer. Furthermore, we did not observe any significant interactions by recent aspirin use, BMI, or smoking, which are key inflammatory factors that may interact with the prostaglandin synthesis pathway (23, 30, 44–48). Our post hoc analysis of aspirin use and risk of ovarian cancer by PGE-M level reported no statistically significant evidence of heterogeneity by PGE-M level; however, our observation of a possible lower risk of ovarian cancer among current low dose aspirin users with low PGE-M levels but not high PGE-M levels merits further evaluation in a larger study.

Research on ovarian cancer biology supports a role for prostaglandins in ovarian carcinogenesis. One in vitro study reported that the COX2 inhibitor, NS-398, reduced PGE2 in ovarian cancer cells (19), and research by the same group and others observed PGE2 in the ascites of patients with ovarian cancer (19), COX2 expression in ovarian tumors (12–15, 19, 49, 50), and poorer survival among patients with COX2⁺ ovarian cancer (15–19). In our recent work, we observed COX1 expression, COX2 expression, or both in many ovarian cancer cases (15, 51); however, when we evaluated the associations between aspirin or nonaspirin NSAID use and ovarian cancer risk by tumor expression of these markers, there was no evidence of heterogeneity (51). Our results are consistent with a recent evaluation of fatty acid metabolites in the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial, which reported no significant associations between COX-derived prostanoids and risk of ovarian cancer (52). It is possible that prostanoid-dependent inflammatory mechanisms are important to ovarian cancer growth and metastasis, but prostanoid-independent inflammatory mechanisms are more relevant to ovarian cancer development.

Epidemiologic research has reported positive associations between PGE-M levels and risk of colorectal cancer (25, 26), colorectal adenoma (27), gastric cancer (28, 29), small-cell lung cancer (30), and pancreatic cancer (31). These associations varied substantially in magnitude, from a 5.6-fold (95%CI, 2.4–13.5) higher risk of colorectal cancer for women comparing PGE-M levels in the highest versus lowest quartile with a 1.63-fold (95%CI, 1.01–2.63) increased risk of pancreatic cancer comparing the highest versus lowest tertile. Our results were more similar to studies of postmenopausal breast cancer, which did not report a positive association overall, but did report associations among specific, albeit small (n < 150), subgroups (e.g., low BMI, ref. 33; nonregular users of NSAIDs, ref. 34). Consistent with existing literature, we observed a positive association between urinary PGE-M and inflammatory factors, including smoking (38, 53) and, to a lesser extent, increasing BMI (34). We also observed a possible positive association between urinary PGE-M and ovarian cancer among normal-weight women who had never smoked. This analysis was motivated by the NHS study of PGE-M, aspirin, and adenoma risk, in which anti-inflammatory drug use was associated with lower adenoma risk among women with mid-high levels of PGE-M but not low levels of PGE-M (27). Multiple obesity pathways are related to prostaglandin synthesis or signaling (44–46) and any modest effects of prostaglandins on ovarian cancer risk may be eclipsed by more extensive dysregulation of inflammation among those with high BMI, so this finding warrants further exploration in future studies.

Urinary PGE-M may reflect both inflammatory and anti-inflammatory factors, including the use of aspirin and nonaspirin NSAIDs. Use of NSAIDs, including aspirin (23, 47, 48), ibuprofen (30), and indomethacin (48), have been associated with lower PGE-M levels. Our three study populations reported very different patterns of aspirin use. Prevalence of aspirin use was nearly 50% among NHS participants, approximately 15% among NHSII participants, and fewer than 5% among SWHS participants. Despite these differences in usage patterns, we observed similar associations between PGE-M levels and ovarian cancer in NHS/NHSII and SWHS overall and when stratified by aspirin use.

This study had several strengths, including prediagnosis urine collection and the ability to account for differences between laboratories via recalibration of all participant PGE-M measures to a common standard. For example, we minimized the potential for reverse causation by collecting the urine biospecimen prior to ovarian cancer diagnosis, and we were able to conduct a sensitivity analysis excluding cases (and matched controls) with diagnosis dates within 1 year of specimen collection. Another strength was detailed covariate information that allowed us to control for important confounders and evaluate effect modification by inflammatory exposures. Furthermore, the inclusion of study populations with different racial/ethnic backgrounds and different aspirin usage patterns increases generalizability of the observed associations.

We also recognize several important limitations of our study. Biomarker validity is one potential concern. While urinary PGE-M reflects systemic PGE2 levels (22, 24), it is unclear whether urinary PGE-M levels are reflective of PGE2 exposure in the peritoneal cavity. However, other systemic markers (e.g., C-reactive protein and androgens) have been associated with ovarian cancer risk (35, 54, 55). Another important limitation is that we only obtained one PGE-M measure, although our prior study demonstrated a reasonable intraclass correlation over 2 years and we did not observe an association in cases diagnosed within 5.7 years of urine collection.

The decision to pool NHS/NHSII/SWHS led to additional strengths and limitations for this analysis. Pooling improved power in our main analysis; however, power remained limited with respect to detecting modest associations, examining associations with specific histotypes, and detecting effect modification. In addition, SWHS, NHS, and NHSII had slightly different urine collection protocols (e.g., all SWHS samples were spot urine and nearly all NHS/NHSII were first morning), thus we were unable to account for differences in these protocols in the analysis. While the NHS/NHSII were able to match on more factors that could explain variability in PGE-M than SWHS (e.g., timing of collection within the menstrual cycle), it is important to note that the SWHS previously detected associations between urinary PGE-M and other cancer types (25, 28, 29, 32, 33).

In summary, we observed no evidence of an association between prediagnosis urinary PGE-M levels and risk of ovarian cancer, despite the modest inverse association of aspirin and nonaspirin NSAIDs with risk of ovarian cancer in these and other populations (1–7). Overall, and particularly when considered in conjunction with our finding that associations between NSAID use and ovarian cancer risk do not differ by tumor expression of COX1 or COX2 (51), the results of this study suggest that regulation of the prostaglandin synthesis pathway may not be the most important link between NSAID use and ovarian carcinogenesis.

A. Beeghly-Fadiel reports receiving speakers bureau honoraria from Gettysburg College and has provided expert testimony for the American College of Obstetricians and Gynecologists. No potential conflicts of interest were disclosed by the other authors.

The authors take full responsibility for the analysis and interpretation of data. The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

Conception and design: M.E. Barnard, X.-O. Shu, W. Zheng, S.S. Tworoger

Development of methodology: M.E. Barnard, E.Y. Akam, S.S. Tworoger

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Beeghly-Fadiel, G.L. Milne, A.H. Eliaseen, X.-O. Shu, Y.-B. Xiang, W. Zheng, S.S. Tworoger

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.E. Barnard, A. Beeghly-Fadiel, A.T. Chan, A.H. Eliaseen, B.A. Rosner, W. Zheng, S.S. Tworoger

Writing, review, and/or revision of the manuscript: M.E. Barnard, A. Beeghly-Fadiel, G.L. Milne, A.T. Chan, A.H. Eliaseen, B.A. Rosner, K.L. Terry, Y.-B. Xiang, W. Zheng, S.S. Tworoger

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E.Y. Akam, A.H. Eliaseen, Y.-B. Xiang, S.S. Tworoger

Study supervision: A.T. Chan, A.H. Eliaseen, X.-O. Shu, S.S. Tworoger

This study was supported by the NCI of the NIH (award numbers: UM1 CA186107 to B. Rosner; P01 CA87969 to A.H. Eliassen, B. Rosner, K. Terry, and S. Tworoger; UM1 CA176726 to A.H. Eliassen; R01 CA137178 to A. Chan; R01 CA49449 to B. Rosner; R01 CA67262 to A.H. Eliassen; UM1 CA182910 to W. Zheng; and R37 CA070867 to W. Zheng). M. Barnard was supported by the NCI of the NIH (award numbers T32 CA09001 and F99 CA212222). A. Chan is a Stuart and Suzanne Steele MGH Research Scholar. We would like to thank the participants and staff of the Nurses' Health Studies and the Shanghai Women's Health Study for their valuable contributions to this research. The authors also thank the following state cancer registries: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, and WY.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.