{"paper_id":"70450999-e54e-468c-bf39-aa29360e6ab3","body_text":"Ovarian cancer is the\nthird most common gynecological cancer diagnosed\nglobally and the most fatal gynecologic cancer. \n − \n \n Racial and ethnic,\nsocioeconomic, and regional disparities in both incidence and survival\nin the U.S. have been documented. \n , \n Ovarian cancer\nrisk increases with age and with genetic (family history, inherited\ngenetic mutations, e.g., BRCA1 and BRCA2 ), gynecologic (endometriosis), hormonal (use of hormone replacement\ntherapy after menopause), and reproductive (older age at first full-term\npregnancy) risk factors. Protective factors\ninclude use of oral contraceptives and having at least one full-term\npregnancy.\nDrinking water can be\na major source of exposure to toxicants,\nincluding known or suspected carcinogens. In the U.S., approximately\n90% of the population is served by community water systems (CWS). CWS are regulated by the U.S. Environmental Protection\nAgency (EPA) under the Safe Drinking Water Act through enforceable\nmaximum contaminant levels (MCLs), standards that consider economic\nfeasibility, technical feasibility/treatment technologies, and public\nhealth benefit for certain health end points. Regulated contaminants fall under six classes, including metals/metalloids\n(arsenic), radionuclides (uranium and gross alpha), inorganic chemicals\n(nitrate), and disinfection byproducts (DBPs), such as trihalomethanes\n(THMs, which include chloroform, dibromochloromethane, bromodichloromethane,\nand bromoform, and are regulated as the sum total, TTHM). \n , \n With limited prior epidemiologic evidence, ovarian cancer was not\nconsidered in the formulation of existing MCLs.\nA positive association\nwas previously observed between drinking\nwater nitrate exposure and ovarian cancer in the Iowa Women’s\nHealth Study, a prospective population-based cohort of postmenopausal\nwomen, and in the Agricultural Health Study, a prospective cohort\nof pesticide applicators and their spouses in Iowa and North Carolina. \n , \n Nitrate can be ingested from drinking water as well as dietary sources,\nand is classified by the International Agency for Research on Cancer\n(IARC) as a probable human carcinogen under conditions that result\nin the endogenous formation of N-nitroso-compounds (NOCs).\nFew other cohort or case-control studies\nhave evaluated drinking\nwater contaminants and ovarian cancer risk. \n , , \n Animal studies of drinking water uranium\nexposure observed uranium accumulation and DNA hypomethylation, a\nprocess linked to carcinogenesis, in the ovaries. Animal evidence also suggests that ingestion of inorganic\narsenic may induce ovarian tumors in the offspring of exposed mice. Uranium is classified as a probable human carcinogen\nby IARC based on limited evidence for lung cancer, and arsenic is\nclassified as a known human carcinogen based on sufficient evidence\nfor bladder, skin, and lung cancers. \n − \n \n Certain THMs have been\nfound to have endocrine-disrupting properties, a mechanism that may\ninduce ovarian carcinogenesis; \n , \n however, the epidemiologic\nevidence for an association between drinking water THM exposures and\novarian cancer is inconsistent. \n ,\nOur objectives were to evaluate associations between\nCWS nitrate, uranium, arsenic, and TTHM exposures with ovarian cancer\nincidence in the California Teachers Study (CTS), a prospective cohort\nof female California teachers and administrators. We aimed to evaluate\nthe effect of single and co-occurring contaminants at levels of exposure\nexperienced by general U.S. populations. To our knowledge, this is\nthe first analytic epidemiologic study to evaluate the effects of\nco-occurring nitrate, uranium, arsenic, and TTHM exposures and ovarian\ncancer risk in U.S. women.\n\nThe CTS is a prospective\ncohort of women that was designed to investigate the etiology of breast\nand other cancers. Public school teachers\nand administrators in the California State Teachers Retirement System\nwere mailed a self-administered questionnaire; 133,470 completed the\nsurvey and enrolled in 1995–1996. We excluded participants who lived outside of California at baseline\n( N = 8,332), consented to breast cancer research\nonly ( N = 18), had a prevalent cancer of any type\nas reported to the California Cancer Registry ( N =\n13,601; does not include nonmelanoma skin cancer), were censored on\nor before the start date ( N = 3), had a risk-eliminating\nsurgery (bilateral or second oophorectomy, N = 11,786)\nor did not provide information on a prior oophorectomy ( N = 796), for a total of 98,934 participants eligible for inclusion\nin this study ( Figure S1 ). Participants\nwere censored at the earliest of their dates of move out of California,\nany cancer diagnosis, a bilateral oophorectomy, death, or the end\nof follow-up (December 31, 2020).\nParticipants provided information\nat enrollment (1995–1996) on sociodemographic characteristics\n(race and ethnicity, age) anthropometrics (height and weight were\nused to calculate body mass index [BMI], kg/m 2 ), smoking\nand alcohol consumption, and personal medical history (menopause status,\noral contraceptive use, total number of live births); educational\nattainment was collected at questionnaire 4 (2005–2008; response\nrate 69%). Dietary intake of frequently consumed foods was collected\nat enrollment via a 1995 103-item Block food-frequency questionnaire\n(FFQ), which was validated in the CTS. \n , \n Total daily\ncaloric intake (kilocalories/day), total intake of vitamin C (mg/day)\nfrom diet and supplements, and total red meat intake (g/day) were\ncalculated as previously described. Participants\nwho had missing or unrealistic daily caloric intakes (<600 or >5,000\ntotal kilocalories/day, including alcohol) were excluded from the\ndietary analyses. Nitrate and nitrite\ncontents of foods that contributed to each FFQ item were estimated\nfrom the published literature. We computed\nthe nitrate and nitrite intake for each FFQ item (mg) by weighing\nthe food-specific values by sex-specific intake amounts from the 1994–1996\nContinuing Survey of Food Intake by Individuals (CSFII) and multiplying by the reported intake (g/day).\nWe summed across line items to calculate the daily nitrate and nitrite\nintake overall (milligrams per day) and from plant and animal sources\nseparately, including from processed meats only. Census block group-level\ncharacteristics of the enrollment addresses (quartiles of neighborhood\nsocioeconomic status (SES), and urbanicity dichotomized as metropolitan\nand nonmetropolitan areas) were previously developed and described\nusing 1990 Census data.\nFirst\nprimary incident ovarian cancers were identified through linkage with\nthe California Cancer Registry, and mortality information was obtained\nfrom linkage with the California state mortality file and the Social\nSecurity Administration death master file. Epithelial ovarian cancer\nwas defined based on the International Classification of Diseases\nfor Oncology, Third Edition, site C569 and included the following\nhistologic types (histotypes): high-grade serous ( N = 199), low-grade serous ( N = 12), endometroid\n( N = 35), mucinous ( N = 29), clear\ncell ( N = 25), and other epithelial ( N = 113). We evaluated risk for all epithelial ovarian carcinomas\n( N = 413) (hereafter all ovarian cancer) and high-grade\nserous cases, the most common histotype, separately.\nAs previously\ndescribed, we obtained a geospatial data set of statewide\ndrinking water boundaries (the Water Boundary Tool) that was cleaned\nand processed by the California Office of Environmental Health Hazard\nAssessment (OEHHA). \n − \n \n \n Monitoring data (1990–2020) for CWS (nitrate, uranium, gross\nalpha, arsenic, TTHM) were obtained from OEHHA, \n , \n and average annual concentrations of each contaminant were computed\nfor each CWS. DBP (i.e., THM) measurements were extracted from post-treatment\nsample points collected within the distribution systems. For all other\ncontaminants, measurements from samples of treated and delivered drinking\nwater were prioritized. Except for the DBPs, when treated samples\nwere not available for a contaminant, we averaged results from raw\nand untreated samples. Uranium concentrations\nwere converted from pCi/L to μg/L using 1.49 as the conversion\nfactor (pCi/L*1.49 = μg/L). CWS are required to report nondetections\nand concentrations when above the detection limits for the purposes\nof reporting (DLR) ( Table S1 ). Samples\nmarked as “below the DLR”, or with concentrations of\nzero, or values unlikely to reflect true concentrations (i.e., equal\nto DLR or 1/2 DLR), were assigned a value based on a single imputation\nthat used Tobit regression, existing measurement data, and assumed\na log-normal distribution. The upper\nbound for imputation was derived from the median of reported concentrations\nbelow the DLR. Among CWS linked to the\nenrollment addresses of CTS participants, the N (%)\nthat reported ≥1 year of detectable data (not imputed) out\nof CWS that reported at least one year of measurement data (including\nimputed values) from 1990 to 2020, was as follows: arsenic (1,163\nout of 1,227 CWS; 95%); uranium (818 out of 901 CWS; 91%); gross alpha\n(1,149 out of 1,224 CWS; 94%); nitrate (1,130 out of 1,232 CWS; 92%);\nand TTHM (1,229 out of 1,239 CWS; 99%). Information on CWS water source type (groundwater, groundwater under\nthe influence of surface water, surface water) and CWS size of the\npopulation served (very small (≤500 people), small (>500–3300\npeople), medium (>3300–10,000 people), large (>10,000–<1,000,000\npeople), and very large (≥1,000,000 people) were also obtained\nfrom OEHHA.\nResidential histories\nwere previously constructed for CTS participants. Briefly, addresses at enrollment (1995–1996) were\nself-reported at the baseline questionnaire, and addresses through\n2019 were obtained from participants who completed follow-up questionnaires\nas well as from the U.S. Postal Service, LexisNexis, Experian, and\nCalifornia Cancer Registry linkages. Self-reported information about\nyears lived at the current address were provided in the fourth (2005–2008;\nresponse rate 69%), fifth (2012–2015; response rate 61%), and\nsixth (2017–2019; response rate 43%) questionnaires. All participants eligible for inclusion in our\nstudy had a geocoded address within CA at baseline. Geocoded addresses\nwere linked to CWS distribution boundaries using QGIS Desktop 3.8.1\nand assigned to the intersecting CWS ( N at enrollment\n= 91,127, 92%) ( Figure S2 ). \n , , , , \n Participants whose enrollment address did\nnot link to a CWS ( N = 7,807, 8%) were assumed to\nbe domestic well users and were excluded from the study ( Figure S1 ).\nIn our main analyses, we computed\n15-year average concentrations\n(1990–2005) of CWS exposures linked to the address at enrollment\nand restricted our analyses to participants with a total residential\nduration at the enrollment address of at least 10 years (based on\ntheir self-reported information about residential duration collected\nat the fourth questionnaire or their residential history), excluding\nparticipants on a CWS who resided less than 10 years at the enrollment\naddress ( N = 27,809). Additionally, we excluded participants\nwho were missing CWS arsenic, nitrate, TTHM, and gross alpha estimates\n( N = 3,437), for a total of 59,881 participants (total\nperson-years = 1,139,582; mean follow-up = 19.0 years) ( Figure S1 ). A subset of these participants also\nhad CWS uranium estimates ( N = 56,314, 94%; total\nperson-years = 1,072,419, mean follow-up = 19.0 years). We also computed\nthe percent of years in 1990–2005 that annual average CWS concentrations\nwere ≥1/2 the MCL, out of the years of monitoring data reported\nfor each CWS (1990–2005). We used the current MCLs for arsenic\n(10 μg/L, established by the Final Arsenic Rule, effective 2006;\nthe prior MCL for arsenic was 50 μg/L), uranium (30 μg/L,\nestablished by the Radionuclides Rule, effective 2003), gross alpha\n(15 pCi/L, established by the Radionuclides Rule), nitrate-nitrogen\n(NO 3 -N, hereafter referred to as nitrate; 10 mg/L, established\nby the Chemical Contaminant Rules, effective 1992), and TTHM (80 μg/L,\nestablished by the Stage 1 and 2 Disinfectants and Disinfection Byproducts\nRules, enacted 2006, effective for all CWS in 2013). \n , − \n \n \n \n We performed Spearman correlation analyses to describe the correlations\nbetween contaminant concentrations.\nIn additional analyses,\nwe linked participants’ addresses\nfrom their postenrollment residential history that included residential\nchanges over follow-up to the corresponding CWS (86,294; Figure S1 ) and the yearly average estimates of\nCWS exposures for each year at the residence(s). We interpolated data\nfor years without measurements using nine-year fixed compliance cycles\nestablished by the U.S. EPA under the Standardized Monitoring Framework;\neach compliance cycle is further divided into three monitoring periods\nof 3 years each. We interpolated data\nusing the average of annual concentrations from the corresponding\nmonitoring period, and if unavailable, we substituted the average\nestimate from the (1) earlier monitoring period or, if unavailable,\n(2) later monitoring period, within the compliance cycle ( Table S2 ).\nAll statistical\nanalyses were conducted in R version 4.3.3 within the CTS Researcher\nPlatform. To interpret the results, we\nconsidered the magnitude of effect estimates and corresponding 95%\nconfidence intervals (CI) and statistical significance ( p < 0.05). We described participant\ncharacteristics and CWS contaminant exposures overall and among all\novarian cancer cases and the high-grade serous histotype. We used\nCox proportional hazards regression models, with time-on-study as\nthe time scale, to estimate hazard ratios (HR) and 95% CIs for the\nassociations between drinking water contaminant exposures and incident\novarian and high-grade serous cancers.\nIn models of the 15-year\naverage concentrations, we parametrized drinking water exposures in\ntwo ways: continuously after a base-2 log transformation and in categories\ndefined by quantiles (0–<25% [reference], 25–<50%,\n50–<75%, 75–<90%, ≥90%). In categorical\nanalyses, we tested for a linear trend using the median of each quantile\nparametrized as a continuous variable. In models of the percent (%)\nof years in 1990–2005 that annual average concentrations were\n≥1/2 of the MCL, we evaluated categories of >0–≤10%\nand >10% of years compared to 0% (reference). Models were adjusted\nfor potential confounders and other known risk factors: baseline age\nand baseline age 2 ( model 1 ), and further\nadjusted for BMI category (<25 kg/m 2 , 25–<30\nkg/m 2 , ≥30 kg/m 2 , or missing), menopause\nstatus (pre-, peri/postmenopause, or missing), ever had live births\n(yes/no/missing), and oral contraceptive use (ever/never/missing)\n( model 2 ). \n − \n \n In our main analyses, we focused\non CWS nitrate, uranium, arsenic, and TTHM exposures. We evaluated\ngross alpha exposures in supplemental analyses to compare to the findings\nfor uranium, as more CWS provided gross alpha monitoring data than\nuranium due to state primacy regulations. In California, CWS were\npermitted by Section 64442­(f) of Title 22 of the California Code of\nRegulations to substitute gross alpha activity measurements for uranium\nmeasurements if the gross alpha concentration did not exceed 5 pCi/L. We additionally evaluated potential nonlinearity\nin the associations using cubic splines.\nTo evaluate the joint\neffects of the drinking water contaminant\nmixture on incident ovarian and high-grade serous cancer, we used\nquantile-based g-computation via the qgcomp package\nin R. \n , \n Nitrate, uranium, arsenic, and TTHM estimates\nwere log2 transformed and divided by the interquartile range (IQR)\nto standardize the concentrations. We estimated the hazard ratios\nper IQR increase in the four-contaminant mixture and evaluated the\ncontribution of each contaminant to the overall effect. We explored\nmixtures composed of different combinations of co-occurring contaminants\n(composed of two or three contaminants) that were adjusted for the\ncontaminants not included in the mixture. Separately, we evaluated\nmodels that were adjusted for all other contaminants.\nIn time-varying\nanalyses, cumulative average concentrations were\nlagged 5 years. We evaluated the adjustment for age as a time-varying\ncovariate in these models; other covariates were otherwise identical\nto the main analyses.\nWe explored potential effect modification\nby menopausal status\nand BMI category. Ovarian cancer is more commonly diagnosed among\npostmenopausal women than premenopausal women, as the risk of developing\novarian cancer increases with age. Obesity\nhas estrogenic effects due to the aromatization of androgens in adipocytes,\nwhich is associated with an increased risk of development and progression\nof breast and other cancers. \n − \n \n \n Some evidence suggests that arsenic and uranium may\nbe obesogenic. \n , \n We also evaluated potential effect\nmodification by smoking status, as cigarette smoking is a source of\nexposure to arsenic, tobacco-specific nitrosamines, and radioactive\nelements. \n , \n We stratified and mutually adjusted analyses\nby urbanicity (metropolitan and nonmetropolitan areas), CWS water\nsource type, and CWS size, due to differences in contaminant exposures\nby these characteristics in this cohort. We used the likelihood ratio test to determine statistical heterogeneity\nfor all stratified analyses, deriving a p-value from the Chi squared\ntest statistic.\nWe evaluated associations between CWS nitrate\nand ovarian cancer\nrisk stratified by tertile of daily vitamin C intake and red meat\nintake (the major source of heme iron) based on previous evidence\nsuggesting modification of cancer risks by these factors due to their\nrole in decreasing and increasing endogenous nitrosation, respectively. \n , \n We also conducted exploratory analyses evaluating associations with\narsenic and uranium exposure stratified by tertile of daily vitamin\nC intake, based on limited evidence that vitamin C intake may decrease\nmetal/metalloid toxicity. In supplemental\nanalyses, we also evaluated associations with total dietary nitrate\nand nitrite as well as plant, animal, and processed meat nitrite sources\nseparately.\nBecause most\nexposure data were postenrollment (15-year averages at the 1995–1996\nenrollment address were calculated using 1990–2005 data), and\noverlapped with the first ten years of follow-up, we performed a sensitivity\nanalysis in which we started follow-up on January 1, 2005, and excluded\nparticipants who were censored before that time. Separately, to confirm\nwhether the observed associations were consistent when all exposures\nwere below regulatory limits, we evaluated single contaminant models\nand mixtures analyses among participants with average levels (1990–2005)\nof all contaminants (nitrate, uranium, gross alpha, arsenic, and TTHM)\n< MCL only.\nIn a posthoc analysis, we evaluated ovarian cancer\nrisk with time-varying exposures (cumulative average exposures lagged\n5 years linked to the residential history) among the same participants\nthat were included in our main analyses (enrollment address duration\n≥ 10 years), allowing direct comparisons between findings from\nthe 15-year averages and the time-varying exposures.\n\nThe median baseline age of\nall participants was 51 years and 54\nyears for those who developed ovarian cancer ( Table \n ). Approximately 42% of participants were\npremenopausal and 48% were peri or postmenopausal (10% missing), while\n33% and 56% of participants who developed ovarian cancer were pre\nor peri/postmenopausal at baseline. Most participants were non-Hispanic\nwhite (85%), followed by Hispanic (5%), Asian (4%), black (3%), Native\nAmerican (1%) and other/multiracial (1%); whereas a higher percent\n(92%) of cases were non-Hispanic white women. Most participants had\na BMI < 25 kg/m 2 , were never smokers, consumed <20g\nalcohol/day, had a bachelor’s degree or higher, and lived in\ncensus block groups in the upper two quartiles of SES and in metropolitan\nareas at enrollment. A higher proportion of cases had never used oral\ncontraceptives (41% and 39% for all ovarian cancers and the high-grade\nserous histotype, respectively) compared to the overall cohort (31%).\nCommunity\nwater system (CWS)\nexposures were linked to the address at enrollment. Analyses were\nrestricted to participants with a residential duration at enrollment\n≥10 years (N = 59,881).\nResidential duration at the enrollment\naddress was determined based on the residential history and self-reported\ninformation about duration at the current home at questionnaire 4.\nSES = socioeconomic status.\nCWS concentrations represent\n15-year\naverage (1990-2005) concentrations.\nMedians and interquartile ranges (IQRs) of average\nCWS exposures\nwere below regulatory limits for all contaminants. We observed modest\ndifferences with overlapping distributions in nitrate, uranium, arsenic,\nand TTHM concentrations comparing all participants to those who developed\novarian cancer ( Table \n ). Nitrate, arsenic, and uranium concentrations were positively correlated\nin pairwise analyses (Spearman’s rho ranging from 0.32 for\nnitrate and arsenic to 0.46 for uranium and arsenic) and negatively\ncorrelated with TTHM concentrations (ranging from −0.13 for\nuranium to −0.33 for nitrate) ( Figure S3 ). Uranium and gross alpha concentrations were highly correlated\n(0.82).\nWe describe the findings from the fully adjusted analyses\n(model\n2), which were similar to model 1 results ( Table \n ). We found positive associations with ovarian\ncancer (HR, 95% CI) per doubling in average (1990–2005) uranium\nconcentrations (HR = 1.07, 95% CI 1.00, 1.15), with the highest risk\nfor the 90th percentile compared to the lowest quartile (HR= 1.53,\n95% CI 1.08, 2.17; p trend = 0.04). A doubling in average nitrate\nlevels was associated with higher risk of high-grade serous cancer\n(HR = 1.09, 95% CI 1.01, 1.17), with the greatest risk observed at\n≥90th percentile of nitrate exposure (HR = 1.54, 95% CI 0.94,\n2.54; p trend = 0.03). A doubling in average arsenic exposure was\npositively associated with total (HR = 1.05, 95% CI 0.96, 1.14) and\nhigh-grade serous (1.07, 95% CI 0.95, 1.20) ovarian cancer. No associations\nwere observed with TTHM exposures. Exposure-response models using\ncubic splines were generally consistent with these findings ( Figures \n and ).\nCWS\nexposures were linked to the address at enrollment.\nAnalyses were restricted\nto participants with a residential duration at enrollment ≥10\nyears .\nCWS concentrations represent 15-year\naverage (1990-2005) concentrations of arsenic (μg/L), uranium\n(μg/L), nitrate-N (mg/L), and TTHM (μg/L).\nResidential duration at the enrollment\naddress was determined based on the residential history and self-reported\ninformation about duration at the current home at questionnaire 4.\nMean follow-up=19.0 years, total person-years: 1,139,582. For uranium,\nmean follow-up=19.0 years, total person-years: 1,072,419.\nModel 1 was adjusted for baseline\nage (years) + baseline age 2 .\nModel 2 = model 1 + BMI category\n(<25 kg/m 2 , 25−<30 kg/m 2 , >30\nkg/m 2 , or missing), menopause status (pre-, peri/post-menopause,\nor missing), live births (no/yes/missing), oral contraceptive use\n(never/ever/missing).\nP\ntrend was evaluated using the\nmedian of each quantile.\nHazard ratios\n(95% CI) for all epithelial ovarian carcinomas by\ncommunity water system (CWS) nitrate, uranium, arsenic and total trihalomethane\n(TTHM) exposures in the California Teachers Study. CWS exposures were linked to the address at enrollment. Analyses\nwere restricted to participants with a residential duration at enrollment\n≥ 10 years. Lines with shaded areas\nrepresent the hazard ratios (95% confidence interval (CI)), based\non cubic splines for Cox proportional hazards models using log2-transformed\nconcentrations with knots at the 10th (reference), 50th, and 90th\npercentiles. A black horizontal line is\nat HR = 1. The histogram represents the frequency distribution of\nCWS estimates in the study sample. CWS\nconcentrations represent long-term average (1990–2005) concentrations\nof arsenic (μg/L), uranium (μg/L), nitrate-N (mg/L), and\nTTHM (μg/L). For plotting purposes, several extremely low concentrations\nwere omitted of arsenic (1 estimate = 0.008 μg/L) and TTHM (9\nestimates <5.20 × 10 –3 ). Residential duration at the enrollment address was determined\nbased on the residential history and self-reported information about\nduration at the current home at questionnaire 4. Models were adjusted for age at baseline, age 2 , BMI category (<25 kg/m 2 , 25–<30 kg/m 2 , >30 kg/m 2 , or missing), menopause status (pre-,\nperi/postmenopause, or missing), live births (yes/no/missing), and\noral contraceptive use (never/ever/missing).\nHazard\nratios (95% CI) for high-grade serous ovarian cancers\nby\ncommunity water system (CWS) nitrate, uranium, arsenic and total trihalomethane\n(TTHM) exposures in the California Teachers Study. CWS exposures were linked to the address at enrollment. Analyses\nwere restricted to participants with a residential duration at enrollment\n≥ 10 years. Lines with shaded areas\nrepresent the hazard ratios (95% confidence interval (CI)), based\non cubic splines for Cox proportional hazards models using log2-transformed\nconcentrations with knots at the 10th (reference), 50th, and 90th\npercentiles. A black vertical line is\nat HR = 1. The histogram represents the frequency distribution of\nCWS estimates in the study sample. 1 CWS concentrations represent\nlong-term average (1990–2005) concentrations of arsenic (μg/L),\nuranium (μg/L), nitrate-N (mg/L), and TTHM (μg/L). For\nplotting purposes, several extremely low concentrations of arsenic\nwere omitted (1 estimate = 0.008 μg/L) and TTHM (9 estimates\n<5.20 × 10 –3 ). Residential duration at the enrollment address was determined based\non the residential history and self-reported information about duration\nat the current home at questionnaire 4. Models were adjusted for age at baseline, age 2 , BMI category\n(<25 kg/m 2 , 25–<30 kg/m 2 , >30\nkg/m 2 , or missing), menopause status (pre-, peri/postmenopause,\nor missing), live births (yes/no/missing), and oral contraceptive\nuse (never/ever/missing).\nIn analyses of joint effects, HRs per IQR increase\nin the contaminant\nmixture of all four contaminants were the largest (all ovarian cancer\nHR = 1.34, 95% CI 0.95, 1.88; high-grade serous cancer HR = 1.77,\n95% CI 1.09, 2.87) compared to models of the contaminant mixture containing\ndifferent combinations of two and three contaminants ( Table \n ). Uranium contributed the largest\nweight (56%) to the positive mixture effect on all ovarian cancers,\nand nitrate contributed the largest weight (48%) to the positive mixture\neffect on the high-grade serous histotype. In single contaminant analyses\nmutually adjusted for other contaminants, we observed that uranium\nhad the largest hazard ratio for all ovarian cancer (HR = 1.18, 95%\nCI 0.99, 1.42), and that nitrate had the largest hazard ratio for\nthe high-grade serous histotype (HR = 1.30, 95% CI 0.99, 1.72) ( Table \n ).\nCWS exposures were linked to the address at enrollment and log2\ntransformed.\nAnalyses were restricted to participants with a residential duration\nat enrollment ≥10 years .\nCWS concentrations represent 15-year\naverage (1990-2005) concentrations of arsenic (μg/L), uranium\n(μg/L), nitrate-N (mg/L), and TTHM (μg/L).\nResidential duration at the enrollment\naddress was determined based on the residential history and self-reported\ninformation about duration at the current home at questionnaire 4.\nMean follow-up = 19.0 years, total person-years: 1,072,419.\nModel 1 was adjusted for baseline\nage (years) + baseline age 2 .\nModel 2 = model 1 + BMI category\n(<25 kg/m 2 , 25−<30 kg/m 2 , >30\nkg/m 2 , or missing), menopause status (pre-, peri/post-menopause,\nor missing), live births (no/yes/missing), oral contraceptive use\n(never/ever/missing).\nConcentrations\nwere log2 transformed\nand subsequently divided by the IQR. Models were adjusted for all\nother contaminants (log2 transformed).\nFor contaminants included in the\nmixture, concentrations were log2 transformed and subsequently divided\nby the IQR. The effect of the drinking water contaminant mixture was\nevaluated using quantile q-computation, and hazard ratios are interpreted\nper IQR increase. Models were adjusted for contaminants not in the\nmixture (log2 transformed).\nCompared to participants who had no years of exposure\nto nitrate\n≥ 1/2 MCL (0%), having >10% years of exposure to nitrate\nwas\nassociated with an elevated risk of high-grade serous histotype (HR=\n1.72, 95% CI 1.20, 2.46), whereas no association was observed with\nall ovarian cancer ( Table \n ). Having >10% of years of exposure to uranium ≥\n1/2\nMCL versus 0% was positively associated with all ovarian cancer (HR=\n1.22, 95% CI 0.94, 1.57) and the high-grade serous histotype (HR=\n1.28, 95% CI 0.90, 1.83). We observed suggestive positive associations\nfor >0% of years of exposure to arsenic ≥ 1/2 MCL and the\nhigh-grade\nserous histotype, and no associations for TTHM.\nCWS exposures\nwere linked to\nthe address at enrollment. Analyses were restricted to participants\nwith a residential duration at enrollment ≥10 years .\nThe\npercent (%) of years was calculated\nas the number of years that average annual concentrations exceeded\n1/2 the MCL / the number of years of monitoring data reported for\neach contaminant per CWS, and modeled as categorical variables comparing\n0% (reference), >0−≤10%, and >10% of years. One\nhalf\nof the MCLs are as follows: arsenic (5 μg/L), uranium (15 μg/L),\ngross alpha (7.5 pCi/L, not including radon and uranium), nitrate-N\n(5 mg/L), TTHM (40 μg/L).\nResidential duration at the enrollment\naddress was determined based on the residential history and self-reported\ninformation about duration at the current home at questionnaire 4.\nMean follow-up=19.0 years, total person-years: 1,139,582. For uranium,\nmean follow-up=19.0 years, total person-years: 1,072,419.\nModel 1 was adjusted for baseline\nage (years) + baseline age 2 .\nModel 2 = model 1 + BMI category\n(<25 kg/m 2 , 25-<30 kg/m 2 , >30 kg/m 2 , or missing), menopause status (pre-, peri/post-menopause,\nor missing), live births (no/yes/missing), oral contraceptive use\n(never/ever/missing).\nIn\ntime varying analyses of cumulative average exposures\nlagged\n5 years ( Table S3 ), we observed a positive\nassociation per doubling in nitrate concentrations for the high-grade\nserous histotype (HR = 1.06, 95% CI 1.00, 1.12) and positive associations\nby quantile of exposure (p-trend = 0.09). We observed a positive association\nwith all ovarian cancer for the 90th percentile of uranium exposure\n(HR = 1.15, 95% CI 0.84, 1.56; p-trend = 0.42) and for the high-grade\nserous histotype with a doubling in arsenic concentrations (HR = 1.06,\n95% CI 0.97, 1.16). Findings were generally similar compared to the\nmain analyses, although associations with nitrate, uranium, and arsenic\nwere attenuated; as with the main analyses, there were no associations\nwith TTHM. In posthoc analyses limited to the same residentially stable\npopulation as the main analyses, the magnitude of associations with\ntime-varying exposures were more consistent with results from the\nmain analyses ( Table S4 ).\nWe noted\nonly minor differences in the associations between a doubling\nin CWS exposures and ovarian cancer risk by menopausal status and\nBMI category ( Table S5 ). We observed no\nevidence of statistical interaction when we stratified by pre- and\nperi/postmenopausal status, though the HRs for arsenic were higher\namong premenopausal women. Arsenic was positively associated with\novarian cancer among participants with a BMI 25–<30 kg/m 2 (HR = 1.28, 95% CI 1.10, 1.49) but not among those with lower\nor higher BMI. We did not observe evidence of statistical interaction\nby smoking status ( Table S6 ).\nMedian\naverage concentrations of nitrate, uranium, and arsenic\nwere higher among participants living in nonmetropolitan areas; whereas\nmedian average TTHM concentrations were higher among participants\nin metropolitan areas. Nitrate, uranium, and arsenic were moderately\npositively correlated with each other, and negatively correlated with\nTTHM, in both nonmetropolitan and metropolitan areas (Figure S4).\nIn single contaminant and mixture analyses, we observed stronger associations\nin the risk of all ovarian and high-grade serous histotype cancers\nper increase in nitrate, uranium, and arsenic among nonmetropolitan\nparticipants compared to metropolitan participants, although there\nwas no evidence of statistical interaction (p-interaction ≥\n0.05) in fully adjusted analyses ( Table S7 ). In analyses stratified by CWS size, we observed stronger HRs for\nall ovarian and high-grade serous cancers per doubling in nitrate,\nuranium, and arsenic among medium-sized CWS compared with large and\nvery large CWS. There was no evidence for statistical interaction\nby CWS size, except for nitrate for all ovarian cancer risk (p-interaction\n= 0.05; Table S8 ). Associations with ovarian\ncancer risk were similar across groundwater and surface water sources,\nand we did not observe evidence for statistical interaction by water\nsource type ( p ≥ 0.05; Table S9 ).\nAssociations with nitrate, uranium, and arsenic\nwere stronger among\nparticipants in the lower two tertiles of vitamin C intake compared\nto those in the highest tertile (p-interactions > 0.05). HRs for\nhigh-grade\nserous cancer were also higher in the lower two tertiles of vitamin\nC intake but there were no statistical interactions (p interactions\n≥ 0.05; Table S10-B ). Though we\ndid not observe a statistical interaction between tertiles of red\nmeat intake and CWS nitrate, we observed somewhat stronger associations\nbetween nitrate and ovarian cancer among participants in the second\ntertile of total daily red meat intake (19.0–41.1 g/day; Table S11 ) compared to the other tertiles. Further,\nwe observed increased risk of high-grade serous cancer at higher quartiles\nof CWS nitrate among participants in the second tertile (p trend <0.05)\nof red meat intake.\nIn analyses of dietary intakes of nitrate\nand nitrite ( Table S12 ), we did not observe\nsignificant associations\nwith all ovarian cancers or the high-grade serous histotype for total\ndietary nitrate, total dietary nitrite, or for nitrite intake from\nplant or processed meat sources. However, we observed increasing risk\nof all ovarian cancer with increasing quartiles of dietary nitrite\nfrom animal sources (p trend <0.05).\nWe observed positive\nassociations per doubling of gross alpha levels\nfor all ovarian cancer (HR = 1.11, 95% CI 1.01, 1.22) (Table S13)\nthat were generally similar to the results for uranium. For regulatory\npurposes, gross alpha radioactivity is measured as the sum of α\nparticle activity, which may be released from uranium and other radionuclides.\nAs many CWS in California (particularly smaller systems) did not provide\nuranium measurements when gross alpha levels were <5 pCi/L, our\nfindings suggest that gross alpha levels, which were highly correlated\nwith uranium, may be a useful proxy for uranium exposure in epidemiologic\nanalyses in California.\nIn sensitivity analyses in which we\nstarted follow-up on January\n1, 2005, we observed results similar to those of our main single\ncontaminant and four contaminant mixture findings ( Table S14; Table S13 for gross alpha). We separately restricted\nour main single contaminant and four contaminant mixture analyses\nto participants whose 15 year average contaminant exposures were below\nthe corresponding MCLs; these findings were consistent with the main\nanalyses ( Table S15 ).\n\nIn this large prospective\ncohort of women in California, we evaluated\nexposures to frequently detected and regulated contaminants in public\nwater supplies and ovarian cancer risk. We observed positive associations\nbetween uranium and all epithelial ovarian cancers and between nitrate\nand the high-grade serous histotype. We observed the strongest positive\nassociations per IQR increase in the mixture that included nitrate,\nuranium, arsenic, and TTHM. The joint effect generally increased as\ncontaminants were added to the mixture model, suggesting that single\ncontaminants (i.e., arsenic and TTHM) that did not have a statistically\nsignificant effect individually may nonetheless contribute to an overall\ndrinking water mixture effect on ovarian cancer risk.\nFew previous\ncohort studies have evaluated the association between\ndrinking water contaminants and epithelial ovarian cancer risk. The\nIowa Women’s Health Study, a prospective cohort of postmenopausal\nwomen, estimated long-term average nitrate and disinfection byproduct\n(TTHM and haloacetic acids) exposures in CWS based on participants’\nenrollment address. They found a significant\nelevated risk of ovarian cancer with increasing quartiles of average\nCWS nitrate exposure (highest quartile ≥ 2.98 mg/L) in models\nadjusted for TTHM. TTHM and haloacetic\nacid exposures were not associated with ovarian cancer risk, and there\nwas no evidence for interaction with nitrate. We observed an elevated\nrisk of high-grade serous ovarian cancer among participants who had\n>10% of years of exposure to nitrate ≥ 5 mg/L (1.68, 95%\nCI\n1,17, 2.39); this is consistent with the evaluation in the Iowa Women’s\nHealth Study, which yielded an elevated risk for ovarian cancer participants\nwho had ingested water with nitrate levels >5 mg/L for at least\n4\nyears compared to those with no years of exposure at this level (HR\n= 1.60, 95% CI 1.06, 2.41). Similarly,\nnitrate was positively associated with ovarian cancer risk in the\nAgricultural Health Study (HR per 5 mg/L = 1.15,\n95% CI 0.96, 1.39), a prospective cohort of pesticide applicators\nand their spouses primarily using private wells for their drinking\nwater in rural areas of Iowa and North Carolina.\nStudies of carcinogenic effects of NOCs in multiple\nanimal species\nprovide biologic plausibility for an epidemiologic association between\ndrinking water nitrate and ovarian cancer risk. \n , , \n Prior epidemiologic evidence for other drinking\nwater contaminants and ovarian cancer risk are limited. To the best\nof our knowledge, no prior studies evaluated uranium in drinking water\nand ovarian cancer risk. Uranium is a potent toxicant associated with\nreproductive toxicity in animal and mechanistic studies, providing\nbiological plausibility for the novel epidemiologic associations we\nobserved for uranium. Animal studies\nshowed that uranium ingested from drinking water accumulates in the\novaries and causes DNA hypomethylation. In cell-based studies, uranium was associated with decreased germ\ncell density and an increased apoptosis rate in human fetal ovaries.\nA study of ovarian cancer mortality rates\nin an area of Chile that\nexperienced large fluctuations in arsenic levels in drinking water\nfound a reduction in ovarian cancer mortality during a period of high\nexposure (mean levels of 870 μg/L) compared to earlier and later\nperiods with lower arsenic levels (<100 ug/L). Animal studies indicate that transplacental exposure to\ninorganic arsenic can induce ovarian tumors in the offspring of mice,\nso timing of exposures may be important. A hospital-based case-control study of women with primary ovarian\ninsufficiency in China found that urinary arsenic levels were higher\namong cases compared to healthy controls matched by age and BMI. Primary ovarian insufficiency may be linked\nto an increased risk of ovarian cancer; evidence is limited. Arsenic is a known carcinogen associated with\nreproductive toxicity in human, animal and mechanistic studies, however\nfuture studies with historical measurements and a greater range in\nexposure are needed to determine whether arsenic is associated with\novarian cancer risk. \n , , −\nWe observed non-statistically significant effect\nmodification of\nthe associations of water uranium and arsenic stratified by vitamin\nC intake; specifically, the risk of all ovarian cancer was attenuated\namong participants in the highest tertile of vitamin C intake. Limited\nevidence suggests that vitamin C intake may decrease metal/metalloid\ntoxicity through multiple potential mechanistic pathways (e.g., oxidative\nstress) and through enhancing metal excretion in urine. Similar to our findings in the CTS, stronger\nassociations were observed between water nitrate and ovarian cancer\nat lower levels of vitamin C intake (10 years on their enrollment residence CWS, and comprehensive\nfollow-up and information on demographic, anthropometric, dietary,\nand medical factors, including annual linkages to California cancer\nand mortality registries. We estimated 15-year average drinking water\nexposures using robust water quality monitoring data and CWS distribution\nboundaries, which have been extensively described and were previously\nvalidated for use in this cohort using self-reported drinking water\ninformation at a later follow-up (Q6).\nOur study also had some limitations. The lack of historical\ndata\nbefore 1990 limited our ability to assess early life and long-term\nexposures in this study. We partly addressed this by limiting our\nanalyses to women who lived a minimum of 10 years at their enrollment\naddress. This approach assumes that the 15-year average reflects earlier\nexposures, which may have led to misclassification of exposure if\nthe concentrations were different prior to 1990. We expect that a\ntemporal overlap of the exposure (1990–2005) and follow-up\n(beginning in 1995) periods did not bias our findings, as we observed\nsimilar results in sensitivity analyses that started follow-up in\n2005. In time-varying analyses, we incorporated residential changes\nover time and computed a cumulative average exposure estimate that\nwas lagged 5 years. A 5-year lag was the maximum we could apply without\nextrapolating to time periods where data were not available; however,\na 5-year lag is likely insufficient to capture the latency period\nfor ovarian cancer. Residence duration\nwas asked in the fourth, fifth, and sixth questionnaires. For other\nquestionnaire addresses and addresses assigned by linkages, the move-in\ndate was estimated as the minimum date among matched records, which\nmay have led to misclassification of exposure. However, in analyses that evaluated time-varying exposures\nin the same residentially stable population as the main analysis,\nour findings were more similar to the results observed from using\nthe 15-year average drinking water exposure metric. Overall, we observed\nconsistent patterns across different drinking water exposure metrics.\nAdditional limitations include our exposure assessment based on\nresidential address rather than self-reported drinking water source.\nDrinking water source and tap water treatment were only collected\nat the sixth follow-up questionnaire (Q6, 2017–2019; response\nrate 43%); therefore, we were not able to use this information in\nour exposure assessment. However, among participants who provided\nthis information at Q6 ( N = 33,276), we found that\nself-reported and geocoded address-assigned water source had high\nagreement, indicating that exposure misclassification\nof the water source is not likely to be large. Specifically, among\nparticipants linked to a CWS based on their Q6 address, 74% reported\ntheir tap water source as municipal water, 2% as private well water,\n15% as bottled water, 4% as other, and 5% as do not know/missing. Participants who drank only bottled water may\nstill have exposure to inorganic contaminants via ingestion from tap\nwater used for cooking and consumption of beverages made with boiling\nwater and to THMs through inhalation and dermal exposure.\nFinally,\ndifferences in water quality by sociodemographic characteristics\n(e.g., by race and ethnicity, neighborhood urbanicity and SES) have\nbeen observed in the CTS and across California. \n , \n The exposure distributions in our study population (an educated\ncohort of women who predominantly lived in metropolitan (including\nsuburban) areas at enrollment and are predominantly non-Hispanic white)\nmay not reflect the full range of exposures to California women. This\nunderscores the need to replicate these analyses in additional study\npopulations to advance our understanding of ovarian cancer etiology.\n\nWe observed positive associations between\nCWS uranium and nitrate\nlevels and the risk of all epithelial and high-grade serous ovarian\ncancer, respectively. We also found a positive effect of the drinking\nwater contaminant mixture on all ovarian and high-grade serous cancer.\nAlthough future research is needed to provide additional epidemiologic\nsupport for these novel findings, our results suggest that uranium\nexposure through drinking water may be a novel risk factor for epithelial\novarian carcinomas and that nitrate in drinking water is a risk factor\nfor high-grade serous cancer. The identification of modifiable risk\nfactors for high-grade serous cancer is particularly important, as\nit has a poor prognosis compared to other histotypes. There is a critical\nneed for more epidemiological studies of gynecologic cancers and drinking\nwater contaminant exposures. If these findings are confirmed, population-level\ninterventions to reduce drinking water nitrate and uranium levels\nmay be potential opportunities to reduce ovarian cancer risk from\nenvironmental exposures.","source_license":"CC-BY-4.0","license_restricted":false}