|Home | About | Journals | Submit | Contact Us | Français|
Epidemiologic data on the association between nonsteroidal antiinflammatory drugs (NSAIDs) and ovarian cancer risk have been inconsistent. The authors prospectively examined the association between regular use of aspirin and nonaspirin NSAIDs and ovarian cancer incidence among 197,486 participants of the Nurses’ Health Study (NHS) and the Nurses’ Health Study-II (NHS-II) over 24 and 16 years of follow-up, respectively. Information on aspirin was initially assessed in 1980 (NHS) and 1989 (NHS-II) and on nonaspirin NSAIDs and acetaminophen in 1990 (NHS) and 1989 (NHS-II) and updated throughout follow-up. The authors used Cox proportional hazards models adjusting for ovarian cancer risk factors. A total of 666 confirmed cases of epithelial ovarian cancer were identified over 2,790,986 person-years of follow-up. The hazard ratios associated with regular use of aspirin, nonaspirin NSAIDs, and acetaminophen were 1.11 (95% confidence interval (CI): 0.92, 1.33), 0.81 (95% CI: 0.64, 1.01), and 1.14 (95% CI: 0.92, 1.43), respectively. The authors did not observe a dose-response relation with increased frequency or duration of regular use of any of these medications and ovarian cancer incidence. The results did not differ substantially by tumor histology. In this large prospective study, the authors found no compelling evidence to support an association between regular use of aspirin, nonaspirin NSAIDs, or acetaminophen and ovarian cancer incidence.
Chronic inflammation was proposed to play a key role in ovarian carcinogenesis (1), and use of antiinflammatory nonsteroidal drugs (NSAIDs) was hypothesized to decrease ovarian cancer risk. NSAIDs are thought to decrease inflammation by inhibiting cyclooxygenases (2–5), enzymes involved in the synthesis of prostaglandins, which in turn can contribute to carcinogenesis by promoting cellular proliferation and inhibiting apoptosis (6, 7). In spite of the accruing experimental evidence implicating cyclooxygenase and its inhibition by NSAIDs in ovarian carcinogenesis (8–17), epidemiologic studies have been inconsistent (18–29).
Previously, we found ovarian cancer risk to be unrelated to aspirin use and inversely related to use of other NSAIDs. However, the lack of a clear dose-response effect and the relatively short follow-up (6 years, 83 cases) limited these analyses (26). Now with longer follow-up and a greater number of accumulated cases, we sought to reevaluate the association between NSAIDs and ovarian cancer in 2 large prospective cohorts, the Nurses’ Health Study (NHS) and the Nurses’ Health Study-II (NHS-II).
The NHS is composed of 121,700 US female registered nurses, aged 30–55 years at entry, who responded to a mailed questionnaire in 1976. The NHS-II included 116,609 US female registered nurses, aged 25–42 years at entry, who responded to a baseline questionnaire in 1989. These cohorts are described in detail elsewhere (30, 31). In both cohorts, follow-up questionnaires are mailed biennially to update information on risk factors and on newly diagnosed diseases. The follow-up rates for the NHS (1976–2004) and NHS-II (1989–2005) were 95.4% and 94.7%, respectively.
We excluded participants with a diagnosis of cancer (except nonmelanoma skin cancer) (n=3,940 (NHS); n =1,045 (NHS-II)) and those who had undergone bilateral oophorectomy (n=9,987 (NHS); n=2,263 (NHS-II)) or pelvic irradiation (n=72 (NHS); n=31 (NHS-II)) prior to baseline. Exclusions were updated biennially. After exclusions, a total of 197,486 (n=84,303 (NHS); n=113,183 (NHS-II)) participants remained in the analysis at the start of follow-up. This study was approved by the Committee on the Use of Human Subjects in Research at the Harvard School of Public Health and the Brigham and Women's Hospital (Boston, Massachusetts).
We identified incident cases of ovarian cancer from 1976 to 2004 (NHS) and from 1989 to 2005 (NHS-II). Discharge summaries and pathology reports were reviewed by a gynecologic pathologist unaware of exposure status. Deaths were identified through the National Death Index, family members, and the US Postal Service. We estimate that 98% of all deaths were ascertained (32, 33).
We identified 1,225 cases in the NHS and the NHS-II (15% by death follow-up). We could not obtain medical records for 136 reported diagnoses and rejected 191 diagnoses upon medical records review. Of the confirmed ovarian cancer diagnoses, 820 were of epithelial origin. We further excluded cases who reported another cancer prior to ovarian cancer diagnosis (n=72).
In the NHS, we first queried about use, number of tablets, and duration of aspirin use in 1980. Except for 1986, information on aspirin use has been updated biennially since 1980. The number of tablets was again assessed in 1982 (tablets/week); 1984 and 1988 (tablets/month); and 1994 and biennially thereafter (tablets/week). The frequency of aspirin use was assessed in 1984, 1988, 1990, 1992, and 1994 (days/month) and biennially since 1996 (days/week). In the NHS-II, we collected information on aspirin use in 1989 and biennially since 1993 and on frequency in 1993 (days/week), 1995 (days/month), and biennially since 1997 (days/week). From 2000 (NHS) and 2001 (NHS-II) onward, participants were additionally asked whether they took standard-dose (≥325 mg) or low-dose (≤100 mg) aspirin. We converted the reported tablets per week of low-dose aspirin into the equivalent tablets of standard dose.
We considered women currently taking aspirin at least twice per week as current regular users, as past users if they had used aspirin regularly in any previous questionnaire cycle, and as never regular users otherwise. To better estimate long-term intake, we calculated the cumulative average number of tablets per week as reported in all available questionnaires up to the start of each 2-year cycle, categorized as <1, 1–3, 4–5, 6–9, or ≥10 tablets per week. We estimated the cumulative duration of regular use in years on the basis of user status and questionnaire return dates (categorized as <1, 1–<5, 5–<10, 10–<20, or ≥20 years) and the cumulative number of aspirin tablets (categorized as <500, 500–999, 1,000–2,999, 3,000–4,999, or ≥5,000 tablets). The total duration and lifetime number of tablets were restricted to NHS participants because the total duration of aspirin use was not available in the NHS-II.
In the NHS, information on nonaspirin NSAIDs and acetaminophen has been consistently assessed since 1990. We queried about the frequency of nonaspirin NSAIDs and acetaminophen use in 1990, 1992 (days/month), and biennially since 1998 (days/week). Across questionnaires, nonaspirin NSAIDs referred to ibuprofen, naproxen, Indocin (Merck & Co., Inc., Whitehouse Station, New Jersey), Relafen (GlaxoSmithKline, Research Triangle Park, North Carolina), and ketoprofen. Information on individual formulations was unavailable. In 2000 and 2002, we additionally asked whether participants used cyclooxygenase-2 inhibitors including celecoxib and rofecoxib.
In the NHS-II, we assessed current regular use of other NSAIDs and acetaminophen in 1989 and biennially since 1993. We collected information on the frequency of use biennially since 1995 and 1999, respectively. Other NSAIDs referred to commonly used NSAID brands, including ibuprofen, naproxen, Midol (Bayer HealthCare, Morristown, New Jersey), Indocin, Relafen, and ketoprofen. In 2001 and 2003, we additionally queried about the use of selective cyclooxygenase inhibitors (celecoxib, rofecoxib).
Use of nonaspirin NSAIDs and acetaminophen was coded similarly to aspirin use (never, past, regular use). The cumulative average frequency of use and duration since 1990 (NHS)/1989 (NHS-II) were categorized as <2, 2–4, or ≥5 days per week and as <1, 1–<2, 2–<5, or ≥5 years, respectively.
We collected information on multiple established and potential risk factors of ovarian cancer at baseline and have updated most of these every 2 years. Age in months was calculated from the date of birth to the date of each questionnaire's return. Height and weight measurements were collected at baseline; current weight was updated biennially. The duration of oral contraceptive use was assessed in 1976 and updated until 1982 in the NHS, as well as biennially since study inception in the NHS-II. Parity was assessed in 1976 through 1984 in the NHS, as well as biennially since 1989 in the NHS-II. Tubal ligation was assessed from 1976 to 1982 and in 1994 in the NHS, as well as biennially since 1989 in the NHS-II. Information on menopausal status, postmenopausal hormone use, and history of hysterectomy was collected at baseline and updated biennially in both cohorts. We estimated the number of ovulatory cycles by subtracting age at menarche, 1 year for each pregnancy, and total duration of oral contraceptive use from age at menopause (or current age, if premenopausal).
We calculated person-years from the baseline questionnaire (1980 (NHS); 1989 (NHS-II)) return date to the date of ovarian cancer diagnosis, death, or end of study follow-up (May 31, 2004 (NHS) and May 31, 2005 (NHS-II)), whichever came first. Hazard rates were calculated by dividing the number of cases by the number of person-years in each category of analgesic use, and hazard ratios were calculated by dividing the rates for each category of use, number of tablets, frequency, or duration of use by the referent category of each exposure metric. We used Cox proportional hazards models (34) to estimate hazard ratios and 95% confidence intervals, stratifying jointly by age in months and time period in 2-year cycles. Multivariate-adjusted models controlled for duration of oral contraceptive use in months, parity, history of tubal ligation, menopausal status, and use of postmenopausal hormones. Further adjustments for age at menarche, age at first birth, hysterectomy, history of arthritis, family history of ovarian cancer, breastfeeding, regularity of menstrual cycles, smoking, and dietary factors did not change estimates substantially. We carried exposure information over for 1 questionnaire cycle if information on analgesic use was missing for the current cycle. Analyses including exposures as reported in each cycle versus analyses carrying exposure information for 1 cycle when current data were missing produced similar results.
We conducted formal tests of heterogeneity estimating Cochran’s Q statistics P values (35). We assessed for linear trend by modeling the number of tablets or days/week, years of use, and number of tablets linearly. In the primary analysis, information reported in 1 cycle was used to define exposure status in the subsequent 2-year cycle; hence, exposure was assessed on average 1 year prior to diagnosis. To assess exposure latency, we conducted secondary analyses considering analgesic intake reported 2 and 4 years prior to diagnosis to predict ovarian cancer incidence. For example, in the 2-year lagged analyses, analgesic intake reported in 1980 was used to assess ovarian cancer incidence from 1984 to 1986, analgesic intake reported in 1982 was used to assess incidence from 1986 to 1988, and so on.
To assess whether the associations between each analgesic and ovarian cancer incidence varied by factors including arthritis (no/yes), body mass index (<30/≥30 kg/m2), parity (nulliparous/parous), oral contraceptive use (never/ever), regularity of menstrual periods (mostly or very regular/mostly or very irregular), and lifetime number of ovulatory cycles (<360/≥360 ovulations), we tested interaction terms in multivariate models using the likelihood ratio test to compare models including the main effects and interaction terms against models including the main effects only. We also conducted secondary analyses to investigate analgesic associations for specific major ovarian tumor subtypes (i.e., serous/poorly differentiated, mucinous, and endometrioid ovarian tumors) and by invasiveness of the tumors (borderline/invasive). We used polytomous regression methods to test for differences in estimates for NSAID use, frequency, and duration and incidence of invasive and borderline ovarian tumors. All analyses were conducted by using SAS, version 9, software (SAS Institute, Inc., Cary, North Carolina). All P values were 2 sided.
We documented 607 and 141 epithelial ovarian cancer cases over 24 and 16 years of follow-up in the NHS and NHS-II, respectively. Our main analyses include a total of 666 (n=552 (NHS); n=114 (NHS-II)) ovarian cancer cases with information on aspirin use and 2,790,986 person-years of follow-up. In both cohorts combined, 84% of tumors were invasive. Serous/poorly differentiated tumors (hereafter referred to as serous) comprised the majority of tumors (55%); endometrioid, mucinous, and clear cell tumors comprised 15%, 9%, and 2% of tumors, respectively.
As shown in Table 1, NHS participants were more likely to use aspirin (31% of person-time) or nonaspirin NSAIDs (29%) than acetaminophen (21%), while NHS-II participants were more likely to regularly use nonaspirin NSAIDs (31%) than acetaminophen (21%) or aspirin (13%). Compared with nonusers, regular users of aspirin were more likely to have a history of arthritis or heart disease, to have ever used oral contraceptives, to use postmenopausal hormones, and to use other types of analgesics. Compared with nonusers, regular users of nonaspirin NSAIDs were more likely to have a history of arthritis, to have ever used oral contraceptives, to use postmenopausal hormones, and to use other analgesics.
Women currently using aspirin at least twice a week were at similar risk of developing ovarian cancer as those who never used aspirin regularly (hazard ratio (HR)=1.11, 95% confidence interval (CI): 0.92, 1.33) (Table 2). There was no evidence of a dose-response association with increasing number of aspirin tablets/week (for ≥10 tablets/week vs. <1 tablet/week: HR = 0.87, 95% CI: 0.61, 1.25; Ptrend= 0.53) (Table 2); duration of use (for ≥20 years vs. <1 year of regular use: HR = 1.02, 95% CI: 0.77, 1.33; Ptrend = 0.60); or number of lifetime tablets (for ≥5,000 tablets vs. <500 tablets: HR = 1.00, 95% CI: 0.79, 1.27; Ptrend= 0.57) (Table 3). Results did not differ significantly by cohort (Pheterogeneity>0.17). Age-adjusted and multivariate-adjusted estimates (Tables 2 and and3)3) and analyses restricted to invasive cases (n=562) (data not shown) produced similar results. Additionally adjusting for use of other NSAIDs did not substantially change the estimates. To address the effect of latent disease on aspirin intake, we lagged exposure time by 2 and 4 years; the results were unchanged. For example, in the 4-year lagged analysis, the multivariate hazard ratio for regular use versus nonuse was 1.00 (95% CI: 0.82, 1.22); for ≥10 tablets/week versus <1 tablet/week, it was 0.80 (95% CI: 0.55, 1.15) (Ptrend=0.24); and for ≥20 years versus <1 year of regular use, it was 1.06 (95% CI: 0.80, 1.41) (Ptrend=0.71).
As shown in Table 4, regular use of nonaspirin NSAIDs also was not significantly associated with ovarian cancer incidence (for regular use vs. no regular use: HR=0.81, 95% CI: 0.64, 1.01). There was no evidence of a dose-response relation with increasing frequency (for ≥5 days/week vs. <2 days/week: HR = 0.88, 95% CI: 0.59, 1.32; Ptrend=0.64) or duration of use (for ≥5 years vs. <1 year of regular use: HR = 0.92, 95% CI: 0.67, 1.25; Ptrend= 0.53). Age-adjusted and multivariate-adjusted results were similar. Results did not differ significantly by cohort (Pheterogeneity>0.43) except for the upper category of nonaspirin NSAIDs’ frequency (Pheterogeneity=0.02), likely because of the limited numbers of cases in this category (n=16 (NHS) and n=7 (NHS-II)). Results from secondary analyses restricted to invasive ovarian cancer cases or to participants using only 1 class of analgesics (ensuring that concurrent use of other classes of analgesics did not influence our estimates) were essentially unchanged. Additional analyses using 2- and 4-year lags between nonaspirin NSAIDs and ovarian cancer risk produced similar associations. For example, in 4-year lagged analysis, the multivariate hazard ratio for regular use of nonaspirin NSAIDs was 0.92 (95% CI: 0.70, 1.19) and for regular use for 5 or more days/week was 0.96 (95% CI: 0.60, 1.53) (Ptrend=0.93).
As depicted in Table 4, current regular use of acetaminophen was not associated with ovarian cancer incidence (for regular use vs. no regular use: HR=1.14, 95% CI: 0.92, 1.43). Increasing frequency (for ≥5 days/week vs. <2 days/week: HR = 1.13, 95% CI: 0.70, 1.82; Ptrend=0.70) or duration of use (for ≥5 years vs. <1 year of regular use: HR = 0.86, 95% CI: 0.55, 1.37; Ptrend=0.83) did not influence the incidence of ovarian cancer. Lagging acetaminophen exposure by 2 and 4 years yielded similar results (data not shown).
There was no evidence that the analgesics–ovarian cancer associations varied by factors including history of arthritis, body mass index, parity, oral contraceptive use, regularity of the menstrual cycle, or lifetime number of ovulations (Pinteraction>0.18), although the power of these analyses was somewhat limited. Further, results did not vary substantially by histologic subtype of tumor.
NSAIDs may affect risk differently according to the invasiveness of the tumor (22). Although estimates for invasive tumors were similar to our main analyses, regular use of NSAIDs (aspirin and nonaspirin) was associated with a nonsignificant decrease in risk of borderline ovarian tumors (n=62) in a comparison of regular versus not regular use, with a multivariate hazard ratio of 0.58 (95% CI: 0.32, 1.06) (Table 5). Results were more pronounced with increased frequency (for ≥4 days/week vs. <2 days/week: HR = 0.24, 95% CI: 0.06, 0.98; Ptrend=0.01) and duration (for ≥2 years vs. <1 year of regular use: HR = 0.47, 95% CI: 0.26, 0.87; Ptrend = 0.03) of regular use of NSAIDs. There was significant heterogeneity of the analgesic–ovarian cancer association according to tumor invasiveness (P<0.01).
We did not assess the reasons for analgesic intake in the NHS cohorts. However, in a subset of 3,876 NHS and 4,024 NHS-II women (36, 37), the most cited reasons for use of aspirin were cardiovascular disease prevention (69% and 43% in NHS and NHS-II, respectively), muscle/joint pain (30% and 31%), and headache (22% and 50%); for use of other types of NSAIDs, the most cited reasons were muscle/joint pain (84% and 68%), backache (26% and 36%), and headache (15% and 46%). The most commonly used nonaspirin NSAID was ibuprofen (57% and 78% in the NHS and NHS-II). Restricting analyses to participants with a history of arthritis and those with a history of cardiovascular disease produced essentially unchanged results, although these analyses had limited statistical power.
The results of this large, prospective study do not support an association between current regular NSAID use and ovarian cancer incidence. Our results may suggest an inverse association between NSAID use and incidence of borderline ovarian tumors, although these findings were based on a limited number of cases.
In the current study, regular use of aspirin was unassociated with ovarian cancer incidence. There was no evidence of a dose-response relation with increasing number of tablets/week or total duration of use. Our findings are in agreement with those of several previous studies (18–22, 26–28) and with the results of 2 quantitative reviews that included 3 prospective and 6 retrospective studies (38, 39). To our knowledge, only 3 retrospective studies reported a significant inverse association between aspirin use and ovarian cancer risk (23–25), including 1 study in which analysis was based on only 1 exposed case (23). Although retrospective studies are subject to differential recall of analgesic use between cases and controls, it is probable that cases’ tendency to better recall analgesic intake would result in odds ratios greater—not lower—than unity. In addition, response rates among controls were lower than those among cases (24, 25), suggesting the potential for bias, although the direction of any such bias is uncertain.
Our findings also do not provide compelling evidence for an association between nonaspirin NSAID use and ovarian cancer. Only a few studies have examined the effects of nonaspirin NSAIDs; our results are in agreement with some (19, 27) (cyclooxygenase-2 inhibitors (25)), but not all (24, 26, 29), previous studies. Heavy use of NSAIDs was associated with a substantial decrease in risk in a large hospital-based case-control study (24). However, it is possible that heavy users of analgesics were overrepresented among hospital controls, as they may be more likely to seek medical care and to be hospitalized. Nonaspirin NSAIDs were also associated with a borderline significant 10% decreased risk in a prospective record linkage study (29), although these estimates were not adjusted for potential confounding factors, such as oral contraceptive use. We previously reported that use of nonaspirin NSAIDs was associated with a 40% reduction in ovarian cancer incidence, although no dose-response with increased frequency of use was found (Ptrend= 0.13) (26). Our current analysis, which includes an additional cohort (NHS-II), 8 more years of follow-up, and approximately 5 times the number of cases, provides no convincing evidence to support an association between nonaspirin NSAIDs and ovarian cancer risk.
We also examined whether the association between analgesics and ovarian cancer differed according to tumor histology. This is particularly informative for mucinous ovarian tumors because of their histologic resemblance to colon cancer (40), which has been inversely related to the use of NSAIDs (41, 42). However, our results did not substantially differ by histologic subtypes of tumors, although only 62 and 40 mucinous cases were available for the analyses of aspirin and nonaspirin NSAIDs, respectively.
Our results suggested an inverse association between regular use of any NSAIDs and the incidence of borderline ovarian tumors, which was more pronounced with increased frequency and duration of use. Borderline ovarian tumors are of considerable interest because, despite being rare and having more favorable prognosis than invasive tumors, they tend to affect younger women who must consider the risks of fertility-sparing treatments. Our findings are in agreement with the results of a recent case-control study (22). Other risk factors for ovarian cancer may also differ for borderline tumors (43–45), although the biologic reasons for these differences remain poorly understood. Our findings are based on a limited number of cases (n=62) and must be interpreted cautiously.
Regular use of acetaminophen was unrelated to ovarian cancer incidence in this analysis. Although acetaminophen and NSAIDs are generally used in similar circumstances, acetaminophen has weak antiinflammatory properties. Thus, examining the association between acetaminophen and ovarian cancer incidence may help to ensure that our main results would not be affected by behaviors associated with the intake of pain relievers. Nonetheless, a protective effect for acetaminophen has been suggested in some retrospective studies (19, 20, 24, 25), which was not replicated in the current analysis.
Potential limitations of our study include the use of self-reported data. In addition, analgesic information was not collected consistently in every questionnaire. Although exposure misclassification is probable, it is unlikely that it differed between those who subsequently did and those who did not develop ovarian cancer. This misclassification may have attenuated our findings and precluded us from observing modest associations. However, our analyses focused on regular users of NSAIDs, who are less likely than sporadic users to misreport intake. Moreover, inverse associations between NSAIDs and colorectal cancer/adenoma risk have been reported in the NHS cohort (46–48), consistent with results from randomized clinical trials (49–52), suggesting that we have reasonably valid measures of NSAIDs use. Our study focused on the most common types of NSAIDs, and it is possible that users of less common formulations of NSAIDs were classified as nonusers, attenuating our results. However, restricting our analyses to participants without a history of arthritis, who may be less likely to use less commonly used/prescribed analgesics, did not substantially change our results. Finally, our study population is largely white, and our results may not be generalizable to women of other ethnicities.
Ours is the largest prospective study conducted to date to examine the association between regular use of both aspirin and nonaspirin NSAIDs and ovarian cancer incidence. The use of repeated measures over a long follow-up allowed us to conduct analyses lagging the time between exposure and ovarian cancer incidence by several years, which is particularly important as the latency period of ovarian cancer is not yet well understood. The prospective nature of our study precludes recall bias, and our ability to identify deaths with high accuracy (32, 33) limits selection bias resulting from loss to follow-up. Moreover, we were also able to account for many known and suspected risk factors of ovarian cancer in these analyses.
The epidemiologic evidence provided by this and other epidemiologic studies that NSAIDs do not reduce the risk of ovarian cancer contrasts with experimental data that NSAIDs may decrease proliferation of cancer cells in culture and inhibit tumor growth in animals. The reasons for the discrepancies between experimental and epidemiologic data are not completely understood, although differences in exposure dosages, generally substantially higher in experimental studies, may play a role. It is also possible that NSAIDs may play a role in the progression of ovarian cancer but are not critical for its initiation. Finally, our data suggest a possible inverse association between use of NSAIDs and the incidence of borderline ovarian tumors, but these results were based on a limited number of cases and need to be replicated in future studies.
Author affiliations: Department of Medicine, Channing Laboratory, Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts (Simone P. Pinheiro, Shelley S. Tworoger, Bernard A. Rosner, Susan E. Hankinson); Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts (Simone P. Pinheiro, Shelley S. Tworoger, Daniel W. Cramer, Susan E. Hankinson); Obstetrics and Gynecology Epidemiology Center, Brigham and Women's Hospital, Boston, Massachusetts (Simone P. Pinheiro, Daniel W. Cramer); and Department of Biostatistics, Harvard School of Public Health, Boston, Massachusetts (Bernard A. Rosner).
The work described in this paper was substantially supported by National Institutes of Health research grants P50 CA105009, CA 50385, and P01 CA87969.
The funding source had no role in the study design, data collection, data analysis, or interpretation of the findings.
Conflict of interest: none declared.