Longer-term aspirin use and subsequent ovarian cancer risk in the Ovarian Cancer Cohort Consortium.

The OC3 is an international consortium of ≥20 prospective cohort studies 12 . For this analysis, we pooled data from 10 studies that collected questionnaire data on frequency of aspirin use at ≥2 time points: the NIH-AARP Diet and Health Study 13 , 14 , Cancer Prevention Study II-Nutrition Cohort 15 , 16 , California Teachers Study 17 , Iowa Women’s Health Study 10 , Nurses’ Health Study 18 , Nurses’ Health Study II 18 , Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial 19 , Southern Community Cohort Study 20 , Sister Study 21 , and Swedish Mammography Cohort 22 . Details of the design and characteristics of these cohorts have been described previously 12 . Data on aspirin use, covariates, and ovarian cancer case characteristics at baseline and follow-up were harmonized centrally at the OC3 Data Coordinating Center (Moffitt Cancer Center, Tampa, Florida) 12 . If frequent aspirin use was not ascertained at cohort entry, the first follow-up questionnaire collecting information on frequent aspirin use became the “baseline” time point for this analysis. Individuals were eligible for inclusion if they had at least one intact ovary, no history of ovarian cancer at baseline, responded to ≥1 questionnaire item on frequency of aspirin use, and were followed for ≥2 years. The OC3 was approved by institutional review boards at participating institutions. Frequency of aspirin use and covariates were self-reported via questionnaires. Each study collected aspirin use data at ≥2 questionnaire cycles (median=4, range=2–14; Figure 1 , Supplementary Table S1 ). For this analysis, frequent aspirin use was defined as use ≥5 days per week, for ≥6 months duration, regardless of dose ( Supplementary Table S1 ). Aspirin dose (low-dose [81–150 mg]; regular-dose [≥325 mg]; collected in 7 of the 10 studies at ≥1 questionnaire cycle) was also harmonized as available ( Figure 1 ). We constructed time-updated exposure metrics for frequent aspirin use, capturing ever use, current vs. former use, cumulative on-study years of use (categorized as ≤2, >2–6, >6 years, with >6 years considered “long-term” use), ever low-dose use, and ever regular-dose use ( Supplementary Table S2 , Supplementary Methods ). Consistent with our prior studies 6 , 8 , never frequent aspirin use was the referent category. Each exposure metric was updated when new questionnaire data became available ( Figure 1 , Supplementary Figure S1 ). The outcome of interest was primary epithelial ovarian cancer, defined as epithelial ovarian, fallopian tube, or peritoneal cancer. Incident cancers were ascertained via questionnaire, medical records, or cancer and death registry linkages. Most studies confirmed self-reported diagnoses through pathology review or linkage to cancer registries. Case characteristics (stage, grade, histotype) were obtained through pathology review, medical record abstraction, or registry linkages. Pooled logistic regression was used to examine the association between frequent aspirin use and incident ovarian cancer 23 – 25 . For this analysis, survival time (i.e., time between baseline and incident ovarian cancer, death, bilateral oophorectomy, loss to follow-up, or administrative censoring) was divided into two-year intervals. Exposure and covariate values were fixed within an interval but could vary across intervals, with updates incorporated when individuals’ statuses changed on their most recent follow-up questionnaire. Incident ovarian cancer within each period was treated as a binary outcome. The two-year intervals were then pooled, and logistic regression with the complementary log-log link was used to assess the risk of ovarian cancer within each interval. This approach estimates beta coefficients interpreted as log odds ratios (ORs) that are analogous to cause-specific hazards ratios 24 , 25 . This approach was utilized instead of Cox proportional hazards regression as it is more computationally efficient when there are many tied event times and time-dependent covariates 23 . For the primary analysis, frequent aspirin use and time-updated covariates were lagged by two years ( Supplementary Figure S1 , Supplementary Methods ). This was done to avoid reverse causation, because aspirin use initiated immediately prior to ovarian cancer diagnosis may have been used to manage pain from ovarian cancer symptoms. Person-time before the age of 30 was excluded due to sparse data. Models were adjusted for age at the start of each interval (continuous), body mass index (BMI; <20, 20-<25, 25-<30, 30-<35, ≥35 kg/m 2 ), duration of oral contraceptive (OC) use (none, <1, 1–4, 5–9, ≥10 years), number of live births (none, 1, 2, 3, ≥4), menopausal hormone therapy (MHT) use (none, former, current, unknown, premenopausal), and year of cohort enrollment (in 5-year categories). Analyses for low-dose use were also adjusted for regular-dose use, and vice versa. All covariates other than year of enrollment were treated as time-varying and were updated when new data became available. Data from all cohorts were pooled. In sensitivity analyses, we 1) stratified analyses by cohort, and combined cohort-specific effect estimates using random effects meta-analysis, 2) used time-updated Cox proportional hazards regression, 3) used lag periods of 0 and 5 years instead of 2 years, 4) used age 50 as the minimum age, 5) excluded individuals with cardiovascular disease (CVD) at baseline, 6) excluded individuals reporting use of non-aspirin nonsteroidal anti-inflammatory drugs (NSAIDs) at baseline, 7) excluded individuals missing covariate data (complete-case analysis), 8) assumed missing follow-up data for aspirin use was either all frequent aspirin use or all non-use (comparison of extremes). Secondary analyses examined associations for frequent aspirin use by ovarian cancer histotype, using an augmented data approach to account for the multi-level outcome and the loglikelihood test for significance of differences across histoypes 26 . We also fit a model with an interaction between frequent aspirin use and decade of age at the start of each interval to assess whether the association between frequent aspirin use and ovarian cancer risk is age-dependent. Finally, we tested for effect modification by obesity [BMI≥30 kg/m 2 ] and an ovarian cancer risk score, defined as the number of ovarian cancer risk factors present (obesity, history of endometriosis, family history of breast/ovarian cancer, nulliparity, no oral contraceptive use, no tubal ligation; Supplementary Table S3 , Supplementary Methods ). Analyses were conducted in SAS Studio 3.8 (SAS Institute Inc., Cary, NC, USA) and Stata 17 (StataCorp LLC, College Station, TX, USA).

There were 675,901 individuals included in this analysis. At baseline, 14% of individuals reported frequent aspirin use ( Table 1 ). By the end of follow-up, 36% of participants reported ever frequent aspirin use; of these ever users, 75% were current users and 25% no longer reported frequent aspirin use ( Supplementary Table S4 ). There were 5,528 incident ovarian cancers diagnosed during follow-up (median follow-up=13 years; median age at diagnosis=70 years among never frequent users, 74 years among ever frequent users). Of these cases, 2,908 (53%) were high-grade serous, 112 (2%) low-grade serous, 398 (7%) endometrioid, 186 (3%) clear cell, 226 (4%) mucinous, and 1,698 (31%) other epithelial or unknown histotype. Overall, there was no association between ever frequent use of aspirin and ovarian cancer risk (versus never frequent use; OR [95% CI]: 0.97 [0.91–1.03], Table 2 ). Similar null associations were observed for former (0.99 [0.88–1.12]) and current (0.97 [0.91–1.03]) use. However, individuals that were long-term (i.e., >6 cumulative years) frequent aspirin users during follow-up experienced a 14% reduction in ovarian cancer risk (≤2 years: 1.01 [0.94–1.08]; >2–6 years: 0.96 [0.84–1.10]; >6 years: 0.86 [0.77–0.97], p-trend=0.03). For long-term use, the association appeared stronger for non-high-grade serous (0.58 [0.45–0.74]) compared to high-grade serous (0.90 [0.77–1.05]) ovarian cancer (p-heterogeneity=0.02, Table 3 ). In the subset of individuals in the aspirin dose analyses (n=355,619; 1,547 ovarian cancer cases), 53% of frequent aspirin users reported low-dose use, 25% reported regular-dose use, and 22% had unknown dose at baseline ( Supplementary Table S4 ). An inverse association was suggested for ever frequent use of low-dose aspirin (0.90 [0.79–1.01]) but not regular-dose aspirin (1.09 [0.94–1.27], Table 2 ). Long-term use of low-dose aspirin was also inversely associated with risk (0.75 [0.56–0.99], Supplementary Table S5 ). In analyses stratified by the ovarian cancer risk score, associations for all aspirin metrics were close to the null for persons with ≤2 ovarian cancer risk factors ( Figure 2 ). However, for persons with ≥3 ovarian cancer risk factors (23% of the study population), ever (0.90 [0.80–1.01], p-interaction=0.02), current (0.89 [0.79–1.01], p-interaction=0.06), long-term (>6 years; 0.65 [0.50–0.85], p-interaction=0.02), and low-dose (0.84 [0.66–1.06], p-interaction=0.13) frequent aspirin use appeared associated with reduced risk ( Figure 2 ). Among individuals with ≥3 ovarian cancer risk factors, there was no evidence of heterogeneity by histotype ( Table 3 ). No interactions were observed with BMI ( Supplementary Table S6 ). When examining associations by age-strata, ever frequent aspirin use was associated with lower risk of being diagnosed with ovarian cancer at age 50-<60 years (0.80 [0.65–0.99]) but not at age 60-<70 years (1.04 [0.94–1.16]) or 70+ years (1.00 [0.92–1.08], p-interaction=0.045, Supplementary Table S7 ). At age 50-<60 years, risk reductions were also observed for current frequent aspirin use (0.76 [0.58–0.95]), use ≤2 years (0.74 [0.57–0.95]) and low-dose use (0.49 [0.31–0.77]). Results for ever frequent aspirin use were consistent in analyses stratified by cohort and meta-analyzed and in analyses using time-updated Cox proportional hazards regression ( Supplementary Table S8 ). Sensitivity analyses that considered varying lag periods, minimum ages, or exclusions for CVD and use of other NSAIDs were also consistent, with the exception of associations for low-dose aspirin use, which became more strongly inverse in analyses lagged by 5 years or beginning at age 50 ( Supplementary Table S9 ). Analyses incorporating extreme assumptions for missing data demonstrated that associations for ever and current frequent aspirin use were similar under both scenarios, suggesting that the imputation approach did not influence the results ( Supplementary Table S9 ). Additional sensitivity analyses described in the Supplementary Methods also produced similar findings ( Supplementary Figure S2 , Supplementary Table S10 ).

In this analysis of approximately 675,000 individuals from 10 prospective cohorts, each of which had repeated assessment of aspirin use and long-term follow-up for incident cancer diagnoses, ever frequent aspirin use was not associated with lower ovarian cancer risk. However, individuals who reported long-term frequent use (i.e., use across >6 years), and low-dose aspirin use had lower ovarian cancer risk during follow-up. These risk reductions were strongest among individuals with 3+ ovarian cancer risk factors, possibly confined to individuals diagnosed at ages 50–60, and robust across numerous sensitivity analyses. These results suggest that frequent, long-term use of low-dose aspirin may lower ovarian cancer risk among specific subgroups of at-risk individuals or within specific age periods, highlighting the opportunity for ovarian cancer precision prevention. Aspirin (acetylsalicylic acid) is a non-selective, irreversible inhibitor of the cyclooxygenase (COX) enzymes that produce pro-inflammatory prostanoids and thromboxanes 27 . Long-term use of aspirin is hypothesized to protect against ovarian cancer by inhibiting these pro-inflammatory mediators involved in inflammatory signaling, cellular proliferation, angiogenesis, invasion, and metastasis 28 , 29 . Additionally, aspirin may act through COX-independent pathways relevant to ovarian carcinogenesis, including inhibition of cell signaling via WNT/β-catenin, NF-κB, mTOR, and AMPK 30 , 31 . Laboratory studies and molecular analyses of ovarian tumors have also found that aspirin may enhance immune surveillance and/or reduce metastatic potential 32 , 33 . While there is biologic and mechanistic plausibility, studying the relationship between aspirin use and ovarian cancer risk has proven challenging. Observational studies have reported both null and protective associations 9 – 11 , 14 , 34 – 43 , and meta-analyses and pooled analyses have found frequent aspirin use to be associated with a 10–20% reduced risk of ovarian cancer 6 – 8 , 44 . However, many of the previous cohort analyses 6 , 8 , 10 , 11 , 14 , 37 , 40 relied only on baseline assessment of aspirin use (i.e., aspirin use at a single timepoint) and may have misclassified relevant exposed or unexposed person-time for individuals who changed medication use during follow-up. Data from the current study suggest that exposure misclassification resulting from the use of baseline data only could be substantial, as 22% of study participants reported frequent aspirin use during follow-up but not at baseline, and 25% of participants who reported frequent aspirin use at one or more time points no longer reported aspirin use by the end of follow-up. However, in the current study, reducing this source of exposure misclassification did not impact the results (HR=0.99 [0.92–1.07] for baseline frequent aspirin use vs. OR=0.97 [0.91–1.03] for time-updated ever frequent aspirin use, Supplementary Table S8 ). Our results for baseline frequent aspirin use are similar, albeit slightly weaker, than results from the two prior OC3 pooled analyses that included a similar set of cohorts but less follow-up time 6 , 8 , suggesting that associations for baseline aspirin use may wane with prolonged follow-up and/or inclusion of cases with older ages at diagnosis. Importantly, we also leveraged time-updated aspirin data to examine specific patterns of aspirin use over time in relation to ovarian cancer risk. Though associations were null for ever and current frequent aspirin use, long-term (>6 years) use was associated with a 14% reduced risk of ovarian cancer overall, and a 35% reduced risk among individuals with 3+ ovarian cancer risk factors. Of the previous cohort studies that incorporated time-updated aspirin use, some 34 , 35 but not all 43 also observed stronger protective associations for consistent and/or continuous long-term aspirin use. Low-dose use, including long-term low-dose use, was also associated with reduced risk in our study; given that the long-term aspirin users in this study primarily used low-dose aspirin ( Supplementary Table S4 ), it is difficult to disentangle whether dose or duration were driving the observed inverse associations for long-term use. Regardless, these findings suggest that frequent, long-term low-dose aspirin use is associated with lower ovarian cancer risk. The observed aspirin-ovarian cancer associations were modified by the ovarian cancer risk score. Specifically, there were null associations for frequent aspirin use for individuals with ≤2 ovarian cancer risk factors but inverse associations for individuals with ≥3 risk factors; a similar pattern was observed in a previous OC3 study of baseline aspirin use 6 . Some previous studies also noted differences in the magnitude of the aspirin-ovarian cancer risk by individual risk factors, including nulliparity, endometriosis, oral contraceptive use, pelvic inflammatory disease, and a composite measure of systemic inflammatory diseases 41 , 45 , 46 . These risk factors are hypothesized to contribute to ovarian cancer risk by increasing chronic inflammation 47 ; it thus seems biologically plausible that aspirin, an anti-inflammatory medication, might be more protective in the presence of these risk factors, and that the timing and duration of aspirin use matter in terms of mitigating risk. Because aspirin has potential serious side effects including increased risk of major gastrointestinal or intracranial bleeding, identifying subgroups of individuals most likely to benefit from aspirin use for primary prevention is of utmost importance 31 . Our results reaffirm that frequent aspirin use may be most efficacious for ovarian cancer prevention among individuals with other ovarian cancer risk factors, suggesting that targeting prevention efforts may be possible. There was also potential effect modification by age at diagnosis, with risk reductions only observed for individuals diagnosed at ages 50-<60, who therefore used aspirin at younger ages. These analyses were exploratory given that we did not know participants’ precise ages at initiation or cessation of aspirin use. However, the results are generally consistent with the recent ASPREE trial which reported no benefit of short-term low-dose aspirin use for cancer prevention when initiated at age 70 or older 5 . For CVD primary prevention, the United States Preventive Services Task Force currently recommends against low-dose aspirin use for persons ages 60 or older, but recommends individual decision-making for certain persons ages 40–59 48 . Understanding the benefits and harms of aspirin use in this same age interval may thus be most clinically relevant, and future work may seek to confirm the inverse association between aspirin use and ovarian cancer in this specific age group. Strengths of this study include the time-updated data for aspirin use and covariates, as discussed above. Missing data were minimized due to the carrying forward and backward of covariate data, and multiple sensitivity analyses were conducted to ensure the robustness of the findings and methodologic approaches. The pooling of data across ten cohorts also improved statistical power to detect associations for the rare outcome of ovarian cancer and bolstered the generalizability of the findings. Limitations include the lack of information on indication for aspirin use and the heterogeneous time intervals of aspirin exposure assessment across cohort, which hindered more detailed assessment of duration and age of use. Additionally, due to the observational study design, despite our efforts to carefully adjust for time-varying confounders, there may have still been residual confounding. In conclusion, this study suggests that long-term, frequent low-dose aspirin use is associated with lower risk of all ovarian cancer histotypes among individuals with multiple epidemiologic risk factors for ovarian cancer. Ovarian cancer prevention may be an added benefit of frequent low-dose aspirin use for specific at-risk subgroups. These results suggest specific patterns of aspirin use that are most strongly associated with ovarian cancer risk and may guide future research to tailor aspirin use recommendations.

Ovarian cancer is a leading cause of gynecologic cancer mortality 1 . Prognosis is typically poor as ovarian cancer is often diagnosed at late stages, when curative treatment options are limited 2 , 3 . Treatments for ovarian cancer are also costly and invasive. Prevention of ovarian cancer thus has the potential to reduce substantial morbidity and mortality. Frequent (i.e., daily/near daily) use of aspirin has shown promise in reducing risk of ovarian and other cancers. Secondary analyses of randomized controlled trials of daily low-dose aspirin use have observed reduced risk of female reproductive cancers after ≥3 years of follow-up 4 , but these trials had too few ovarian cancers to study ovarian cancer specifically 4 , 5 . Pooled analyses of observational studies have also found aspirin use to be associated with reduced ovarian cancer risk 6 – 8 , with the strongest associations observed for frequent and/or low-dose use 7 – 11 . A limitation of the observational evidence is that most prospective analyses only included a single assessment of aspirin use. This exposure metric fails to capture the time-varying nature of aspirin use and misclassifies exposure for participants who initiate or stop aspirin after ascertainment. Incorporating time-updated measures of aspirin use into prospective analyses can reduce exposure misclassification, thereby reducing bias in estimates of the aspirin use-ovarian cancer association, and allow for detailed examination into how the timing and duration of aspirin exposure may influence disease risk. In this study, we incorporate time-updated data into prospective analyses of frequent aspirin use and ovarian cancer risk in 10 cohort studies from the Ovarian Cancer Cohort Consortium (OC3) 12 .