Benign Breast Disease, NSAIDs, and Postmenopausal Breast Cancer Risk in the CPS-II Cohort

Benign breast disease (BBD) is classified into three pathological categories associated with progressively greater breast cancer risk: non-proliferative BBD, proliferative BBD without atypia, and atypical hyperplasia, which elevates risk up to four-fold (1, 2). Regardless of BBD category, BBD is frequently multifocal and comprised of a spectrum of pathologic lesions, including fibroinflammatory changes. We propose that nonsteroidal anti-inflammatory agents (NSAID) may lower breast cancer risk among women with BBD but have less impact among those without BBD. Consequently, studies of NSAIDs and breast cancer risk that do not stratify patients by BBD status may underestimate protection among those with BBD (3, 4).

The hypothesis that the increased breast cancer risk associated with BBD is mediated at least partly by inflammatory mechanisms that can be inhibited by anti-inflammatory agents is supported by prior reports. Higher levels of circulating c-reactive protein are linked to elevated BBD risk (5), and compared with normal donated tissues, lobules in BBD biopsies contain increased immune cells (6). We reported that among 3,080 Mayo Clinic patients who had a BBD biopsy, users of non-aspirin NSAIDs experienced lower breast cancer risk than nonusers (OR, 0.63; 95% CI, 0.46–0.85; P = 0.002), irrespective of BBD category, and with a suggested dose–response effect (7). Use of aspirin showed suggestive protection but did not reach statistical significance. Another analysis of 1,467 postmenopausal women with a history of BBD found that aspirin use was associated with lower BC risk, whereas risk for non-aspirin NSAIDs were suggestive of protection, based on analyses of limited power (8). Finally, inflammation has emerged as a “hallmark of cancer,” which may potentiate progression of early lesions (9), and two cohort analyses found that increased levels of urinary metabolites of inflammatory prostaglandins were associated with higher breast cancer risk (10–12).

Understanding whether anti-inflammatory strategies represent a promising approach for breast cancer chemoprevention among patients with BBD, specifically, is important for maximizing risk to benefit ratios because NSAIDs may produce adverse events, particularly among older women. Concerns about bleeding risks related to regular NSAID consumption, particularly among the elderly, have led to reappraisal of broad recommendations for their use in preventing cardiovascular disease (13, 14) in favor of targeted use among subsets of patients likely to receive net benefit (15, 16). Most BBD patients are ages 55 years or less (17), and thus at lower risk of bleeding-related complications, further supporting the potential for targeted “interception” (18) with anti-inflammatory agents. Accordingly, we analyzed data for BBD history, NSAID use, and incident breast cancer in the CPS-II cohort study to test the hypothesis that women with BBD represent a subgroup at increased risk of breast cancer in which anti-inflammatory agents may lower risk.

This analysis included postmenopausal women (age at baseline; median: 67 years; range: 45→90 years) residing in 21 states who participated in the CPS-II nutrition subcohort (19). Women completed a mailed baseline questionnaire (1992), and follow-up surveys every 2 years, beginning in 1997 with response rates exceeding 80%. We limited inclusion to women who completed surveys in 1997 (n = 87,254) because this was the first survey that asked participants about their history of BBD and details about NSAID use. We excluded 7,476 women who reported a prior history of breast cancer, and 7,703 with a history of other cancers; 688 unconfirmed as postmenopausal; 9,053 with uncertain BBD status and 1,968 who were censored in 1997; leaving 60,366 women in the final analysis. Follow-up information was available through June 30, 2015. Completion and return of surveys were accepted as written informed consent and the study was conducted in accordance with the U.S. Common Rule. The study was approved by the Institutional Review Board of the Emory University School of Medicine. Data were analyzed and reported according to the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines.

The 1997 survey queried whether “a physician has ever told you that you had fibrocystic/other benign breast disease?” and year of diagnosis, defined as before 1992, 1992 to 1993, 1994 to 1995, 1996 to 1997. To leverage our prospective study design, BBD status and other exposures were updated biennially via questionnaire. NSAID use was queried as, “During the past year, on average, how frequently have you taken the following: [generic name, plus examples].” Medications included aspirin regular or extra strength (e.g., Bufferin, Anacin, Bayer, Excedrin, Ecotrin), baby or low-dose aspirin, ibuprofen (Motrin, Advil, Nuprin, Mediprin), naproxen (Aleve, Naprosyn, Anaprox, Vimovo), other anti-inflammatory agents (Mobic, nabumetone, meloxicam, diclofenac, indomethacin), and Celebrex. Response options included “Never or less than once per month” or a write-in grid of “days per month” and “pills per day.” Pills per month among those reporting NSAID use was estimated as the product of pills per day times days per month. Use was also defined as: no recent use (no use in last two surveys), current use only (no use on last survey but using currently), or use in the two most recent surveys (doses dichotomized as less than or greater than 15 pills per month). Covariates were obtained via biennial questionnaires related to demographics; weight gain from age 18 to baseline; oral contraceptive use; smoking; age at menarche; age at menopause; use of menopausal hormones; family history of breast cancer; combined parity/age variable; mammographic screening; history of hypertension; history of diabetes; history of cardiovascular disease; energy expenditure [weekly metabolic equivalents hours (MET)] and alcohol consumption.

Cases were initially self-reported on a biennial follow-up survey. Cases who died before they could self-report were identified via linkage with the National Death Index. Verification of these cases and histopathologic characteristics of the cancer such as invasive versus in situ behavior and hormone receptor status were obtained from medical records and subsequent linkage with state cancer registries. Dates of deaths were obtained through linkage with the National Death Index.

Patients were followed from the 1997 survey until: death (unrelated to breast cancer), diagnosis of cancer other than breast cancer, initiation of tamoxifen therapy or self-reported but unverified breast cancer, loss to follow-up, diagnosis of endpoint (i.e., ductal carcinoma in situ or invasive breast cancer) or June 30, 2015. Patients with self-reported but unverified breast cancer (N = 243) were right-censored at their last date known to be cancer-free. Participants missing two consecutive surveys were censored at date of last completed survey; for individuals missing one survey, data were carried forward from the previous survey cycle. BBD status was analyzed as a time-dependent variable; specifically, participants who reported a new BBD diagnosis after baseline were moved in the “Yes, BBD” category from that point in follow-up time forward. Person-time for NSAID use and time varying covariates (i.e., mammographic screening, smoking, menopausal hormone use, METs, alcohol use, and chronic conditions) were updated at each survey and considered as “exposed” until the next survey, censoring or date of breast cancer diagnosis. Aspirin and non-aspirin NSAID use were determined as reported pills-per-month (PPM) usage and type of use and time-dependent usage variables were derived accordingly (Table 1).

Analyses were performed using Cox proportional hazards regression accounting for the time-dependent nature of the BBD and NSAID exposures. Age at baseline was included as a 27-category stratification term, modeled as <55, >78 years due to limited breast cancer diagnoses in high and low age groups, and as single year strata in between. We compared characteristics of participants who self-reported BBD at baseline or any time during follow-up versus those who did not, and then assessed associations of established breast cancer risk factors with incident breast cancer in multivariable analyses adjusted for potential confounders. Next, we assessed associations of NSAID use, overall and by formulation, frequency, recency, and pills per month in relation to breast cancer risk in the entire cohort and then within subsets defined by history of BBD. We assessed whether associations of NSAID use with risk of breast cancer differed by BBD status by fitting both the NSAID use variable of interest and BBD status as nominal main effect terms, and then fitting and testing the appropriate cross-product interaction term(s) using a single degree-of-freedom (for binary NSAID use variables) or multiple degree-of-freedom (for multicategorical NSAID use variables) Wald test. We assessed the functional form of the association between breast cancer risk and pills per month using restricted cubic splines.

We calculated population attributable fractions (PAFs) to estimate the population health impact that NSAID use may have on BC risk in BBD patients, based on the observed exposure probabilities and hazard ratios in our cohort. To translate the PAF to number of BCs eliminated for every 10,000 women with BBD, we assumed a lifetime risk of BC of 12.5% in the general population and used properties of the exponential distribution to back-transform this population survival function to a hazard function, assuming in turn 20, 25 and 30 years of post-biopsy follow-up. This hazard function was then multiplied by the observed BBD hazard ratio for breast cancer from our cohort and transformed back into a BBD-specific survival distribution. We then multiplied this survival function by the PAF and then by 10,000.

Exploratory analyses re-examined associations by subtypes of breast cancer characteristics (stage and ER status), such that patients developing breast cancer not of the subtype of interest or of missing subtype were right censored at date they were last known to be cancer-free. All data preparation and analyses were run in SAS version 9.4 (SAS Institute, Inc.). Associations with nominal (uncorrected) P-values <0.05 were considered statistically significant.

The data underlying this article will be shared on reasonable request to the corresponding author.

Participants included 23,615 who self-reported BBD and 36,751 without a history of BBD with median ages at baseline of 65 and 68 years (range: 45→90 years), respectively. Women with BBD reported current use of menopausal estrogens and menopausal estrogens plus progestin more frequently and had a higher percentage of recent mammographic screening at baseline than women without BBD (Table 2). Distributions of most variables were similar across BBD status. Distributions of demographic and clinical characteristics across levels of any NSAID use are provided in Supplementary Table S1.

Self-reported history of BBD was associated with increased risk of breast cancer overall (HR, 1.45; 95% CI, 1.36–1.55), with stronger risks for in situ (HR, 1.85; 95% CI, 1.57–2.17) versus invasive breast cancer (HR, 1.38; 95% CI, 1.27–1.49; P_int = 0.0012), but similar risks for ER-positive (HR, 1.33; 95% CI, 1.22–1.46) and ER-negative breast cancers (HR, 1.45; 95% CI, 1.13–1.85, P_int = 0.56). Associations of breast cancer risk with other demographic and clinical characteristics are displayed in Supplementary Table S2.

A total of 3,896 incident breast cancers were diagnosed over an average of 12.72 years of follow-up. In multivariable analysis without BBD stratification, use of NSAIDs (HR, 0.95; 95% CI, 0.88–1.02) and non-aspirin NSAIDs (HR, 1.00; 95% CI, 0.88–1.02) was not associated with breast cancer risk. Use of regular dose aspirin was associated with marginally reduced risk (HR, 0.90; 95% CI, 0.82–0.98), with heterogeneity in effect by recency and dose (P = 0.01), ranging from current use only and less than 15 PPM (HR, 0.75; 95% CI, 0.62–0.91) to current use only and 15+ PPM (HR, 1.03; 95% CI, 0.86–1.24; Table 1). Long-term aspirin use at 15+ PPM was also associated with decreased risk (HR, 0.85; 95% CI, 0.75–0.97).

In stratified analysis, lower breast cancer risk was found among women with reported BBD who used NSAIDs compared with never users (HR, 0.87; 95% CI, 0.78–0.97), but this association was not evident in women not reporting BBD (HR, 1.02; 95% CI, 0.92–1.13; P_interaction = 0.04, Table 3). The magnitude of effect was similar for ER-positive breast cancers (HR, 0.85; 95% CI, 0.74–0.97) and for ER-negative breast cancers (HR, 0.87; 95% CI, 0.59–1.29). Among women with BBD, most patterns of NSAID use (type, pills per month, recency) were associated with a greater reduction in breast cancer risk compared with nonusers than was found among women without BBD for similar comparisons. Among women with a reported history of BBD, use of aspirin-only or use of non-aspirin-NSAIDs-only was associated with statistically significantly lower breast cancer risk, whereas among women without reported BBD, breast cancer risk was not significantly reduced for these categories of use. Examination of restricted cubic splines revealed a consistent protective effect of NSAID use in BBD patients regardless of number of pills taken per month, albeit with wide confidence intervals that include the null value (Supplementary Fig. S1).

We estimate that universal NSAID use among patients with BBD would reduce the overall number of breast cancer events by 4.1%. Assuming a lifetime breast cancer risk of 12.5% and a breast cancer relative risk of 1.5 comparing women with BBD to the general population (as found in this and other large cohorts; ref. 20), this would result in 39 fewer breast cancers per 10,000 women with BBD assuming 20 years of post-biopsy follow-up, 48 breast cancers assuming 25 years of follow-up, and 57 breast cancers assuming 30 years of following.

Our analysis shows that postmenopausal women diagnosed with BBD who use NSAIDs experienced reduced breast cancer risk, whereas women who were never diagnosed with BBD and used NSAIDs did not experience lower breast cancer risk. Magnitudes of association were similar for ER-positive versus ER-negative breast cancers. Although the magnitude of protection was modest, this could impact breast cancer incidence importantly at the population level, given the high BBD prevalence and breast cancer incidence. Further, identifying subsets of patients with BBD with active pro-carcinogenic inflammation, particularly at younger ages, could focus preventive therapy on a group most likely to benefit and least likely to experience complications.

NSAIDs had a greater protective effect on breast cancer risk for patients with BBD compared with patients with non-BBD across most strata of NSAID use. In considering future breast cancer prevention strategies, defining a dose that reduces breast cancer risk while producing limited adverse effects would be critical. In one report, aspirin doses of 81 or 325 mg daily produced nearly equal reductions in prostaglandin metabolites in urine, suggesting anti-inflammatory effects at lower doses (21).

We suggest the magnitude of association between NSAID use and lower breast cancer risk among patients with BBD is underestimated in this and other studies. Radiologic features of BBD are extremely common, suggesting that many patients with BBD are not biopsied or specifically identified (22). Misclassification of controls as patients without BBD, when in fact BBD is present, could mask the protective effect of NSAID use in this group. Further, if NSAID use reduces breast cancer risk mainly among women with active inflammation, only a subset of patients with BBD may benefit from NSAIDs, but at levels greater than reported herein. Further, as NSAID use is not motivated by breast pathology, the timing of use does not necessarily coincide with the presence of active inflammation in BBD.

Data indicate that both inflammation and exposure to postmenopausal estrogens increase risk of developing BBD and breast cancer, and that there is crosstalk between these and other putative pro-carcinogenic pathways (23, 24). Prior cohort analyses have linked increased risk of BBD diagnoses to elevated serum levels of c-reactive protein, estrogen, insulin, and lower concentrations of adiponectin (5). In other analyses, elevated urinary levels of metabolites of prostaglandins predicted increased breast cancer risk among postmenopausal women who did not use NSAIDs (11), independent of circulating estrogen levels (10). Further, inflammation in fat is associated with enhanced aromatization of androgens to estrogens (25, 26) and postmenopausal NSAID use is linked to reduced circulating estrogen levels (27), suggesting relationships between hormonal and inflammatory mechanisms. Inflammation is also associated with metabolic dysregulation and insulin resistance, which may increase carcinogenesis (28, 29). Thus, blocking deleterious pro-carcinogenic inflammation could reduce breast cancer risk via direct and indirect mechanisms, acting both within the breast and systemically.

Detection of BBD in this analysis was associated with menopausal hormone therapy and mammographic screening. These data are consistent with the association of menopausal hormone use and increased breast cancer risk (30, 31), the lower risk of BBD among tamoxifen users (32) and the link between higher circulating hormone levels and proliferative BBD (5, 33). Mammographic screening is temporally linked to a large increase in the observed incidence of breast cancer, generally ascribed to increased detection (34), with potentially similar impact on BBD (35). However, differences in the frequency of screening did not account for differences in breast cancer risk between NSAID users with and without BBD in this analysis.

Associations of established breast cancer risk factors (including BBD) with incident breast cancer in this analysis are consistent with the literature. An analysis of 13,485 women with detailed pathology review in the Mayo BBD cohort (1967–2001) found an overall standard incidence ratio (SIR of breast cancer = 1.64; 95% CI, 1.56–1.74), with significantly increased risk for all categories of BBD, including non-proliferative lesions (17). Further, data suggest that both non-proliferative and proliferative BBD lesions are found more frequently in countries with higher versus lower rates of breast cancer (36). Thus, diagnoses of BBD represent an important challenge and opportunity for cancer control.

In a previous analysis of 3,080 Mayo Clinic patients with pathology-confirmed BBD between 1967 and 2001, we found that NSAID use was associated with a reduced risk of breast cancer (HR, 0.76; 95% CI, 0.59–0.99), with significant dose–response associations based on length and frequency of use. Results were strongest for non-aspirin NSAID use (HR, 0.63; 95% CI, 0.46–0.85), but were suggestively protective for aspirin use as well (HR, 0.88; 95% CI, 0.64–1.21). Our current findings for women with self-reported BBD show similar direction of effect for all NSAID use (HR, 0.87; 95% CI, 0.78–0.97), but with a slightly stronger effect for aspirin use than non-aspirin NSAID use. These mild differences based on type of NSAID use could be attributable to differences in the form of BBD ascertainment (confirmed vs. self-report), cohort period, reasons for NSAID use, differences in unmeasured factors, or simple random fluctuation.

Our PAF calculations are based on the premise that all women with BBD adopt regular NSAID use. We recognize that universal adoption is not feasible nor recommendable due to medical contraindications for some women and/or patient preferences. However, as noted above, we expect that the magnitude of association between NSAID use and breast cancer risk is underestimated in women with BBD due to undiagnosed disease, which would counteract the decreased public health impact due to nonuniversal adoption.

Strengths of this analysis include the large cohort size and assessment of time dependent exposures. Limitations include reliance on self-reported information, including BBD diagnoses. Further, generalizability is limited to white, postmenopausal, heavily screened women as reflected in our study population. Reported P values are uncorrected for multiple testing, which could lead to an increase in the probability of false positive results. However, the fact that our results are consistent with the published literature lends credence to our findings.

In conclusion, our analysis suggests that NSAID use may lower breast cancer risk among women with BBD but not necessarily among those without a history of BBD. These data have potential implications for understanding inconsistencies in reported associations of NSAID use and breast cancer risk in prior studies that did not stratify patients by BBD history. As aspirin has not shown desired reductions in cancer incidence and mortality in large trials (37), we propose that optimization of patient selection for aspirin use is critical. Thus, studies with controlled dosing of anti-inflammatory agents and collection of biological samples from women of diverse racial and ethnic backgrounds may be needed to define effects of drugs on deleterious pro-carcinogenic inflammation in the breast to guide development of precision prevention.

A.C. Degnim reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

M.E. Sherman: Conceptualization, resources, formal analysis, supervision, funding acquisition, writing–original draft, writing–review and editing. R.A. Vierkant: Conceptualization, formal analysis, supervision, methodology, writing–original draft, writing–review and editing. M. Masters: Formal analysis, writing–original draft, writing–review and editing. D.C. Radisky: Writing–review and editing. S.J. Winham: Writing–review and editing. A.C. Degnim: Writing–review and editing. C.M. Vachon: Writing–review and editing. A.V. Patel: Resources, data curation, funding acquisition, project administration, writing–review and editing. L.R. Teras: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

The opinions expressed are solely the responsibility of the authors and do not necessarily represent those of the Mayo Clinic, the American Cancer Society or the American Cancer Society – Cancer Action Network.

The authors express sincere appreciation to all Cancer Prevention Study-II participants, and to the members of the study and biospecimen management group. The authors would also like to acknowledge the contribution from central cancer registries supported through the Centers for Disease Control and Prevention's National Program of Cancer Registries and cancer registries supported by the National Cancer Institute's Surveillance Epidemiology and End Results Program. The study protocol was approved by the institutional review boards of Emory University, and those of participating registries as required. The authors assume full responsibility for all analyses and interpretation of results.

Funded in part by the American Cancer Society intramural research program (data collection and L.R. Teras, M Masters, A.V. Patel) and the NCI at the NIH (grant number P30 CA15083: M.E. Sherman, R.A. Vierkant, D.C. Radisky, S.J. Winham, A.C. Degnim, C.M. Vachon).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.