Breast Cancer Risk and Use of Nonsteroidal Anti-inflammatory Agents After a Benign Breast Biopsy

Over one million women are diagnosed with benign breast disease (BBD) in the United States annually, which defines a large and expanding pool of patients at increased breast cancer risk (1). Implementing effective breast cancer risk reduction strategies among these women could substantially reduce breast cancer incidence and mortality.

BBD includes morphologically and biologically diverse lesions, several of which may be present concurrently in a single biopsy. To facilitate clinical management, BBD biopsies are classified as nonproliferative, proliferative without atypia or atypical hyperplasia, denoting progressively increasing levels of breast cancer risk. Over 90% of BBD biopsies do not show atypia, and despite conferring only a 1.5- to 2.0-fold increase in breast cancer risk, nonatypical BBD represents the most frequent diagnosis preceding breast cancer (1, 2). Atypical hyperplasia increases breast cancer risk approximately 4-fold, but comprises less than 10% of BBD diagnoses (1, 2). BBD categories define risk among groups of women, but individual risk varies greatly within BBD categories (3), and we hypothesize that several putative carcinogenic mechanisms underlie risk of developing BBD and its progression to breast cancer.

Increased circulating estrogen levels are associated with proliferative BBD, delayed age-related involution of normal lobules from which BBD arises, and increased breast cancer risk (4–6). Endocrine-based prevention lowers risk of estrogen receptor (ER)-positive breast cancer by approximately 50% (7); however, putative nonhormonal carcinogenic mechanisms, including chronic inflammation and immune responses, are not directly targeted by these agents. Sterile inflammation and aging (“inflammaging”) and other processes such as obesity can contribute to a proposed “inflammogenesis” model of breast cancer (8–10). In this model, cyclooxygenases catalyze prostaglandin production, which drives inflammation and fibrosis, and activates multiple downstream carcinogenic mechanisms. In preclinical breast cancer models, activation of inducible COX-2 increases procarcinogenic inflammation; treatment with COX-2 inhibitors, such as nonsteroidal anti-inflammatory agents (NSAID), reduces inflammation and inhibits breast cancer development (11–15). In human studies, inflammatory mediators have been strongly implicated in breast carcinogenesis (16, 17), and elevated COX-2 expression in atypical hyperplasia has been associated with increased breast cancer risk (18). Furthermore, inflammation increases estrogen production via activation of aromatase and regular NSAID use is associated with lower circulating estrogen levels (19). Nonetheless, meta-analyses of studies of average risk women in the general population provide modest evidence that NSAIDs lower breast cancer risk, with significant heterogeneity in results among individual studies and multiple limitations, as noted in these reviews (20–26). However, a notable limitation in nearly all studies of NSAID use and breast cancer risk is a lack of information regarding whether women had procarcinogenic inflammation in their breast tissues when medications were used.

Cyst formation and chronic inflammation are integral components of BBD, as reflected in the historical term “chronic cystic mastitis”. Cysts lined by apocrine cells frequently contain fluid with high levels of androgens, inflammatory mediators and enzymes with both prostaglandin synthesizing (COX-2) and degrading functions (15-hydroxy prostaglandin dehydrogenase; refs. 27–29). Cysts are often associated with duct ectasia, consisting of dilated structures containing fluid and macrophages and encircled by mononuclear cells and fibrosis. An analysis of 1,467 patients with BBD, including 91 women who developed incident breast cancer, found that aspirin use was associated with significantly reduced breast cancer risk [OR = 0.46; 95% confidence interval (CI) = 0.22–0.98], whereas nonaspirin NSAIDs did not show significant protection (OR = 0.82; 95% CI = 0.41–1.62; ref. 30). These data suggest that NSAID use reduces breast cancer risk among patients with BBD, but the small sample size and lack of information on medication doses, follow-up time, and detailed BBD features limit conclusions. Accordingly, we analyzed associations between NSAID use and breast cancer risk in the Mayo Clinic BBD cohort study, which includes detailed breast cancer risk factor data with pathology review and long-term follow-up (2, 31).

The Mayo BBD Cohort currently includes 13,455 women, ages 18–85 years, who underwent BBD biopsies between 1967 and 2001 at Mayo Clinic (Rochester, MN; refs. 2, 31). Women who had not been diagnosed with invasive or in situ breast cancer prior to or 6 months following BBD biopsy, had not undergone mastectomy or breast reduction and had not used prevention therapy were eligible. This analysis includes 3,080 (69%) of 4,482 potential participants biopsied between January 1, 1992 and December 31, 2001 who completed the study questionnaire that included self-reported NSAID use; prior to 1992 NSAID use was not queried. Mean age at biopsy (SD) was 50.3 (13.5) years with mean follow-up of 16.4 (6.0) years. Incident breast cancers (including ductal carcinoma in situ) were ascertained from study questionnaires, tumor registry, and review of medical records. The study protocol, including patient contact, written consent in the context of completing questionnaires, and patient follow-up, was approved by the Mayo Clinic Institutional Review Board, and in accord with the Common Rule.

Demographic and clinical characteristics and established breast cancer risk factors were collected using study questionnaires and comprehensive medical record review (2). Family history of breast cancer was categorized as strong, weak, or negative. A strong family history was defined as the patient having (1) at least one first-degree relative with breast cancer diagnosed before age 50 years or (2) two relatives with breast cancer at any age, with at least one being a first-degree relative. Breast cancer risk factors were collected using a baseline study questionnaire and comprehensive medical record review. Participants were asked, “Have you ever regularly take any nonsteroidal anti-inflammatory drugs (see names below), either over-the-counter or by prescription? (exclude occasional use of less than once per month). Specific drug names were listed, followed by fillable tables, in which the name of each drug was listed over three column headers “Average days per month used,” “On days used, average number of pills taken,” and “Total number of years taken.” Each of the three metrics of use was associated with “two-column bubble forms” for entering numerals from 00 to 99.”

Study coordinators classified agents into the following categories: ibuprofen and other prostaglandin synthesis inhibitors; naproxen, celecoxib, rofecoxib and other COX inhibitors; aspirin; other NSAIDs; and non-NSAIDs. Non-NSAID drugs were excluded from analyses.

Diagnostic BBD biopsies were reviewed in detail and specific proliferative (e.g., ductal hyperplasia) and nonproliferative lesions (e.g., cysts, apocrine metaplasia, duct ectasia, columnar cell change) were recorded using a standardized form, masked to follow-up (DWV; refs. 2, 31, 32). An overall BBD classification based on the highest risk category was also assigned: nonproliferative, proliferative without atypia, or atypical hyperplasia. Level of lobular involution (LI), a feature inversely related to breast cancer risk, was assessed as: noninvoluted (0%); partial involution (1%–75%); or complete involution (>75%; ref. 32).

Associations for different non-aspirin NSAID classes were similar (see below) and these data were combined. We examined ever use overall and by number of years used, average number of days used per month, and total number of doses. Total dose was calculated by multiplying together average number of days used per month, average number of pills taken on days used and total number of years taken. Metrics of NSAID use were stratified as never used, and among users, in tertiles. We compared frequencies of key epidemiological features among NSAID users and nonusers using χ² tests of significance. Among women with breast cancer, we compare NSAID use by breast cancer subtype using χ²tests.

Women were followed from date of BBD biopsy to either date of breast cancer diagnosis or censoring events, including death, prophylactic mastectomy, reduction mammoplasty, diagnosis of lobular carcinoma in situ or last contact. Patients with lobular carcinoma in situ (n = 5) were censored because of routine use of mastectomy during the period of the study. We compared breast cancer risk across use of NSAIDs using Kaplan–Meier curves (Supplementary Fig. S1) and then in Cox proportional hazards regression models with estimated HRs and 95% CIs. We performed unadjusted analyses and multivariate Cox regression models, adjusted for covariates, including: age at biopsy, BBD classification, level of LI, body mass index (BMI; kg/m²) and number of years between biopsy and questionnaire. We assessed dose–response effects related to frequency of NSAID use with breast cancer risk using tests for trend, modeling the ordered frequency of use variable as a one degree-of-freedom term in the Cox regression models. Analyses were carried out overall and by subgroups defined by demographic and clinical variables. We confirmed proportional hazards assumptions by fitting time-by-NSAID use interactions. We ran additional Cox regression analyses to assess, in turn, the potential confounding effects of Breast Imaging, Reporting and Data System (BI-RADs) mammographic density (33) and use of menopausal hormone replacement. We fit Fine and Gray proportional subdistribution hazards models to determine the potential impact of treating death as a competing risk rather than a censoring variable (34).

To account for the potential for NSAID use recall bias, we ran Cox regression sensitivity analyses that excluded follow-up prior to date of questionnaire, such that individuals with censoring or breast cancer events that preceded date of questionnaire were excluded from these analyses. We ran sensitivity analyses for specific drug classes, adjusted for use of other drug classes. All statistical tests were two-sided, and all analyses were carried out using SAS Studio, Release 3.7 (SAS Institute, Inc.).

Of the 3,080 (69%) women who provided NSAID use data, 312 women were diagnosed with incident breast cancer (230 invasive) over a median of 9.9 (SD, 5.9) years of follow-up. Breast cancer risk was not significantly different among women who provided NSAID data and those who did not and were excluded from the analysis (Cox model P = 0.46). Approximately 38% of women reported some level of regular NSAID use with minimal variation by year of BBD biopsy and similar levels of use for specific drug classes (Table 1). Increased total lifetime number of NSAID doses was higher among women 45 years of age and older at biopsy (P < 0.001), obese women (P < 0.001), and women with proliferative BBD (either with or without atypia) versus nonproliferative BBD (P = 0.02). Positive family history was marginally associated with increased NSAID use. Year of biopsy, level of LI, and age at first birth were unrelated to NSAID use.

Regular NSAID use overall was associated with lower breast cancer risk (HR = 0.76; 95% CI = 0.59–0.99; P = 0.039) with trends suggesting lower risk with increased use (comparing highest tertile for each category of use versus non-use for each metric). Specifically, for number of years used: HR = 0.76; 95% CI = 0.52–1.10; P_trend = 0.058, for days used per month: HR = 0.49; 95% CI = 0.18–1.33; P_trend = 0.015 and for total number of doses taken: HR = 0.65, 95% CI = 0.43–0.98; P_trend = 0.019 (Supplementary Table S1). Lower risk was associated with regular use of nonaspirin NSAIDs (but not aspirin), yielding HR = 0.63 (95% CI = 0.46–0.85; P = 0.002; Table 2). Compared with nonuse, all metrics of non-aspirin NSAID use showed dose–response effects, including years of use [HR = 0.55; 95% CI = 0.31–0.97; P_trend = 0.003], days used per month (HR = 0.51; 95% CI = 0.33–0.80; P_trend = 0.001) and total lifetime doses (HR = 0.53; 95% CI = 0.31–0.89; P_trend = 0.003). Similar patterns for ever use and increased use were found for Ibuprofen and naproxen and their respective related agents, whereas regular aspirin use was not significantly related to breast cancer risk (Supplementary Table S1).

Reduced breast cancer risk with regular non-aspirin NSAID use was found among women with nonproliferative BBD (HR = 0.52; 95% CI = 0.31–0.87; P = 0.013); associations for proliferative BBD with or without atypia were similar in direction, but failed to reach significance (Table 3). Similarly, significantly decreased risk with regular nonaspirin NSAID use was found among women ≤50 years of age (HR = 0.56; 95% CI = 0.36–0.87; P = 0.010), parous women (HR = 0.64; 95% CI = 0.46–0.89; P = 0.007), and obese women (HR = 0.45; 95% CI = 0.0.25–0.80; P = 0.007), whereas risks were nonsignificantly reduced among women who were >50 years of age, nulliparous, or nonobese, albeit in some analyses that included few events. Lower breast cancer risk was associated with several features included in the spectrum of nonproliferative BBD (cysts, apocrine metaplasia, duct ectasia, and columnar cell change) and were markedly lower for women with ≥3 such lesions (HR = 0.33; 95% CI = 0.17–0.65; P = 0.001). Among patients with BBD who later developed breast cancer, risks for NSAID use did not vary by estrogen receptor, progesterone receptor, or nodal status of the breast cancer (Supplementary Table S2).

Associations of ever versus never regular non-aspirin NSAID use with risk of breast cancer were similar when subsetting to proliferative BBD and atypical hyperplasia (see Table 3) and in analyses adjusted additionally for breast density (HR = 0.62; 95% CI, 0.46–0.84) and menopausal hormone use (HR = 0.61; 95% CI, 0.45–0.83). Accounting for death as a competing risk did not substantively change results (HR = 0.65; 95% CI, 0.48–0.88). Analyses that excluded follow-up prior to date of completion of the questionnaire (which effectively removed 197 participants who were diagnosed with breast cancer prior to completing the questionnaire and 604 participants right-censored prior to completion) yielded similar results (HR 0.57; 95% CI, 0.35–0.93). See Supplementary Table S3 for full results. Associations persisted when breast cancer risk related to a specific agent was adjusted for use of other agents.

Well-tolerated targeted breast cancer prevention strategies could offer great benefits for individual women diagnosed with BBD and reduce the burden of breast cancer in the population. We report that women in the Mayo BBD cohort who regularly used nonaspirin NSAIDs experienced significantly lower breast cancer risk, with suggestively greater effects among subsets of patients.

Regular use of non-aspirin NSAIDs was associated with 37% lower breast cancer risk in our analysis, with greater effects in relation to increased years or frequency of use and greater number of lifetime doses consumed. Sensitivity analyses that excluded women who were diagnosed with breast cancer prior to reporting NSAID use or for whom data were provided by next-of-kin yielded similar results, limiting concerns of bias. Adjusting for mammographic density, use of menopausal hormone therapy or use of multiple different NSAIDs did not alter these conclusions. Risk associations did not vary by estrogen receptor or nodal status of breast cancers.

In a prior report, with limited dosing information, aspirin use among women with BBD showed a significant reduction in breast cancer risk, whereas nonaspirin NSAID use was associated with nonsignificantly lower risk (30). We did not identify significant risk associations for aspirin use; however, 54.1% of aspirin users in our analysis were 55 years of age or older compared with 21.4% of nonaspirin users, and we did not collect data on dosage of the tablets consumed; thus, if risk reduction with aspirin use is greatest among younger women or only realized with high dose formulations, a protective effect may not have been found. Consistent with this view, risk associations were greatest among nonaspirin NSAID users who were less than 50 years of age. Several features of the premenopausal period are associated with inflammation, including menses, which is associated with increased circulating inflammatory cytokines (35), and postpartum involution, in which breast tissue undergoes dramatic remodeling. In preclinical models, postpartum involution drives the development of aggressive breast cancer and this effect can be mitigated with NSAIDs (14). In humans, dysregulated post-partum inflammation may contribute to transiently elevated breast cancer risk after childbirth (36). Effects of nonaspirin NSAID use on breast cancer risk were also greater among obese versus nonobese women. Obesity is associated with low-grade chronic inflammation, which may manifest in adipose rich tissues, such as the breast, and increase breast cancer risk through hormonal and nonhormonal pathways (37). Mechanistic studies show that sera from obese patients induce dramatic increases in macrophage production of COX-2, which can be linked to increased aromatase expression in preadipocytes, and can be inhibited with the NSAID celecoxib (38).

Meta-analyses among average risk women have generally found that aspirin and non-aspirin NSAID use are associated with lower breast cancer risk; however, results are variable (20–25) and magnitudes of effects are generally modest. Misclassification of drug use is an important concern in observational epidemiologic studies evaluating breast risk; however, random misclassification would generally result in under-estimation of true relationships and over-reporting among cases (recall bias) would likely produce a similar effect. Furthermore, NSAID use in large epidemiologic studies cannot be temporally related to inflammation in the breast, which would be anticipated to dilute protective effects. Previously, we reported that immune cells are increased in BBD biopsies versus normal tissues donated for research, and that specific immune responses may predict breast cancer risk among women with BBD (18, 39, 40), suggesting that targeting inflammation may have potential utility for prevention among women with BBD.

Our data suggest that lower breast cancer risk with NSAID use was more evident for non-proliferative BBD, which is the most common category of BBD diagnosis. Nonproliferative BBD frequently accompanies proliferative BBD or atypical hyperplasia, although pathology reports may only report the highest risk lesion. Apocrine cysts are among the most frequent manifestations of nonproliferative BBD, and while cysts are not breast cancer precursors, cyst fluid is rich in hormones and inflammatory mediators, which may create a microenvironment conducive to breast cancer development (27–29, 41). Furthermore, cysts may form and collapse, resulting in inflammation and scarring, even when no longer recognizable microscopically. Women with ≥3 nonproliferative histopathologic BBD lesions in BBD biopsies, a potential marker of diffuse bilateral disease, showed markedly lower breast cancer risk with nonaspirin NSAID use (HR = 0.33; 95% CI = 0.17–0.65; P = 0.001).

Our study leverages strengths of the Mayo BBD cohort, which includes collection of breast cancer risk factors, tissues and follow-up. Our ability to demonstrate a dose–response relationship between non-aspirin NSAID use and lower breast cancer risk among women with BBD is an additional strength. However, this population is comprised almost exclusively of white women, which limits generalizability. Furthermore, medication use was self-reported at a single time point and did not distinguish use of low-dose aspirin from higher doses, which may have limited our ability to identify a significant association for use of this drug. In addition, stratified analyses assessing NSAID associations in certain subsets of women resulted in modest sample sizes and decreased statistical power. We did not collect information about use of statins or bisphosphonates, which are frequently used at older ages, and could affect breast cancer risk. Results of multivariable analyses adjusted for potential confounders, produced only modest effects on risk estimates. Although anti-inflammatory agents are used sporadically for acute illnesses or injuries, our analysis focused on regular use and found strongest effects for higher cumulative doses.

In summary, our analysis of women with BBD demonstrates that regular use of nonaspirin NSAIDs is associated with an approximately 37% reduction in breast cancer risk. Given that BBD is associated with increased breast cancer risk, and frequently includes inflammatory lesions in breast tissues, our findings suggest that NSAIDs may represent an important prevention strategy, and therefore merits further clinical and mechanistic investigation. We propose that clinical prevention trials evaluating the efficacy of NSAIDs for breast cancer prevention among women with BBD that incorporate correlative studies to examine interactions between inflammation and estrogen pathways in breast tissues are warranted. A recently completed phase II trial of celecoxib 400 mg twice daily for 6 months, demonstrated good tolerance with favorable modulation of selected biomarkers (42).

M.H. Frost reports grants from NCI during the conduct of the study. C.M. Vachon reports grants from NIH/NCI during the conduct of the study; grants from Grail "STRIVE mammography study" outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

M.E. Sherman: Conceptualization, resources, formal analysis, validation, methodology, writing-original draft, project administration, writing-review and editing. R.A. Vierkant: Conceptualization, formal analysis, supervision, writing-original draft, project administration, writing-review and editing. S. Kaggal: Formal analysis. T.L. Hoskin: Formal analysis, writing-review and editing. M.H. Frost: Data curation, supervision, project administration, writing-review and editing. L. Denison: Writing-review and editing. D.W. Visscher: Formal analysis, methodology. J.M. Carter: Writing-original draft, writing-review and editing. S.J. Winham: Formal analysis, supervision, writing-original draft, project administration, writing-review and editing. M.R. Jensen: Formal analysis. D.C. Radisky: Formal analysis, funding acquisition, writing-original draft, writing-review and editing. C.M. Vachon: Formal analysis, writing-original draft, writing-review and editing. A.C. Degnim: Conceptualization, resources, supervision, funding acquisition, writing-original draft, project administration, writing-review and editing.

This work was supported by the NCI of the NIH (CA229811 to M.E. Sherman and A.C. Degnim, and CA1187112 to A.C. Degnim).

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