Lifetime use of nonsteroidal anti-inflammatory drugs and breast cancer risk: results from a prospective study of women with a sister with breast cancer

Nonsteroidal anti-inflammatory drugs (NSAIDs) are a class of drugs commonly used to treat pain, fever and inflammation [1]. Based on their long-term safety and preliminary efficacy data, there has been interest in aspirin and other NSAIDs as possible candidates for breast cancer prevention [2]. NSAIDs exert pharmacological effects primarily by blocking cyclooxygenase (COX) enzyme activity to suppress the prostaglandin synthesis pathway [1, 3]. COX-independent effects of NSAIDs have been described, but the chemopreventive potential of NSAIDs is thought to be closely related to their inhibition of the COX-2 enzyme [4]. Overexpression of COX-2 has been implicated in various processes related to breast carcinogenesis such as inhibition of apoptosis, promotion of angiogenesis, and impairment of immune surveillance [3]. Notably, upregulation of PGE_2, the most abundant COX-2—derived prostaglandin [5], can also induce CYP19 transcription and increase aromatase activity, leading to increased estrogen production in mammary adipose tissue [6].

Numerous epidemiologic studies have investigated the relation between use of NSAIDs and breast cancer. Although previous meta-analyses have suggested a modest reduction in breast cancer risk in relation to use of aspirin and other NSAIDs [7, 8], a randomized clinical trial of low dose aspirin [9] and recent analyses of large cohort studies [10, 11] do not support a protective association. Moreover, results concerning the influence of frequency, duration or types of NSAIDs are largely inconsistent even within studies that found an inverse association between use of NSAIDs and breast cancer [12‐15]. Assessment and quantification of NSAID use are challenging in epidemiologic studies. Except for a few studies based on healthcare databases which have their own limitation in assessing over-the-counter NSAIDs [16, 17], information on use of NSAIDs is collected using questionnaires that vary across studies, possibly accounting for inconsistent results.

To further clarify the role of NSAID use in breast cancer prevention, we examined the association between lifetime use of NSAIDs and breast cancer risk in a large prospective cohort of women with a sister with breast cancer.

Approximately 11 % of premenopausal women and 21 % of postmenopausal women had at least 49 pill-years of NSAID use in their lifetime (Table 1). High NSAID use was associated with older age, non-Hispanic white race and higher BMI. History of benign proliferative breast disease and other medical conditions managed with NSAIDs was more common among the high NSAID user group. Among premenopausal women, the high NSAID users were also more likely to be current smokers.

There were no significant associations between use of any NSAID and breast cancer risk, when analyzing duration and frequency separately (Table 2). However, for pill-years of NSAID use, which is a combined measure of duration and frequency, the highest category of non-aspirin, non-COXib NSAIDs was associated with reduced risk compared to the lowest category (multivariable-adjusted HR = 0.84, 95 % CI: 0.71–0.99). Neither pill-years of aspirin nor pill-years of COXibs were inversely associated with breast cancer risk. Results were similar for low- and regular-dose aspirin use and these were not evaluated separately in subsequent analyses.

Approximately 26 % of all incident breast cancers were in situ tumors, and of the remaining invasive cases, 1203 women were diagnosed at stages I or II, and 216 women were diagnosed at later stages (stage III or IV). Associations with NSAID use did not differ among in situ tumors, localized and advanced breast cancers (Additional file 1: Table S2). Invasive breast cancers were also categorized based on ER, PR and HER2 status (Table 3). In situ tumors were not included in this analysis because they are not routinely evaluated for these molecular markers. The majority (75 %) of invasive cancers were ER+/PR+/HER2-, followed by ER-/PR-/HER2- (11 %) and ER+/PR+/HER2+ (10 %). There were too few ER-/PR-/HER2+ tumors to evaluate (N = 53). There was an indication of inverse associations between long-term regular use of any NSAIDs and ER+/PR+ breast cancer regardless of HER2 status. While most associations were not statistically significant, there was a significant inverse trend for increasing pill years of non-aspirin, non-COXib NSAIDs and ER+/PR+/HER2- breast cancer (p = 0.039). The relationship between use of NSAIDs and breast cancer risk did not vary by degree of family history of breast cancer or BMI (data not shown).

In contrast to results for postmenopausal women, lifetime use of NSAIDs was associated with reduced risk of breast cancer among premenopausal women (Table 4). The multivariable-adjusted HRs for the highest versus the lowest category of pill-years of NSAID use were 0.66 (95 % CI: 0.50–0.87) among premenopausal women versus 0.95 (95 % CI: 0.82–1.09) among postmenopausal women. The heterogeneous associations between use of NSAIDs and breast cancer risk by menopause status were most apparent for aspirin (p for homogeneity of trends = 0.038): The HRs associated with the highest category of pill-years of aspirin use were 0.57 (95 % CI: 0.33–0.98) in premenopausal women and 1.03 (95 % CI: 0.87–1.23) among postmenopausal women. Although results were not statistically significant, the inverse associations with the highest category of pill-years of use of COXibs and other non-aspirin, non-COXib NSAIDs also tended to be more evident among premenopausal women.

Several exploratory analyses were conducted to better understand the unexpected subgroup finding. First, our analysis used age as the time scale for calculating HRs rather than time since enrollment, based on the results from a simulation study of different time scales in Cox regression analysis [23]. Since women were required to be breast cancer free at enrollment, those who enrolled at later ages could have a lower hazard than those of the same age who were enrolled at younger ages [24]. However, additional adjustment for age at study entry had negligible impact on the risk estimates for pre- and post-menopausal women.

Second, median pill-years of NSAID use were lower for premenopausal than for postmenopausal women (18.7 vs 28 years). To make the analyses in the two groups more comparable, we further divided the highest category of NSAID use (≥21 pill-years for COXibs and ≥ 49 pill-years for other types of NSAIDs) into two groups using the median among postmenopausal women (not shown), but the association with breast cancer risk remained unchanged (HR for ≥ 91 versus <0.75 pill-years of NSAID use = 1.01, p = 0.89 in postmenopausal women). We also examined breast cancer risk in relation to pill-years of NSAID use within the past 5 years (Additional file 1: Table S3). Again, the results were similar, suggesting an inverse, albeit not significant, association between use of different NSAIDs and breast cancer risk only among premenopausal women (HR for ≥ 14 versus <0.75 pill-years of NSAID use = 0.84, p = 0.09 in premenopausal women vs. 1.02, p = 0.78 in postmenopausal women). Third, separate analysis, stratified by menopause status at study follow-up, was conducted because 46 % of premenopausal women at baseline (including 30 % of premenopausal women who developed breast cancer) became postmenopausal by the end of study follow-up. This analysis did not affect the lack of the association among postmenopausal women: the HRs for the highest pill-years of aspirin were 0.55 (95 % CI: 0.24–1.23) vs 0.99 (95 % CI:0.83–1.17) for those who were premenopausal versus postmenopausal at follow-up (p for homogeneity of trends =0.067).

Lastly, we performed post-hoc analysis to explore the influence of other reproductive factors. Results stratified by parity suggested stronger inverse associations between high pill-years of non-aspirin, non-COXib NSAID use and breast cancer risk among nulliparous premenopausal women compared with among parous premenopausal women (p for interaction = 0.081; Additional file 1: Table S4). The inverse association between high pill-years of NSAID use and breast cancer risk was also more evident among premenopausal women who had not given birth by the age of 30, except for use of COXibs, but there were only a few women with incident breast cancer in this exploratory subgroup analysis. In postmenopausal women, we did not find evidence that these reproductive events modified the relationship between use of NSAIDs and breast cancer risk.

In this analysis of a prospective study of women with a sister with breast cancer, lifetime use of NSAIDs generally did not reduce the risk of developing breast cancer. However, among premenopausal women, our results suggest that lifetime use of different NSAIDs, particularly aspirin, may be associated with reduced breast cancer risk.

A wealth of experimental data supports the chemopreventive potential of NSAIDs. Although the mechanistic pathways are not well established, the primary mechanism appears to involve suppression of COX-2-PGE₂ signaling, which can modulate cellular proliferation, survival and migration, and impair tumor immunosurveillance through multiple signaling pathways [3]. COX-2-PGE₂ signaling has been directly implicated in breast cancer. Targeted overexpression of the COX-2 gene sufficiently induced mammary carcinogenesis and COX-2 inhibition reduced the incidence and extent of breast tumors in breast cancer animal models. [25‐27]. PGE₂ can also upregulate aromatase expression in stromal fat cells in the breast and consequently increase local estrogen production, suggesting that NSAIDs may be especially protective for ER+ breast cancer [28, 29]. However, there is also evidence from animal studies for the role of COX-2 in the development of ER- and HER2+ breast cancer through activation of Akt pathway [30].

There is yet insufficient epidemiological evidence to draw conclusions about the role of NSAIDs in breast cancer risk [2]. Overall, results of recent meta-analyses have concluded that aspirin and other NSAIDs might provide a slight reduction in breast cancer risk [7, 8, 31], but the inverse association tends to be more evident in case–control studies than in cohort studies [10, 15, 17, 32‐34] or in the single randomized controlled trial of low-dose aspirin [9]. Although our study found some evidence that breast cancer risk was reduced among women with the highest lifetime use of non-aspirin, non-COXib NSAIDs, this finding was only significant when we categorized by pill-years of use and appears in association with tumors that are ER+/PR+/HER2-. In general, our findings are consistent with the previous prospective studies that did not support overall benefits of NSAID use for breast cancer prevention.

Some studies have suggested subgroup effects according to tumor and population characteristics. A stronger inverse association with NSAID use has been found for ER+ breast cancer in some [14, 35] but not all studies [12, 22, 36‐38]. The limited available data, including ours, do not support different associations by HER2 status [38, 39]. An extended follow-up analysis of randomized clinical trials has suggested that daily aspirin use may reduce cancer mortality by reducing the incidence of metastatic cancer [40]. We are not yet able to study breast cancer mortality in our cohort. However, associations with use of NSAIDs were similar for in situ, localized and advanced breast cancer in our study.

Menopause status is considered an important effect modifier of several known breast cancer risk factors [41, 42]. We therefore evaluated the association between lifetime use of NSAIDs and breast cancer risk by menopause status and observed a protective association between high lifetime use of NSAIDs and breast cancer risk only among premenopausal women. Consistent with our findings, Zhang et al. reported that the odds ratios for regular use of NSAIDs were 0.62 (95 % CI: 0.41–0.94) among premenopausal women, but 0.90 (95 % CI: 0.69–1.16) in postmenopausal women in a large case–control study [22]. However, regular use of aspirin was similarly inversely associated with breast cancer regardless of menopause status in the Black Women’s Health Study [43], and no relationship between NSAID use and breast cancer was found among premenopausal women in the Nurses’ Health Study II [11]. Reasons for the inconsistent findings are difficult to pinpoint, but it is notable that our prospective study shares marked similarities to Zhang’s study [22] in terms of the assessment of NSAID use, where histories of medications were obtained through questions about an extensive list of major indications, and then detailed information for each episode of use was recorded. On the other hand, regular use (≥2 or ≥3 times per week) of aspirin and other NSAIDs was ascertained based on the baseline questionnaires, and the duration of use was estimated based on the responses to the same questions in each follow-up questionnaire in the aforementioned prospective cohort studies [11, 43].

We found that the protective association with NSAID use among premenopausal women was primarily restricted to the highest exposure category of 49 pill-years of NSAID use. The pill-year measure combines frequency and duration so the same number of pill-years can be derived from different combinations of frequency and duration of use, analogous to pack-years of smoking, a commonly used exposure measure. To address the possibility that older women may achieve the highest category by having more years of use but lower frequency of use, we further investigated the association between pill-years of NSAID use and breast cancer risk after dividing the highest exposure category into two groups (< 91 or ≥91 pill years). However, the HRs did not vary in the two strata.

Another possible explanation for the observed difference may lie in the interaction between NSAIDs and timing of exposure. The development of human breast tissue progresses over the lifespan, and menopause represents one lifecycle event associated with lobular involution and decrease in glandular or dense parenchymal features. Because hormone responsiveness and cell proliferation are the highest in the undifferentiated lobules [44‐46], chemopreventive agents like NSAIDs may be more effective during the period susceptible to tumorigenesis. In the same context, we also evaluated other reproductive events affecting breast development such as parity and age at first term pregnancy. Our exploratory analysis suggested that premenopausal women who were nulliparous or did not give birth before age 30 years old might have greater benefit from lifetime use of NSAIDs. Further studies are necessary to determine whether women’s reproductive history can be used to inform clinical effectiveness of aspirin or other types of NSAIDs on breast cancer prevention.

Strengths of the present study include its prospective design, the large sample size and detailed information on potential risk factors for breast cancer. We also were able to characterize lifetime use of NSAIDs based on comprehensive questionnaire data about both medical history and medication use. In addition to questions about the general use of pain and inflammation medications, participants could report NSAID use in connection with medical histories, which included questions about all medications used in relation to 25 specific health conditions, and as part of a listing of any other medications that were currently being taken or were used within 1 year of enrollment. Despite various approaches (e.g., query of NSAID use prompted by specific medical conditions and visual cue cards for medications) used to improve the accuracy of recall, however, some nondifferential measurement error is unavoidable and is hard to quantify.

A few other issues concerning the interpretation of our study need to be discussed. First, we defined lifetime NSAID use based on the information collected at baseline and did not include information on changes in NSAID use during the study follow-up. However, given the relatively short time to breast cancer diagnosis in our study, it seems unlikely that change in NSAID use pattern is sufficiently large to influence the association between use of NSAIDs and breast cancer risk. Second, the present study involves multiple exposure and outcome measures, and the resulting larger number of statistical tests may have yielded spurious associations. Although the subgroup analyses were guided by biological plausibility and conducted in order to obtain a more comprehensive picture of the relationship between use of NSAIDs and breast cancer risk, findings from the subgroup analyses should be interpreted with caution. Lastly, our study population is unique in that all women had a sister with breast cancer at the time of enrollment. Because women with a family history of breast cancer on average have a 2–4 fold increased risk of developing breast cancer [47], women with an affected sister are a priority population for chemoprevention strategies. At the same time, the increased breast cancer risk in our cohort compared to the general population may also raise concern about potential lack of generalizability of findings. However, this increased risk is not due solely to rare high penetrance genes, but rather to possibly increased frequency of the same polymorphisms and environmental risk factors that affect most women. In fact, the prevalence of high penetrance breast cancer genes in our cohort was <1 % among those tested for BRCA1/BRCA2 mutations. Women in our study were similar to the general population in terms of prevalence of obesity, physical activity and other healthy lifestyle behaviors [48].

In this analysis of more than 50,000 women at higher risk of breast cancer on the basis of family history, there was little evidence that use of NSAIDs can protect against breast cancer. However, among premenopausal women, use of NSAIDs, particularly aspirin, was associated with reduced risk of breast cancer, raising the intriguing hypothesis of chemopreventive benefit for this subgroup of at-risk women.