Aspirin Use and Risk of Breast Cancer: Systematic Review and Meta-analysis of Observational Studies

Breast cancer, one of the most frequently diagnosed cancers among women of all racial and ethnic groups, leads to the second most common cancer-associated death among U.S. women (1). Despite many efforts, few modifiable risk factors for breast cancer have been identified. Recently, the potential anticancer properties of aspirin, commonly known as pain reliever, have attracted more interest (2). Aspirin display anticancer activity by inhibiting cyclooxygenase (COX; ref. 3), thus decreasing the formation of downstream tissue-specific signaling lipids known as prostanoids. Prostanoids play an important role in carcinogenesis by affecting cellular proliferation, apoptosis, and angiogenesis (4). Although protection of aspirin against cancer has been showed, the protective effects are seen mainly in colorectal, esophageal, gastric, and endometrial cancers (3, 5). Previously, we have investigated the effect of aspirin intake on the mortality in breast cancer, but only found that aspirin use has a small effect on the survival of breast cancer patients (6).

A number of observational studies have investigated the effect of aspirin use on the risk of breast cancer in the past three decades, but their results are conflicting rather than conclusive. A meta-analysis (7) including 33 studies has evaluated the association and found that a decreased risk of breast cancer for aspirin users was found in the pooled analysis of all studies [odds ratio (OR) = 0.86; 95% confidence interval (CI) = 0.81–0.92] but not in a randomized controlled trial (OR, 0.98; 95% CI, 0.87–1.09). The investigators concluded that regular use of aspirin may be associated with reduced risk of breast cancer. However, the subjects of two included articles (8, 9) were overlapped in another two included articles (10, 11). The duplicated data could bias the results. We also noted that two studies (12, 13) examining the association between aspirin use and mortality in breast cancer patients but not between aspirin use and breast cancer risk were also included in that meta-analysis. In addition, previous meta-analysis has not performed a subgroup analysis according to hormone receptor status, menopausal status, or cancer stage. Since the meta-analysis, seven large observational studies have published. Therefore, we performed a meta-analysis with all available studies to explore the association between aspirin use and risk of breast cancer. Besides, we also performed a dose–response analysis to further evaluate the potential dose–response relations.

We searched PubMed (from 1980 to present) and Embase (from 1977 to present) using the following terms: “aspirin” or “acetylsalicylic acid” or “non-steroidal anti-inflammatory drug” or “non-steroidal anti-inflammatory drugs,” and “breast cancer” and “risk” or “incidence”. The latest date of this search was April 2015. All cohort or case–control studies evaluating the relationship between aspirin use and risk of breast cancer were eligible, without language restriction. Reference lists from relevant articles were examined manually to further identify potentially relevant studies. All searches were conducted independently by two reviewers; differences were checked by the two and resolved by discussion. When two or more studies presented possible overlap, we included the one with largest populations.

All the studies were included in this meta-analysis if they met the following criteria: (i) the exposure of interest was aspirin use; (ii) the outcome of interest was breast cancer; (iii) the study design was case–control or cohort; and (iv) risk estimates and 95% CIs were reported (or information to calculate them).

Two independent investigators extracted data from the eligible articles. The extracted data included the last name of first author, year of publication, origin of the study, follow-up period, study design, sample size, aspirin use, risk estimates, and corresponding 95% CIs, and covariates adjusted for in the multivariable analysis. For studies provided more than one risk estimate, we extracted the one that was adjusted for the greatest number of confounding factors. Discrepancies were resolved by consensus, involving a third investigator.

Two investigators assessed the methodological quality of included studies independently using the nine-star Newcastle Ottawa scale (NOS; ref. 14). Each study was evaluated based on eight items, categorized into three broad perspectives, including selection, comparability, and outcome, for cohort studies or exposure for case–control studies. We considered studies with a score of 7 or greater as high quality. Discrepancies were resolved by discussion or through consultation with a third investigator.

Because outcomes were relatively rare, the ORs and HRs were considered approximations of relative risks (RR). Summary estimates of RRs and 95% CIs were obtained using a random-effects model where the restricted maximum likelihood estimator was used to evaluate the interstudy heterogeneity (15, 16). We calculated prediction interval (PI) of summary estimate for the random effects model to depict the uncertainty around the estimate (17). If studies did not report a summary risk estimate for aspirin use, a summary risk estimate was calculated using risk estimates for each of the aspirin use categories (18–33). Interstudy heterogeneity was estimated using a χ²-based Q test (34), with a P value of <0.10 considered statistically significant (35). We also calculated the I² quantity (34), which lies between 0% and 100%. A value of 0% indicates no observed heterogeneity and larger values indicate increasing heterogeneity. Metaregression was conducted to further explore the sources of heterogeneity. Sensitivity analyses were performed to reflect the influence of individual data on summary HRs. Finally, the potential for publication bias was examined using Begg and Egger regression test (36). Where publication bias was found, the trim-and-fill method was used to estimate the potential influence of this bias on pooled summary estimates (37).

A two-stage random-effects dose–response meta-analysis was performed to compute the trend from the correlated log RR estimates across levels of aspirin use into account the between-study heterogeneity (38). A potential nonlinear relation between aspirin use and risk of death was investigated using restricted cubic splines with three knots at the 25th, 50th, and 75th percentiles of the exposure distribution. Then, the study-specific estimates were combined using the restricted maximum likelihood method in a multivariate random-effects meta-analysis (39). A P value for nonlinearity was calculated by testing the null hypothesis that the coefficient of the second spline is equal to zero. For each study, we calculated the median level of aspirin use for each category by assigning the midpoint of upper and lower boundaries in each category as the average aspirin use level. When the highest category was open-ended, we assigned the lower end value of the category multiplied by 1.5. Studies were not eligible if the required data were not reported or could not be estimated. All of the statistical analyses were done with R software, version 3.1.1, using the packages metafor (40) and dosresmeta (41). All statistical tests were two-sided.

Figure 1 presents the process of study selection. Two thousand six hundred and ten abstracts and titles were identified and assessed, and 45 studies were evaluated in detail with regard to their fulfillment of the inclusion criteria. Four articles were excluded as the exposure of interest was not aspirin use, or no usable data were reported (42–45). Nine studies were excluded because their data duplicated or overlapped with other articles (8, 9, 46–52). Nevertheless, one of these studies (46) was included in the subgroup analysis by hormone receptor status, and another study (48) was included in dose–response analysis. Finally, 17 cohort studies (10, 18, 20–25, 29–31, 53–58) and 15 case–control studies (11, 19, 26–28, 32, 33, 59–66) involving 1350,913 participants were selected for meta-analysis. Table 1 shows the characteristics of the included studies. Supplementary Tables S1 and S2 summarize the methodological quality of all the included studies. The NOS results showed that the average score was 6.9 (range 4–9) for cohort studies and 6.3 (range 4–9) for case–control studies. There were 18 studies with a score of 7 or more.

Figure 2 presents the pooled RR of the association between aspirin use and risk of breast cancer. When the association of aspirin use on breast cancer risk was analyzed as users versus nonusers, a RR of 0.90 (95% CI, 0.85–0.95, P < 0.01; 95% PI, 0.68–1.19) was found. In stratified by study design, a significant benefit for aspirin users was only found in population-based case–control (PCC) studies (RR, 0.80; 95% CI, 0.73–0.88, P < 0.01; 95% PI, 0.65–0.98) and hospital-based case–control (HCC) studies (RR, 0.82; 95% CI, 0.75–0.91, P < 0.01; 95% PI, 0.75–0.91; Fig. 2). In the subgroup analysis by estrogen receptor (ER) or/and progesterone receptor (PR) status, a decreased breast cancer risk was found for aspirin users with ER-positive tumors (RR, 0.90; 95% CI, 0.81–0.99, P = 0.04; 95% PI, 0.68–1.19), PR-positive tumors (RR, 0.87; 95% CI, 0.77–0.98, P = 0.02; 95% PI, 0.67–1.13), and ER- and PR-positive tumors (RR, 0.88; 95% CI, 0.79–0.99, P = 0.04; 95% PI, 0.68–1.14; Fig. 3). When stratifying by menopausal status, the results revealed that postmenopausal women who used aspirin had a RR of 0.86 (95% CI, 0.80–0.93, P < 0.01; 95% PI, 0.75–1.00) for breast cancer risk (Fig. 3). Further analysis by cancer stage showed that aspirin use was only associated with decreased risk of in situ breast cancer (RR, 0.79; 95% CI, 0.71–0.88, P < 0.01; 95% PI, 0.71–0.88; Fig. 3).

The dose–response effects of aspirin on breast cancer risk were assessed with 14 studies (10, 11, 18, 20, 22, 24, 27–29, 48, 54, 55, 60, 63). Among these studies, exposure to aspirin was expressed in duration of exposure (11, 18, 20, 24, 27–29, 55, 63) or frequency (e.g., pills/week; refs. 10, 22, 27, 48, 54, 55, 60, 63). If exposure to aspirin was expressed in a times/week (or days/week) scale (10, 22, 55), we assumed the scale equals to pills/week. A linear relationship was found for both duration (P for nonlinearity = 0.29) and frequency (P for nonlinearity = 0.88) of aspirin use (Fig. 4). A 1-year increment in duration of exposure to aspirin conferred a RR of 0.98 (95% CI, 0.97–1.00, P = 0.02) for breast cancer risk (Fig. 4A). The risk of breast cancer was decreased by 4% for every 3 pills/week increment in aspirin use (RR, 0.96; 95% CI, 0.92–0.99, P = 0.02; Fig. 4B).

From the results of the leave-one-out sensitivity analysis, the summary RR was not materially altered (data not shown). We explored the source of heterogeneity by country (United States and others), study design (cohort, NCC, PCC, and HCC), publication year, methodological quality (continuous variables), and sample size (≥8,000 and <8,000 subjects) with metaregression. The results revealed that none of them contributed to the source of heterogeneity (data not shown).

The results of Begg (P < 0.05) and Egger tests (P = 0.01) have shown the evidence of publication bias. Then, trim-and-fill method was used to correct the result, and nine potential missing studies were required in the right side of the funnel plot in order to make the plot symmetric (Supplementary Fig. S1). However, the correction achieved a nonsignificant reduction of breast cancer risk for aspirin users (RR, 0.95; 95% CI, 0.88–1.02, P = 0.12; 95% PI, 0.65–1.38).

The current meta-analysis investigated the relationship between aspirin use and risk of breast cancer involving 1350,913 participants. The summary results, as derived from 17 cohort studies and 15 case–control studies, indicated that the average effect of aspirin use was associated with decreased risk of breast cancer. However, there was evidence of publication bias and the association between aspirin use and breast cancer risk no longer existed after correction using the trim-and-fill method. We also noted significant heterogeneity between included studies. We explored the source of heterogeneity using meta-regression and subgroup analysis but failed to find convincing explanations for the significant heterogeneity. When stratified by study design, a significant benefit for aspirin users was only found in PCC and HCC studies but not in cohort and nest case–control (NCC) studies. PCC and HCC studies give a lower level of evidence than cohort and NCC studies and might provide spurious results because of selection bias and recall bias, which might have contributed to the different results. Considering above mentioned, we conclude that aspirin use may not be associated with decreased risk of breast cancer overall.

In subgroup analysis by ER or/and PR status, a benefit for aspirin use on breast cancer risk was only found in hormone receptor–positive tumors. Evidence suggested that endogenous estrogens are important risk factors for breast cancer (67). Postmenopausal women who are regular users of aspirin showed lower estrogen levels compared with nonusers, which could potentially lower the risk of breast cancer (68). As our result indicated, the effect for ever use of aspirin was beneficial among postmenopausal women but not premenopausal women. The different effect of aspirin may be due to postmenopausal women tend to have more hormone receptor–positive tumors (69). Further analysis by cancer stage showed that aspirin use was only associated with decreased risk of in situ breast cancer. Although COX2 overexpression has been observed in both invasive and in situ breast tumors, a higher frequency of COX2 expression was observed in in situ tumors suggesting that the potential therapeutic impact of COX2 inhibition may be more relevant for in situ breast cancer than invasive tumors (70).

We further explored the dose–response relationship between aspirin use and risk of breast cancer. Our results indicated a linear and borderline significant relationship between aspirin use and breast cancer risk. The risk of breast cancer was decreased by 2% for 1-year increment of aspirin use. There was a 4% breast cancer risk reduction for every 3 pills/week increment in aspirin use. Considering the small effect, long-term or high-frequency aspirin use is not recommended for prevention of breast cancer.

There is significant heterogeneity between included studies. However, we failed to find the source of heterogeneity. Several reasons may account for the heterogeneity. First, seven cohort studies reassessed exposure to aspirin during the course of follow-up, while ten cohort studies assessed the exposure only at baseline (Supplementary Table S1), which can lead to immortal time bias (71). Second, misclassification of aspirin use is likely to impact on the effect estimates of aspirin use. Most studies assessed the exposure to aspirin based on self-reported, while some studies obtained exposure data from prescription database or medical record. Self-reported data are not expected to be accurately recalled. Third, studies conducted among populations are varying in age and ethnicity. All of these may partly explain the significant heterogeneity between studies.

The potential limitations of our study should be considered when interpreting the results. First, there was evidence of publication bias and heterogeneity among included studies. Second, although included studies had adjusted for important risk factors, unmeasured factors related to aspirin use may also have influenced results of individual studies. Third, in the subgroup analyses, the limited study number with relatively small sample size in each subgroup may not have enough statistical power to explore the real association. Fourth, the included studies were major conducted in Western countries, and hence the results should be extrapolated to other populations with caution.

In conclusion, aspirin use may not affect overall risk of breast cancer, but decrease risk of in situ breast tumors or hormone receptor–positive tumors and reduce breast cancer risk in postmenopausal women. Considering the heterogeneity and publication bias, our results on the effect of aspirin use on risk of breast cancer overall or in the subgroups should be confirmed in future studies with well-controlled confounding factors, long enough follow-up time, and more accurate assessment of aspirin use, including quantity, frequency, and duration of use.

No potential conflicts of interest were disclosed.

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.