Aspirin might reduce the incidence of breast cancer: An updated meta-analysis of 38 observational studies.

BACKGROUND: Many epidemiologic studies were performed to clarify the protective effect of regular aspirin use on breast cancer risks, but the results remain inconsistent. Here, we conducted an updated meta-analysis of 38 studies to quantitatively assess the association of regular aspirin use with risk of breast cancer.
METHOD: We performed a bibliographic database search in PubMed, Embase, Web of Science, Cochrane library, Scopus, and Google Scholar from January 1939 to December 2019. Relative risk (RR) estimates were extracted from eligible case-control and cohort studies and pooled using a random effects model. Subgroup analysis was conducted based on study design, aspirin exposure assessment, hormone receptor status, menopausal status, cancer stage as well as aspirin use duration or frequency. Furthermore, sensitivity and publication bias analyses were performed.
RESULTS: Thirty eight studies of 1,926,742 participants involving 97,099 breast cancer cases contributed to this meta-analysis. Compared with nonusers, the aspirin users had a reduced risk of breast cancer (RR = 0.91, 95% confidence interval [CI]: 0.87-0.95, P value of significance [Psig] < .001) with heterogeneity (P value of heterogeneity [Phet] < .001, I = 82.6%). Subgroup analysis revealed a reduced risk in case-control studies (RR = 0.83, 95% CI: 0.78-0.89, Psig < .001), in hormone receptor positive tumors (RR = 0.91, 95% CI: 0.88-0.94, Psig < .001), in situ breast tumors (RR = 0.79, 95% CI: 0.71-0.88, Psig < .001), and in postmenopausal women (RR = 0.89, 95% CI: 0.83-0.96, Psig = .002). Furthermore, participants who use aspirin for >4 times/wk (RR = 0.88, 95% CI: 0.82-0.96, Psig = .003) or for >10 years (RR = 0.94, 95% CI: 0.89-0.99, Psig = .025) appeared to benefit more from the reduction in breast cancer caused by aspirin.
CONCLUSIONS: Our study suggested that aspirin use might be associated with a reduced risk of breast cancer, particularly for reducing the risk of hormone receptor positive tumors or in situ breast tumors, and the risk of breast cancer in postmenopausal women.

Introduction

Breast cancer is the most frequent cancer and the leading cause of cancer death among women.[ Therefore, effective breast cancer prevention strategies are urgently needed. Aspirin is commonly used to treat pain, fever, and inflammation. Based on its long-term safety and preliminary efficacy data, aspirin has been investigated extensively as a potential cancer chemopreventive agent, reducing incidence in oesophageal, colorectal, gastric, pancreatic, and prostate cancer.[ Aspirin may inhibit tumor growth by modulating cellular proliferation and apoptosis, predominantly via suppression of endogenous production of prostaglandin from inhibition of cyclooxygenase (COX) enzyme activity, particularly COX-2. Convincing laboratory evidence has emerged to demonstrate that COX-2 was overexpressed in breast cancer, but not in normal breast issue, which made aspirin a potential chemopreventive agent of breast cancer.

Numerous epidemiologic studies have investigated the relationships between use of aspirin and risk of breast cancer. Some studies have suggested a modest reduction in breast cancer risk in relation to use of aspirin,[ but a randomized clinical trial of long term use of aspirin and some of recent studies did not support a protective role.[

A meta-analysis[ of 13 cohort studies recently published had found a borderline significant inverse association between aspirin use and breast cancer risk. The researchers had observed potential associations between aspirin intake frequency and breast cancer risk, and the duration of aspirin intake and breast cancer risk. However, results concluded only from cohort studies were less robust, as case–control studies tended to obtain more detailed data for aspirin use, which included the definition and updated assessment of aspirin exposure, appropriate adjustment in baseline characteristics. With the exception of that, estrogen receptor, progesterone receptor, and menopausal status as well as different stages of cancer were not evaluated in the study. Furthermore, 6 cohort studies[ were leaved out in the meta-analysis, and the subjects of one included study[ was overlapped with another one.[ Both the missing and duplicated data could bias the results. Besides, 6 relevant studies were published recently.[ So far, the effect of aspirin on the occurrence of breast cancer has remained uncertain.

The purpose of this meta-analysis was to investigate the association between aspirin use and breast cancer risk of all eligible studies by grouping type of study design, aspirin exposure assessment, hormone receptor status, menopausal status, cancer stage, frequency or duration of aspirin use.

Methods

Search strategy

We searched PubMed, Embase, Web of Science, Cochrane library, Scopus, and Google Scholar without language restriction from January 11, 1939 to October 19, 2019 with the following search terms: (“neoplasms” [Mesh] OR neoplas∗ OR “tumor” OR “tumour” OR “cancer” OR tumorigen∗ OR sarcoma∗ OR malignan∗ OR adenocarcinoma∗ OR “tumors”[tw] OR “tumours”[tw] OR “cancers”[tw]) AND (“aspirin”[Mesh] OR “anti-inflammatory agents, Non-Steroidal”[Mesh] OR “anti-inflammatory drugs, Non-Steroidal”[tw] OR “NSAIDs”[tw] OR “acetylsalicylic acid”[tw] OR salicyl∗ OR “Cyclooxygenase 2 Inhibitors”[tw] OR “cyclooxygenase inhibitors”[tw] OR “COX inhibitors”[tw]) AND (“breast”[Mesh] OR mamma∗). The reference lists of previous systematic reviews on the same topic were reviewed to obtain additional eligible publications. We attempted to contact the authors if we required additional information. Two reviewers identified the publications independently and discussed to resolve the differences.

Study identification

Studies were included if they met the following criteria: evaluate the association between aspirin use and risk of breast cancer; use a randomized controlled trial (RCT) or case–control or cohort study design; provide the odds ratio (OR) or relative risk (RR) with confidence interval (CI) or data necessary to calculate them (raw data, P value, or variance estimate). When authors reported the same population in more than one publication, only the most recent report, or the most complete one, was included.[ Whereas, studies controlling for aspirin use in statistical models without numerically reporting effect measures were excluded.[ Reviewers resolved all the discrepancy by discussion during the study identification process.

Data extraction

Each eligible study was carefully reviewed by 2 independent investigators. The extracted data included the last name of first author, year of publication, location, study design, study time, setting, sample size, cancer cases, the number of people using aspirin, and nonuse, any matching factors and definition for aspirin user. For studies providing >1 risk estimate, we extracted the one that was adjusted for the greatest number of confounding factors.[ We resolved discrepancies through discussion or the third investigator.

Statistical analysis

The analyses were conducted in 4 parts. First, we used meta-analysis to pool the estimates of RRs and 95% CIs. As low morbidity of breast cancer, we treated ORs as proxy measures of RRs.[ Estimates of summary risks that were not represented in original articles were calculated based on each of aspirin use categories. Statistical inconsistency among included studies was tested by Cochran Q test at the P < .05 level of significance. We also calculated the quantity I2 that describes the percentage variation across studies that is attributed to heterogeneity. When significant heterogeneity was found, the random-effects model was used for meta-analysis. Otherwise, the fixed-effects model was adopted.

Second, subgroup and sensitivity analyses were performed. We evaluated for potential source of heterogeneity stratified by the following rules: study design: cohort studies versus case–control studies; aspirin exposure ascertain: questionnaire versus interview versus automated databases; hormone receptor status: positive versus negative; menopausal status: premenopausal versus postmenopausal; cancer stage: in situ cancer versus invasive cancer; aspirin use frequency: ≤4 times/wk versus >4 times/wk; aspirin use duration: ≤10 years versus >10 years. To reflect the influence of individual study on summary RRs, a sensitivity analysis was performed.

Third, the potential for publication bias was investigated using Begg and Egger regression test. Egger test was performed to provide quantitative evidence, and P < .05 indicated the existed publication bias. Where publication bias was found, the trim-and-fill method was used to estimate the potential influence of this bias on pooled summary estimates. Stata 12.0 (Stata Corp, LLP, College Station, TX) was used for all analyses.

Results

Eligible studies

The overview of our search process was illustrated in Fig. 1. We identified 4186 articles through electronic databases research. After titles and abstracts review, there were 52 potentially eligible studies. We found an additional 11 papers after reviewing articles, reference lists, and other sources. Of the 63 publications, we reviewed in full-text and excluded 25 records according to the above inclusion criteria. The studies were excluded for no useable data reported[ or as the exposure of interest was not aspirin use.[ Eleven studies were excluded for overlapping with others.[ As only a few studies reported subgroup results, we used data in duplicate studies if studies with the largest number of cancer cases didn’t report those to allow studies to be as inclusive as possible.[ At last, 38 studies met the predetermined criteria for inclusion, with 22 cohort studies[ and 16 case–control studies.[ Among the 1,926,742 participants of this meta-analysis, there were 97,099 incident breast cancers. The 38 studies were published between January 1939 and April 2019. The range of enrollment periods for participants across studies was 1971 to 2017. Breast cancer was screened along with pathology reports and medical records, or through linkages with cancer registries in 36 studies (2 studies had no report of this).[ Aspirin use was measured from questionnaire data in 18 studies, interviews in 10 studies, and automated database in 10 studies. Details of these studies are shown in Table 1    .

Figure 1

The flow diagram of search strategy. Flow chart of study identification and selection.

Table 1

Characteristics of studies included in the meta-analysis.

Overall meta-analysis

Figure 2 illustrated the forest plot of RRs estimates with 95% CIs from individual studies and overall meta-analysis of all 38 studies. The overall summary RRs demonstrated aspirin reduced the incidence of breast cancer (RR = 0.91, 95%CI: 0.87–0.95, Psig < .001), but there was heterogeneity between studies (Phet < .001, I2 = 83.0%). We explored the source of heterogeneity by subgroup and sensitivity analyses.

Figure 2

Forrest plot of the association between aspirin use and risk of breast cancer. Black squares and horizontal lines represent the study relative risk and 95% CIs. The size of the square is proportional to the weights of the individual studies. The diamond represents the pooled estimate (center) and 95% CI (width). CI = confidence interval; I squared = estimate for the proportion of variability between studies that is due to inter-study heterogeneity; RR = relative risk.

Subgroup analyses

Table 2 summarized the results of subgroup and publication bias analyses by various factors. The pooled results were 0.96 (95% CI: 0.91–1.02, Psig = .164) for cohort studies and 0.83 (95% CI: 0.78–0.89, Psig < .001) for case–control studies, suggesting a protective effect in case–control studies. We observed substantial statistical heterogeneity in both groups (cohort: Phet < .001, I2 = 85.9%; case–control: Phet = .003, I2 = 56.7%).

Table 1 (Continued)

Characteristics of studies included in the meta-analysis.

Regarding the exposure assessment, 18 studies used mailed questionnaires or in-hospital questionnaires, 10 studies used interview typically performed by trained personnel, and the rest of studies used either automated databases or medical records. Breast cancer was less likely to occur after aspirin use if exposure definition was interview (RR = 0.81, 95% CI: 0.73–0.89, Psig < .001), however, this association was weaker for questionnaires (RR = 0.93, 95% CI: 0.87–1.00, Psig = .046). We found that aspirin was not significantly associated with the breast cancer if recorded by automated databases (RR = 0.96, 95% CI: 0.90–1.02, Psig = .154). Significant heterogeneity was observed when exposure definition were questionnaires (Phet < .001, I2 = 87.6%) and automated databases (Phet < .001, I2 = 78.4%), but not in interview (Phet = .078, I2 = 41.9%).

Twelve studies evaluated the relationships between exposure to aspirin and breast cancer risk in subgroup analysis based on hormone receptor status. All of them reported the relationships in hormone receptor positive tumors, and 10 studies reported the relationships in hormone receptor negative tumors. Aspirin was associated with a decreased breast cancer risk in hormone receptor positive tumors (RR = 0.91, 95% CI: 0.88–0.94, Psig < .001), with statistical heterogeneity (Phet < .001, I2 = 58.5%). While in hormone receptor negative tumors, no associations were observed (RR = 1.06, 95% CI: 0.98–1.15, Psig = .130), with no statistical heterogeneity (Phet = .060, I2 = 46.4%).

When stratified by menopausal status, there were 7 studies concerned premenopausal women and 15 studies concerned postmenopausal. A significant risk reduction of breast cancer caused by aspirin was observed in postmenopausal women (RR = 0.89, 95% CI: 0.83–0.96, Psig = .002), but not in premenopausal women (RR = 0.88, 95% CI: 0.72–1.08, Psig = .223). Both groups had significant heterogeneity (postmenopausal: Phet < .001, I2 = 72.1%; premenopausal: Phet = .007, I2 = 66.1%).

When stratified by cancer stage, 13 studies were found for reporting invasive breast cancer, and 3 of them reported in situ breast cancer at the same time. A reduction in risk was observed for in situ breast cancer (RR = 0.79, 95% CI: 0.71–0.88, Psig < .001), with no substantial statistical heterogeneity (Phet = .410, I2 = 0.0%). For invasive breast cancer, no significant association was found (RR = 1.00, 95% CI: 0.94–1.06, Psig = .988) and a substantial statistical heterogeneity was observed (Phet = .002, I2 = 61.7%).

We further explored the relationships between aspirin use dosage and breast cancer risk. The risk of breast cancer was significantly reduced in participants who use aspirin for >4 times/wk (RR = 0.88, 95% CI: 0.82–0.96, Psig = .003) or <4 times/wk (RR = 0.95, 95% CI: 0.91–0.99, Psig = .029). Besides, a significant risk reduction of breast cancer caused by aspirin was observed in participants who use aspirin for >10 years (RR = 0.94, 95% CI: 0.89–0.99, Psig = .025). And borderline significant inverse associations were observed between breast cancer risk and aspirin use for shorter than 10 years (RR = 0.97, 95% CI: 0.94–1.00, Psig = .045). But, significant heterogeneity was observed in all these subgroups (all Phet < .05).

Sensitivity analysis and publication bias

From the results of the leave-one-out sensitivity analysis, the summary RR was not materially altered (data not shown).

The publication bias was not observed (P of Egger test = .999) in overall results.

Relevant summary of RRs for subgroup were provided in Supplementary Table 1–5.

Discussion

Based on the overall meta-analysis, we observed a 9% relative decrease in the risk of breast cancer for aspirin users. In this study, aspirin was associated with decreased breast cancer risk in hormone receptor positive tumors. We observed a decreased risk related to in situ breast cancer after aspirin use. We also found a reduction in risk of breast cancer for postmenopausal women. When the analysis was stratified by study design, a reduced risk in case–control studies was observed.

Our result of overall aspirin use and breast cancer risk was different from the article reported previously,[ which reported that a borderline significant inverse association (RR = 0.94, 95% CI: 0.87–1.01) was observed between overall aspirin use and breast cancer risk. The difference might be explained that previous study just summarized the results of cohort studies. And we observed the similar result (RR = 0.96, 95% CI: 0.91–1.02) for cohort studies. Our result found that aspirin was associated with a decreased breast cancer risk. This could be related to aspirin not only preventing breast tumor cell growth through the induction of apoptosis, but also significantly reducing the self-renewal capacity and growth of breast tumor-initiating cells/breast cancer stem cells and delaying the formation of a palpable tumor.[ In addition, recent study showed aspirin may suppress tumor cell-induced angiogenesis or normalizing tumor blood vessels.[ Similar results of subgroup analysis were reported in the meta-analysis by Zhong et al.[ They concluded that aspirin use might decrease risk of in situ breast tumors or hormone receptor positive tumors and reduce risk of breast cancer in postmenopausal women.

We observed a risk reduction of 17% in breast cancer risk for aspirin users from case–control studies, while cohort studies gave no evidence. As a result of recall bias, selection bias, and healthy-user bias introducing differential measurement error, case–control studies gave a lower level of evidence than cohort studies. However, the invalid effects reported in some cohort studies could be accounted for less detailed records of aspirin exposure, a lack of updating of exposure between initial recruitment and subsequent diagnosis of cancer. In a word, the estimates were heterogeneous and further well-characterized research is needed before reliable conclusions can be drawn.

When the analysis was examined by aspirin exposure definition, the risk of breast cancer was reduced in the interview and automated database subgroups. In contrast to self-administered questionnaires that relied heavily on the subject's ability to recall, automated database provided detailed information on dates of use and dosage of drugs used. Even so, as the weak compliance from patients may over-represent the results of automated database, we found that the estimates for the subset of studies using interview was more reliable. Considering above mentioned, we concluded that aspirin use might be associated with decreased risk of breast cancer.

Previous studies have found that in breast cancer of varying stages, COX-2 expression was shown to be inversely correlated with ER expression, and stable transfection of ER in breast cancer cells led to the repression of COX-2, in turn leading to a reduction in proliferation and migration of these cells. Huang et al[ had already demonstrated that a novel aspirin derivative, PA-2 induced a potent cytokinetic effect on ER+ breast cancer cells by impacting potent inhibition of tumor growth. Our analysis also observed aspirin to be associated with a reduced risk of hormone receptor positive tumors.

Postmenopausal women who were regular users of aspirin showed lower incidence to suffer from breast cancer. This might be ascribed that postmenopausal women tended to suffer from hormone receptor positive tumors. Aspirin can decrease aromatase activity via suppression of COX expression and prostaglandin synthesis, which may decrease estrogen concentrations and potentially risk of breast cancer.[ Postmenopausal women who were regular users of aspirin showed lower estrogen levels than nonusers.

Beneficial efficiency was found between aspirin use and the risk of in situ breast tumors. Evidence suggested that COX-2 over-expression has been observed in about 40% of cases of invasive breast carcinoma, but at a higher frequency of in situ tumors, suggesting that the potential therapeutic impact of COX-2 inhibition may be more relevant for in situ breast cancer than invasive tumors,[ which might explain our findings.

It was widely accepted that any potential protective effects of aspirin use against cancers were likely to involve a considerable duration. We also observed statistically significant associations between any frequency and longer duration of aspirin use and risk of breast cancer in subgroup analysis. The overall association between aspirin use and cancer risk of lower frequency appeared to be similar to higher frequency but less striking, the same as when stratified by duration. Our results were consistent with the previous study. Lu et al[ confirmed that breast cancer risk decreased as aspirin intake frequency and a trend of decreasing risk for more years of aspirin intake increased. They also concluded that the optimal aspirin dose for preventing breast cancer may be in the scope of <325 mg per day, 2 to 7 times/wk, along with long-term medication (>5 years).

It is noted that there are always concerns about the use of aspirin and the risk of serious bleeding. According to previous data, only a very modestly higher risk of extracranial and major gastrointestinal bleeds happened on aspirin users when compared with those taken placebo.[ Experts consider that aspirin does increase the risk of bleeding, but the magnitude of this effect is likely to be over-estimated. The secondary cardiovascular disease events from aspirin are thought to be outweighed by the benefits gained from it.

In our study, we noted significant heterogeneity between included studies. We explored the source of heterogeneity using subgroup and sensitivity analysis. It was visible that the heterogeneity had declined in nearly all subgroups we stratified except for subgroup of exposure ascertain. Sensitivity analyses conducted by excluding one study at a time indicated that each individual dataset had no significant influence on the overall results. In summary, the heterogeneity between articles might come from different study designs, hormone receptor status, menopausal status, stages of cancer, aspirin use frequencies, and durations.

Our pooled study had some strengths as well as limitations. First, after searching electronic database, we carefully read the reference list of previous systematic review, making our search broad and comprehensive. This allowed for a large sample size and suitable statistical power to evaluate overall main effect associations, although the case numbers were still small and limited our power to fully perform subset of our studies with extensive questions. Second, subgroup and sensitivity analyses based on biologically important variables gave us ability to further explore the source of heterogeneity and discuss the underlying pharmacological mechanism. However, we failed to find convincing explanations for the significant heterogeneity. Third, we used Egger test to provide quantitative evidence for publication bias. Recall bias, selection bias, and healthy-user bias could not avert although recall bias about aspirin use could be reduced by use of prescription databases. However, the misclassification of aspirin use might have a crucial impact on the effect estimates of aspirin use. Besides, the subgroup analyses based on other factors, such as obesity status, hormone replacement treatment were restricted by few studies reported interested topics.

Conclusions

In summary, findings from this pooled analysis support the hypothesis that aspirin use provides potential benefits in preventing breast cancer, particularly in hormone receptor positive tumors or in situ cancers and breast cancer in postmenopausal women. Future observational studies should confirm well-controlled confounding factors, and consistent assessment of aspirin use, including exposure dose, frequency and duration of use. Random controlled trials are also urgently needed.

Author contributions

Aihua Tan conceived this study. Aihua Tan and Yueqing Cao were responsible for the literature retrieval and data extraction. Yueqing Cao performed the statistical analysis. Aihua Tan oversaw the statistical analysis. Yueqing Cao wrote the paper. Aihua Tan and Yueqing Cao contributed to data interpretation and critical revision. Aihua Tan and Yueqing Cao reviewed the final version of the manuscript.

Table 1 (Continued)

Characteristics of studies included in the meta-analysis.

Table 1 (Continued)

Characteristics of studies included in the meta-analysis.

Table 1 (Continued)

Characteristics of studies included in the meta-analysis.

Table 2

Subgroup analyses and summary for publication bias.