No association between low-dose aspirin use and breast cancer outcomes overall: a Swedish population-based study

We used a cohort study design to investigate the association between low-dose aspirin use and risk of breast cancer-specific death as primary outcome. Aspirin use was assessed through records of drug dispensing shortly before and after breast cancer diagnosis as well as during the entire follow-up. The study population was identified through three regional breast cancer quality-of-care registers linked with nationwide health-care registers under the acronym BcBaSe Sweden (the Stockholm-Gotland, Uppsala-Örebro and Northern regions) and included all women diagnosed with a primary invasive breast cancer during the period 1 April 2006–31 December 2012. The starting date was given by the availability of information on drug dispensing of low-dose aspirin from July 2005 (see later) plus a period of 9 months for prediagnostic exposure assessment. Compared to the mandated reporting to the National Swedish Cancer Register [20], the completeness of the breast cancer quality-of-care registries is in excess of 90% [21]. Furthermore, health care in Sweden is tax-funded with specialized care in oncology being accessible to all residents. By means of record linkage to the National Swedish Cancer Register, we excluded women with a record of an earlier breast cancer diagnosis while women with other previous cancer diagnoses were not excluded (Additional file 1: Figure S1). Women with breast cancer stage I–III at diagnosis were included in the main analysis, whereas women with stage IV disease at diagnosis were analyzed separately. Among women with stage I–III breast cancer diagnosed in the Stockholm-Gotland region, we also analyzed time to local recurrence or distant metastases as a secondary outcome. Information available from the regional quality-of-care breast cancer registers included age, region, date and clinical TNM classification of disease at diagnosis, estrogen receptor (ER) status, human epidermal growth factor receptor (HER2) status and neoadjuvant and intended adjuvant treatments (chemotherapy, radiotherapy, endocrine therapy and/or trastuzumab). The concordance between register records of intended adjuvant treatment and actually administered treatment has been shown to be high (90%) [22]. Based on the variables in the regional quality-of-care register, we classified the breast cancers as luminal (ER⁺, HER2^–/HER2⁺), nonluminal HER2 (ER^–, HER2⁺) or ER^–HER2^–.

The Swedish Prescription Register (SPR) has recorded all filled prescriptions at Swedish pharmacies from 1 July 2005 onward [23]. From the SPR, we ascertained any dispensing of low-dose aspirin to assess aspirin use. Our definition of dispensed aspirin was limited to daily doses of 75 or 160 mg (ATC codes B01AC06, B01AC30 and B01AC56). These doses are available only by prescription and represent 90% of all aspirin sold nationally (by prescription or over the counter) [24]. We did not consider exposure during 90 days before or after breast cancer diagnosis since women may have taken more aspirin before diagnosis due to local symptoms or may have been told to avoid aspirin in conjunction with surgery. Aspirin use (yes/no) was assessed during a 6-month period before breast cancer diagnosis (≥ 1 dispensing from 9 to 3 months before breast cancer diagnosis) and during a 6-month period shortly after breast cancer diagnosis (≥ 1 dispensing 3–9 months after breast cancer diagnosis). In addition, to estimate cumulative use, aspirin use was assessed during the entire follow-up post diagnosis (from 3 months after breast cancer diagnosis and onward) as a time-varying exposure. For each dispensing, SPR contains information on the date, number of packages, package size and milligrams (mg) per tablet dispensed. There is also a text variable with the doctor’s prescription (e.g., “1 tablet a day”). We calculated the prescribed daily dose as the number of prescribed tablets per day times the number of mg per tablet. The prescribed daily dose was divided into ≤ 75 mg per day and > 75 mg per day. The cumulative number of days of aspirin use was calculated as the number of tablets dispensed divided by the number of tablets prescribed per day and updated at every dispensing. Women who changed doses during follow-up (n = 258, 6.3% of aspirin users) were excluded from the time-varying analyses. The cumulative number of days of use was grouped as follows: none, < 6 months, 6 months–2 years and > 2 years.

Using the national registration numbers assigned to all Swedish residents [25], the cohort was linked to nationwide registers including the SPR (described earlier), the National Patient Register (NPR) and the longitudinal integration database for health insurance and labor market studies (LISA). In the NPR, the Swedish National Board of Health and Welfare has compiled data on individual hospital discharges. Each record contains medical data including diagnoses at discharge according to the International Classification of Diseases (ICD) and dates of admission and discharge. Since 2001, this register also records visits to nonprimary outpatient care, with an estimated proportion of register dropouts of only 2% [26]. The NPR was used to assess comorbidity at diagnosis as well as locoregional recurrence or distant metastases during follow-up (see later). The LISA database includes information on highest achieved educational level (≤ 9 years, 10–12 years, > 12 years), used as a proxy for socioeconomic status [27]. Comorbidity was assessed based on NPR records during 5 years before diagnosis and was classified into two major groups: diseases for which low-dose aspirin use is recommended (cardiovascular, inflammatory and cerebrovascular disorders); and diseases where aspirin may be counterindicated (peptic ulcer disease, chronic liver failure and asthma) (Additional file 1: Table S1). Since the NPR is confined to records of hospital admissions and/or nonprimary outpatient visits, and not visits to the general practitioner, the assessment of comorbidity is likely to mainly reflect severe disorders requiring specialized care. We also assessed dispensing of other nonsteroidal anti-inflammatory drugs (NSAIDs), statins and metformin (through ATC codes M01A, C10AA and A10BA02) from the SPR as potential confounders covering prescriptions from both primary and specialized care. Use of these drugs were assessed as ever/never use during the same time windows as aspirin (dose and duration were not considered).

Women diagnosed with stage I–III breast cancer were followed from 9 months after breast cancer diagnosis until death or 31 December 2012, whichever occurred first. In the analysis of stage IV patients, and in analyses of risk of first recurrence/metastasis among stage I–III patients in the Stockholm-Gotland region (42% of the cohort), patients were followed from the date of diagnosis. For classification of breast cancer-specific death, the cohort was linked to the Cause of Death register [28]. When breast cancer was recorded as the main underlying cause of death, it was regarded as breast cancer specific. Date of recurrence/metastasis was defined as the date of the first record in the NPR with a diagnosis code of locoregional recurrence or distant metastasis (ICD C77–C79). As breast cancer patients are routinely followed in specialist (nonprimary) care, visits due to recurrence/metastases should be recorded in the NPR. This complementary analysis, however, was restricted to women residing in the Stockholm-Gotland region because the completeness of recurrence and metastasis coding in the NPR has been found to be particularly high in this region.

We analyzed the association between low-dose aspirin use, risk of breast cancer-specific death and time to recurrence/metastasis (Stockholm-Gotland region) using a Cox proportional hazards model yielding hazard ratios (HRs) with 95% confidence intervals (CIs) as measures of association. When we classified aspirin use we used a 180-day lag period in order to disregard changes in prescription patterns during the last 6 months prior to death as these may reflect changes in personal drug administration routines due to end-of-life care [29‐31]. The main models were stratified on tumor stage to allow for different baseline hazards and to account for deviations from the proportional hazards assumption. We used a crude model with no adjustments, a second model adjusted for age at diagnosis, stage, year of diagnosis, region, education level and comorbidity before diagnosis, and a third model adjusted for the same variables plus statin, metformin and NSAID use as well as oncological treatment (neoadjuvant/adjuvant therapy in four categories: chemotherapy, radiotherapy, endocrine therapy and/or trastuzumab) to illustrate the effect of stepwise addition of potential confounding factors. We adjusted for confounders on the basis of a priori selection of factors suggested in other studies to be potential confounders in analyses of aspirin use and breast cancer progression/death. In analyses of aspirin use after breast cancer diagnosis, the models were additionally adjusted for low-dose aspirin use before diagnosis. The underlying timescale was time since breast cancer diagnosis. The proportional hazards assumption was formally tested using the Grambsch and Therneau test [32]. Adjusted graphical predictions of breast cancer-specific survival according to aspirin use after diagnosis (3–9 months) were obtained from a flexible parametric survival model [33] with five degrees of freedom to model the baseline hazard function with a restricted cubic spline. For comparison, this adjusted survival function was presented together with survival estimates obtained using the Kaplan–Meier method.

We further analyzed aspirin use before and after diagnosis among subgroups of women by clinical and tumor characteristics and oncological treatment and risk of breast cancer-specific death. Thus, second sets of models were fitted for stage of disease and ER status, HER2 status, subtype and oncological treatment. Interaction tests were performed with likelihood ratio tests.

All analyses were performed with Stata 14 software (StataCorp. 2015. Stata Statistical Software: Release 14; StataCorp LP, College Station, TX, USA).

The cohort consisted of 21,414 women with breast cancer diagnosed in stage I–III (Table 1). The median age at breast cancer diagnosis was 63 years (range 19–102) with a median follow-up of 3.8 years (range 0.75–7.75). Overall, 2660 women (12.4%) were treated with low-dose aspirin before breast cancer diagnosis. Aspirin users were older at diagnosis (median age 75 years, range 31–102), and more often diagnosed with stage II–III than stage I tumors compared with nonusers (Additional file 1: Table S2). Considering the entire follow-up after breast cancer diagnosis, 4091 women (19.1%) used low-dose aspirin.

The majority of the women (12,546; 58.6%) were diagnosed with stage I breast cancer, 7879 women (36.8%) had stage II and 989 women (4.6%) had stage III disease. The most common clinical subtype was the luminal subtype (ER⁺, HER2^–/HER2⁺), recorded in 15,529 women (72.5%), whereas 857 women (4%) had nonluminal HER2 (ER^–, HER2⁺) and 1739 women (8.1%) had ER^–HER2^– breast cancer.

There were no associations between aspirin use from 9 to 3 months before breast cancer diagnosis and risk of breast cancer death when adjusted for age, stage of primary breast cancer, education, region, year of primary diagnosis and comorbidity before breast cancer diagnosis (HR 0.92, 95% CI 0.77–1.09) (Table 2). Further adjustment for use of other medications (statins, metformin and NSAIDs) before breast cancer diagnosis and neoadjuvant/adjuvant oncological treatments (chemotherapy, endocrine therapy, radiotherapy and/or trastuzumab) did not change the result (HR 0.93, 95% CI 0.77–1.12). The dose of aspirin (≤ 75 or > 75 mg/day) before diagnosis did not modify the null association. In patient subgroups by clinical characteristics, however, reduced risks of breast cancer death were suggested among women with ER⁺ tumors (HR 0.74, 95% CI 0.57–0.97) and among those with intended endocrine treatment (HR 0.75, 95% CI 0.59–0.96) (Additional file 1: Table S3).

Aspirin use during the period 3–9 months after breast cancer diagnosis did not affect the risk of breast cancer-specific death in a full multivariate model also adjusting for prediagnostic aspirin use (HR 1.00, 95% CI 0.74–1.37) (Table 2). This is also shown graphically in an adjusted survival curve and compared with the univariable Kaplan–Meier method (Fig. 1). When aspirin use over the entire follow-up post diagnosis was considered, the result was similar (HR 0.99, 95% CI 0.79–1.23) (Table 2). In general, the dose and duration of aspirin use after diagnosis was not associated with breast cancer-specific death. However, in the subgroup of women treated with aspirin > 75 mg daily during the entire follow-up, an increased risk of breast cancer-specific death was observed (HR 1.62, 95% CI 1.09–2.40). In subgroups of patients defined by clinical and tumor characteristics (stage, ER status, HER2 status, breast cancer subtype and oncological treatment), aspirin use after diagnosis was associated with a reduced risk of breast cancer-specific death among women with stage I tumors (HR 0.53, 95% CI 0.29–0.96) (Table 3). There was also a borderline significantly increased risk among women with stage III tumors.

In the Stockholm-Gotland regional cohort, there were 9226 women with stage I–III disease, of which 1048 women (11.4%) were aspirin users before breast cancer diagnosis (Additional file 1: Table S4). During follow-up, 2800 women not using aspirin (34.2%) and 347 women treated with aspirin (33.1%) had a record of a first recurrence or distant metastasis. Aspirin use was not associated with risk of recurrence/metastases in univariate analysis (HR 0.98, 95% CI 0.88–1.09) or in a model adjusted for age at diagnosis, stage, year of diagnosis, region, educational level and comorbidity before breast cancer diagnosis (HR 0.97, 95% CI 0.86–1.10).

In a separate analysis, we investigated aspirin use before breast cancer diagnosis and time to breast cancer death among 621 women with stage IV disease at diagnosis. Time to breast cancer death was not significantly different for aspirin users with stage IV disease (n = 61) when compared to nonusers (n = 334) (HR 0.91, 95% CI 0.67–1.23) in adjusted analyses.