CVD indicates cardiovascular disease.
eTable 1. Demographic Characteristics of Post-Menopausal Breast Cancer Survivors Diagnosed Between 1991 and 2010 (Followed Through 2011)
eTable 2. Tumor Characteristics of Breast Cancer Survivors
eTable 3. Adjusted Hazards Ratios for CVD Events Among Those With Adjuvant Radiotherapy
eTable 4. Adjusted Hazards Ratios for CVD Events by Endocrine Treatment Use Among Women With MPR>80% for Tamoxifen and AIs
Customize your JAMA Network experience by selecting one or more topics from the list below.
Haque R, Shi J, Schottinger JE, et al. Cardiovascular Disease After Aromatase Inhibitor Use. JAMA Oncol. 2016;2(12):1590–1597. doi:10.1001/jamaoncol.2016.0429
Copyright 2016 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
What is the long-term effect of aromatase inhibitors (AIs) on cardiovascular disease risk in breast cancer survivors?
In this retrospective cohort study of 13 273 women with 72 886 person-years of follow-up, AI-only users had a similar risk of cardiac ischemia and stroke as tamoxifen-only users.
The risk of cardiac ischemia and stroke was not elevated in AI-only users compared with tamoxifen-only users, providing reassurance that AIs may not increase the risk of the most fatal cardiovascular events.
Cardiovascular disease (CVD) is an important cause of death in older patients with breast cancer. However, limited information exists on the long-term effect of aromatase inhibitor (AI) use on CVD risk in breast cancer survivors. To this point, no other population-based studies have been able to adjust for CVD risk factors or cardiovascular medications.
To determine the long-term influence of adjuvant endocrine therapies on CVD in a cohort of postmenopausal breast cancer survivors in analyses that accounted for major CVD risk factors, medication use, chemotherapy, and radiotherapy.
Design, Setting, and Participants
A retrospective cohort of postmenopausal women with breast cancer diagnosed from January 1, 1991, to December 31, 2010, and followed up through December 31, 2011 (maximum, 21 years [72 886 person-years]), was evaluated using records from a managed care organization with nearly 20 community hospitals in California. A total of 13 273 postmenopausal women with hormone receptor–positive breast cancer without prior CVD were included. Cardiovascular disease incidence was compared across endocrine therapy categories. Information on demographics, comorbidity, medication, use, and CVD risk was captured from electronic health records. Multivariate Cox proportional hazards models using time-dependent endocrine drug use variables and propensity scores were conducted. Data analysis was conducted from September 15, 2014, to February 1, 2016.
Women were grouped by endocrine therapy status (tamoxifen citrate only, AI only, both, or neither).
Main Outcomes and Measures
Person-year rates of CVD for each therapy group.
During 72 886 person-years in 13 273 women (mean [SD] age, 66.8 [8.1] years) with follow-up through 2011, we observed 3711 CVD events. In multivariable analyses (reported as hazard ratio [95% CI]), AI-only users had a similar risk of cardiac ischemia (myocardial infarction and angina) (adjusted, 0.97 [0.78-1.22]) and stroke (adjusted, 0.97 [0.70-1.33]) as tamoxifen-only users (reference). However, we found an increased risk of other CVD (dysrhythmia, valvular dysfunction, and pericarditis) (adjusted, 1.29 [1.11-1.50]) in women who used AIs only or sequentially after tamoxifen (1.26 [1.09-1.45]) vs tamoxifen (reference) as well nonhormone users (1.18 [1.02-1.35]).
Conclusions and Relevance
The risk of the most serious cardiovascular events (cardiac ischemia or stroke) was not elevated in AI-only users compared with tamoxifen users. The finding that other CVD events combined were greater in AI users requires further study.
Aromatase inhibitors (AIs) are considered superior to tamoxifen citrate in reducing recurrence risk in postmenopausal women with hormone receptor–positive breast cancer.1,2 However, although tamoxifen has favorable effects on lipid profiles, AIs have unfavorable effects, raising questions on whether these 2 therapies differentially affect cardiovascular disease (CVD) risk.3-9
Because CVD is an important cause of death in older breast cancer survivors, accurate information on such events associated with tamoxifen and AI use is needed to inform clinical decision making.6,7 Unfortunately, available evidence is mixed and/or inconclusive. For tamoxifen, no cardiac protective effect was seen in meta-analyses10 of placebo-controlled, adjuvant breast cancer trials. By contrast, in the Adjuvant Tamoxifen: Longer Against Shorter study,11 using tamoxifen for 10 years was associated with fewer instances of CVD compared with drug discontinuation.
Concerns regarding the adverse influence of AIs on CVD were raised by 2 meta-analyses12,13 of adjuvant breast cancer trials in which AI use was associated with higher CVD risk (odds ratio, 1.26; 95% CI, 1.10-1.43). The CVD outcomes in most of the trials were based on unverified self-reports collected as information on toxic effects. Conversely, no increase in myocardial infarction or stroke risk with AIs was found in recent placebo-controlled, breast cancer prevention trials.14,15 Given inconclusive evidence, we examined the long-term influence of adjuvant endocrine therapies on CVD in a cohort of 13 273 postmenopausal breast cancer survivors accounting for major CVD risk factors, medication use, chemotherapy, and radiotherapy.
This study was conducted at Kaiser Permanente Southern California, a managed care system with 16 hospitals and approximately 4.0 million members. Electronic health records captured information on pharmacy, outpatient and inpatient diagnoses, census, and California state mortality data. The health plan’s National Cancer Institute Surveillance, Epidemiology, and End Results (SEER)–affiliated cancer registry was used to identify patients. Kaiser Permanente Southern California’s institutional review board approved this study; written informed consent was waived. All analysis was conducted at Kaiser Permanente Southern California; no data were shared.
We assembled a retrospective cohort of women with a first diagnosis of primary breast cancer between 1991 and 2010 and observed them through December 2011. For eligibility, women had to be older than 18 years, have pharmacy benefits, and have estrogen– or progesterone receptor–positive breast cancer. Of 24 837 breast cancer survivors meeting the criteria, women with prior CVD (n = 4174) and premenopausal women (n = 7390) were excluded, leaving a final cohort of 13 273 postmenopausal women (Figure). Age at breast cancer diagnosis was used as a proxy for menopausal status (age >55 years was considered postmenopausal).
Electronic inpatient and outpatient diagnoses, based on International Classification of Diseases, Ninth Revision (ICD-9), identified toxic cardiovascular effects and categorized them into 4 major groups: (1) cardiac ischemia (acute myocardial infarction and angina), (2) stroke, (3) heart failure and cardiomyopathy, and (4) other events (dysrhythmia, valvular dysfunction, and pericarditis). The selection of the ICD-9 codes was based on studies16-19 demonstrating their validity and expert opinion from the study oncocardiologist (A.B.).
Women were assigned to 1 of the 4 outcome categories based on the date of their first diagnosis; inpatient codes were prioritized over outpatient codes. Those with multiple CVD diagnoses on the same day were assigned the diagnosis with the greatest mortality risk.16 Thus, such women were censored (ie, follow- up ceased) on the diagnosis date of their first CVD.
Information on first-course cancer therapy (ie, surgery, radiotherapy, and chemotherapy) was extracted from the SEER registry. Data on tamoxifen and AIs (ie, letrozole, anastrozole, and exemestane) were extracted from pharmacy records including the dispensing dates and number of days the drugs were supplied. Total duration of therapy was calculated by summing the total days supplied starting from the earliest dispensing date after breast cancer diagnosis and ending at 1 of the study end points (defined below).
Women were classified into 1 of 4 endocrine therapy categories: (1) tamoxifen only, (2) AI only, (3) both, and (4) nonusers of endocrine therapy. Women were categorized as both if they used tamoxifen and AIs regardless of drug sequencing. Of 2682 women in this category, 474 (17.7%) used AIs first and therapy was then switched to tamoxifen. We retained such women in the analyses because our goal was to examine the cumulative effect of endocrine therapy. Women who never received endocrine therapy or who received these agent for less than 6 months were categorized as nonusers.
Endocrine therapies were also examined as binary time-dependent variables (ie, 0 up to start date and 1 after start date of each therapy) in the multivariate Cox proportional hazards models. We calculated the person-years of endocrine therapy exposure for each patient by summing their total exposure to each endocrine treatment (denominator) to calculate the CVD rates. Patients in whom endocrine therapy was switched contributed person-time to the both category. The time between cancer diagnosis and start date of the first endocrine therapy contributed to the nonuser category, thereby reducing immortal time bias. This enabled the patient’s person-time to contribute to both the endocrine treatment user and nonuser categories. The CVD outcomes (numerator of the rates) were assigned to the drug exposure category based on the history of all endocrine therapy used.
Covariates included age at diagnosis, diagnosis year, breast cancer stage, race/ethnicity (from the SEER registry), geocoded median household income, body mass index (BMI), medical center, tumor characteristics, and primary cancer treatment (surgery, radiotherapy, and chemotherapy). Comorbidities, captured in the year before breast cancer diagnosis, included hypertension, diabetes mellitus, and the Charlson comorbidity index score.20 Data on pharmacy use related to CVD therapy and/or prevention were also extracted. These drug covariates were coded as binary (ever or never).
Follow-up commenced on the breast cancer diagnosis date and ended on the date of one of the study end points (first CVD diagnosis date, death, termination of health plan membership, or study’s end [December 31, 2011]), whichever occurred first. Thus, the maximum follow-up length was 21 years. Differences in demographics, health care utilization, and tumor characteristics by endocrine treatments were first examined by χ2 or Fisher exact tests. All P values presented were 2-sided. We also calculated person-year rates of each CVD diagnosis. For flexibility of comparisons, we presented descriptive statistics for each endocrine therapy and nonuser group. In the multivariate models, all comparisons used the tamoxifen-only group as the reference.
Hazard ratios (HRs) and 95% CIs were estimated using Cox proportional hazards models with time-dependent medication use variables as described above; these models were adjusted for the aforementioned covariates. All variables selected for the model were based on clinical importance and descriptive statistics. The model accounted for the following covariates: CVD medications, race/ethnicity, cancer stage and patient’s age at breast cancer diagnosis, nodal status, geocoded income, diabetes mellitus, hypertension, first-course chemotherapy use, and radiotherapy. We checked for collinearity and conducted tests for interaction (deemed significant at P < .05). The proportional hazards assumption was tested via graphic plots and Schoenfeld residuals. Time since breast cancer diagnosis was used as the time scale. We also conducted complementary Cox proportional hazards models that incorporated the inverse probability of treatment weighting based on the propensity scores for having tamoxifen or AIs constructed using logistic regression. All analyses were performed using SAS, version 9.3 (SAS Institute Inc). Data analysis was conducted from September 15, 2014, to February 1, 2016.
Sensitivity analyses were conducted for the other category of CVD (atrial fibrillation ICD-9 codes 427, 427.31, and 427.32), and to assess the impact of radiotherapy (subset of 7982). Additional sensitivity analyses were based on adherence to endocrine therapies incorporating the subset of women with more than an 80% medication possession ratio, which is considered standard adherence.21 The medication possession ratio was estimated as the number of days supplied (excluding last refill) divided by the number of days between the first and last dispense dates. We also conducted additional sensitivity analyses based on women diagnosed with stage 0 to III disease (n = 12 924).
Patients’ baseline demographic characteristics, comorbidities, and CVD risk factors are displayed in eTable 1 in the Supplement. In the study cohort of 13 273 women (mean [SD] age, 66.8 [8.1] years) with incident breast cancer, 4207 (31.7%) received tamoxifen only, 3807 (28.7%), AI only, 2682 (20.2%), both, and 2577 (19.4%), no endocrine therapy. Mean (SD) duration of therapy was 2.8 (2.0) years with tamoxifen and 2.3 (1.8) years with AIs. Tamoxifen-only users, compared with AI-only users, tended to be older, in the lower quartile of median household income, and less likely to be overweight or obese. Given the later dates of marketing of AI, almost all users initiated therapy after 2000. Those exposed to tamoxifen only and those who received both treatments tended to be healthier, with 3512 (83.5%) and 2072 (77.3%) having no prior comorbidities, respectively, than 2568 (67.5%) of those who used AIs only or 1892 (73.4%) nonusers.
Endocrine therapy was strongly associated with tumor characteristics (eTable 2 in the Supplement); AI-only users had larger tumors diagnosed at higher stage and received chemotherapy more frequently than did tamoxifen-only users. A larger proportion of AI-only users received breast-conserving surgery compared with tamoxifen-only users.
Use of statins and other antilipemic agents, antihypertensive, and antidiabetic medications were significantly greater in the AI-only group (Table 1). Cardiac medication use in the nonuser group was generally similar to that of tamoxifen-only users.
The maximum follow-up was 21 years (median, 4.5 years). A total of 6544 (49.3%) of the 13 273 women in the cohort reached the end of the study, and 3711 (28.0%) patients developed CVD in the ensuing 72 886 person-years of follow-up. The median times to incident CVD were 5.0 years (interquartile range [IQR], 2.5-8.4 years) for tamoxifen only, 2.3 years (IQR, 1.2-4.0 years) for AI only, and 3.6 months (IQR, 1.0-40.8 months) for nonusers.
The highest crude person-year rate was seen for other CVD (25.3 per 1000 person-years), followed by cardiac ischemia (11.5 per 1000 person-years), heart failure/cardiomyopathy (8.6 per 1000 person-years), and stroke (5.4 per 1000 person-years) (Table 2). The lowest rate for cardiac ischemia was seen among AI-only users (8.1 per 1000 person-years) vs tamoxifen-only users (11.1 per 1000 person-years). Similarly, the lowest stroke rates were observed in the AI-only group (4.0 per 1000 person-years) vs tamoxifen only (5.9 per 1000 person-years).
Table 3 presents the overall and adjusted HRs and 95% CIs of the CVD events. In the full cohort, no significant difference in risk of cardiac ischemia or stroke was seen comparing those who used AIs only and both treatments vs tamoxifen-only users. In the Cox proportional hazards models that incorporated the inverse probability of treatment weighting based on the propensity scores for the endocrine treatments, we again found no associations (Table 3). The concordance measures and 95% CIs for the Cox proportional hazards models for each outcome were above 80%, demonstrating the overall good discriminant capability of the models. In a sensitivity analysis based on women with good drug adherence (medication possession ratio, >80%), the HRs were similar to those in the full cohort (eTable 4 in the Supplement). In other sensitivity analyses, Cox proportional hazards models based for women with stage 0 to III breast cancer (n = 12 924), the adjusted HRs were again similar to results of the full cohort (n = 13 273) (Table 3).
A somewhat higher risk of heart failure and cardiomyopathy was seen in AI-only users (adjusted HR, 1.10; 95% CI, 0.86-1.40) and in users of both treatments (adjusted HR, 1.27; 95% CI, 0.99-1.60) compared with tamoxifen-only users (Table 3). In addition, women exposed to AIs only and those who received both endocrine treatments had 29% and 26% higher risks of the category of other CVD (dysrhythmia, arrhythmia, and pericarditis), respectively, vs tamoxifen-only users. The HRs were similar when we ran the multivariable model with and without BMI, suggesting that BMI was not a strong confounder. Of note, the fraction of overweight/obese women was similarly high in all 4 groups (70% in all 4 groups) (eTable 1 in the Supplement). In a sensitivity analysis limited to atrial fibrillation, a higher HR was noted for AI-only users, but the 95% CIs included the null.
When we examined the adjusted HR for each outcome by duration of tamoxifen (<1, 2-3, 4-5, and ≥5 years) and AI (<1, 2-3, and ≥3 years), we did not observe any statistically significant trends, although the HRs for other CVD outcomes increased with greater duration of AIs (vs nonusers of endocrine therapy) (Table 4).
We conducted additional sensitivity analyses in a subset of 7982 patients who received radiotherapy. The adjusted HRs were greater in women who underwent left-sided breast radiotherapy. For example, among the AI-only cohort, in those who received left-sided breast radiotherapy the adjusted risk for stroke was 15% higher (HR, 1.15; 95% CI, 0.69-1.92) and 21% higher for heart failure/cardiomyopathy (HR, 1.21; 95% CI, 0.79-1.83) compared with those risks in tamoxifen-only users, but the 95% CIs included the null (eTable 3 in the Supplement).
In 13 273 postmenopausal women with hormone receptor–positive breast cancer, AI users had similar risk for cardiac ischemia (myocardial infarction and angina) and stroke as did tamoxifen users. These results provide reassurance that AIs may not increase risk of the most fatal CVDs. The association between AI use and other cardiac events (dysrhythmia, valvular dysfunction, and pericarditis) and heart failure/cardiomyopathy were unexpected and require further study. Based on our literature review, to our knowledge, our analysis provides the first comprehensive assessment of AI influence that accounted for CVD risk factors and cardiovascular medication use.
The present study differs from findings of meta-analyses12,13 of randomized clinical trials (mainly based on secondary analyses of CVD) comparing tamoxifen with AI use. In contrast to those trials, in the present study, we excluded women with a history of CVD and accounted for major disease risk factors. Specifically, because antilipemic (statins), antihypertensive, and antidiabetes drugs were more commonly used by women who received AI therapy compared with those who received tamoxifen, we adjusted for these medications. The American College of Cardiology/American Heart Association22 recently revised the guidelines for CVD primary prevention by expanding the indications for statins. For example, statin use is now recommended for women older than 70 years regardless of laboratory findings or family history. The present study results suggest that oncologists should familiarize themselves with these new guidelines for implementation in breast cancer survivors.
Although early reports3,5 suggested adverse influence of AIs on lipoproteins, more recent studies4 question that paradigm. No adverse lipid effects for anastrozole were observed in the Study of Anastrozole With the Bisphosphonate Risedronate trial23,24 or for extended adjuvant exemestane therapy25 after tamoxifen. Two studies26,27 found adverse AI effects after tamoxifen use, but such results are consistent with either an AI effect and/or an effect of tamoxifen washout. In the MA.17 substudy,28 although significant increases in total cholesterol and low-density lipoprotein cholesterol levels were seen with letrozole following tamoxifen use, no difference in CVD incidence in the letrozole vs placebo groups was found.
Two observational studies examining AI use and CVD provided conflicting information. Ligibel and colleagues29 found similar risks of myocardial infarction and stroke risk among AI vs tamoxifen users or no therapy in a retrospective study of 44 463 women. However, information on CVD risk factors or medication use was not available. In contrast, in a population-based cohort of 2542 patients with breast cancer, researchers30 found a significantly increased CVD risk among AI users (HR, 1.42; 95% CI, 1.09-1.84), but outcomes were self-reported and covariate medication use was not collected.
The association of AIs with other CVD outcomes, including heart failure and cardiomyopathy, was unexpected in our study. Our literature review could not identify a potential mediating mechanism through which estrogen reduction by AIs would influence these outcomes.
Our study has strengths. Although our median follow-up time was 4.5 years, 1991 (15.0%) of the participants had more than 10 years of follow-up, and 101 618 (80.0%) breast cancer survivors stayed within the health plan. Even after disenrollment, we ascertained CVD deaths. Given the managed care setting, bias through variable medical coverage was avoided. Stroke, myocardial infarction, angina, and heart failure are major clinical conditions that warrant attention and therapy regardless of breast cancer recurrence state and, therefore, were accurately captured in the electronic health records. Comprehensive pharmacy data handled as time-dependent medication variables, the set of confounders, and sensitivity analyses including propensity score methods enhanced our study design.
A unique study strength is the comprehensive attention to CVD prevention in the Kaiser Permanente Southern California health plan. Compared with multicenter, randomized adjuvant trials, mainly conducted more than 2 decades ago, breast cancer survivors in this health plan likely received more attention to blood pressure control and CVD risk factor management. Our results are generalizable to the larger California population given that racial/ethnicity distribution in the Kaiser Permanente Southern California database is similar to that of the community.
Certain limitations must be considered. We did not have data on smoking, physical activity, or over-the-counter drugs, and many records were missing BMI. However, we adjusted for hypertension and diabetes mellitus, which are correlated with BMI, and used propensity score analysis. Furthermore, the fraction of overweight/obese women was balanced across each group, and the HR estimates were the same when we did and did not include BMI in the multivariate models. Nevertheless, residual confounding by BMI is possible. We could not examine the risk by type of chemotherapy because it was not available in the electronic health record. Although anthracyclines may increase CVD risk, their use was limited in the Kaiser Permanente Southern California health plan. Similarly, approximately 10% of the cohort was HER2 positive, and a fraction of these might have been candidates for trastuzumab therapy. Although we adjusted for HER2 status, residual confounding by trastuzumab might persist. However, because adjuvant trastuzumab became available in 2005, relatively few person-years in this cohort were influenced by differential use. Although women who used AIs were approximately twice as likely to receive adjuvant chemotherapy compared with tamoxifen only users, cardiac ischemia and stroke rates were similar to those of tamoxifen only users. Longer observation times may be necessary to identify more CVD outcomes, particularly in the AI exposed group. In addition, we could not capture chemotherapy regimens among women who had a recurrence (approximately 15% over a 20-year follow-up period in this health plan). In addition, although it was necessary to ascertain the cause of death from California’s death records, it is possible that there was some misclassification.31
Aromatase inhibitor use was not associated with an increased risk for cardiac ischemia or stroke in analyses that controlled for CVD risk factors and covariate medications. The association of other CVD outcomes with AIs was unexpected and requires further study. Overall, the findings provide reassurance when considering AI use for adjuvant breast cancer therapy and potentially for ductal carcinoma in situ therapy or breast cancer prevention.
Corresponding Author: Reina Haque, PhD, Department of Research and Evaluation, Kaiser Permanente Southern California, 100 S Los Robles, Second Floor, Pasadena, CA 91101 (email@example.com).
Accepted for Publication: February 11, 2016.
Published Online: April 21, 2016. doi:10.1001/jamaoncol.2016.0429.
Author Contributions: Drs Haque and Shi had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Haque, Shi, Amundsen, Chlebowski.
Acquisition, analysis, or interpretation of data: Haque, Shi, Schottinger, Chung, Avila, Xu, Barac.
Drafting of the manuscript: Haque, Avila, Xu, Barac.
Critical revision of the manuscript for important intellectual content: Haque, Shi, Schottinger, Chung, Amundsen, Barac, Chlebowski.
Statistical analysis: Haque, Shi, Xu.
Obtained funding: Haque, Chlebowski.
Administrative, technical, or material support: Haque, Schottinger, Avila, Amundsen, Chlebowski.
Study supervision: Haque.
Conflict of Interest Disclosures: Kaiser Permanente Southern California received research funding from Novartis and Astra-Zeneca on topics not related to the present study (Drs Haque and Schottinger). Dr Chlebowski has received speaker’s fees and honorarium from Novartis, and personal fees from Genentech, Pfizer, Amgen, Genomic Health, and Novo Nordisk. Dr Barac has received honoraria, consulting/travel fees, or research funding unrelated to the study topic from Genentech/Roche, Cell Therapeutics (CTI BioPharma), Corelab, JenaValve, Biotronik Mitralign, and St Jude. No other disclosures were reported.
Funding/Support: This study was funded by California Breast Cancer Research Program grant 190OB-0201, Cardiovascular Toxicity Following Aromatase Inhibitor Use, with supplemental support from National Institutes of Health/National Cancer Institute grant R01CA136743, Antidepressants & Breast Cancer Pharmacoepidemiology (Dr Haque).
Role of the Funder/Sponsor: The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Previous Presentation: Preliminary results of this study were presented at the 2014 San Antonio Breast Cancer Symposium; December 11, 2014; San Antonio, Texas.
Create a personal account or sign in to: