Do adjuvant aromatase inhibitors increase the cardiovascular risk in postmenopausal women with early breast cancer

Meta-analysis of randomized trials

Abstract

BACKGROUND.

Despite the advantages from using aromatase inhibitors (AIs) compared with tamoxifen for early breast cancer, an unexpectedly greater number of grade 3 and 4 cardiovascular events (CVAE) (as defined by National Cancer Institute of Canada-Common Toxicity Criteria [version 2.0] was demonstrated.

METHODS.

Phase 3 randomized clinical trials (RCTs) comparing AI with tamoxifen in early breast cancer were considered eligible for this review. The event-based risk ratios (RRs) with 95% confidence intervals (95% CIs) were derived, and a test of heterogeneity was applied. Finally, absolute differences (ADs) in event rates and the number of patients needed to harm 1 patient (NNH) were determined.

RESULTS.

Seven eligible RCTs (19,818 patients) reported CVAE results. When considering all RCTs, the AD of the primary endpoint (CVAE) between the 2 arms (0.52%), tamoxifen versus AI, was statistically significant (RR, 1.31; 95% CI, 1.07–1.60; P = .007). This translated into an NNH value of 189 patients; when only third-generation AIs were considered, the difference (0.57%) remained significant (RR, 1.34; 95% CI, 1.09–1.63; P = .0038). Thromboembolic events were significantly more frequent in the tamoxifen arm, regardless of the strategy adopted (RR, 0.53; 95% CI, 0.42–0.65; P < .0001), without significant heterogeneity (P = .21). An AD of 1.17% and an NNH value of 85 patients were observed.

CONCLUSIONS.

According to the results from this meta-analysis, the risk of grade 3 and 4 CVAEs in patients who were receiving AIs was higher compared with the risk in patients who were receiving tamoxifen, and the difference reached statistical significance. However, the AD was relatively low, and from 160 to 180 patients had to be treated to produce 1 event. Cancer 2008. © 2007 American Cancer Society.

Breast cancer is the second most frequent cause of cancer-related death for women in the U.S.1 For 30 years, tamoxifen has been the most widely used endocrine drug for the management of all stages of breast carcinoma in women with estrogen-dependent tumors, regardless of age or menopausal status.

Aromatase inhibitors (AIs) have demonstrated superiority to tamoxifen in the setting of metastatic breast cancer.2–4 Although tamoxifen has been the cornerstone of treatment for hormone-sensitive early breast cancer, recently reported randomized clinical trials (RCTs) clearly indicate the benefit of AI.5–15 These benefits, despite strategic differences across RCTs and patient populations, are consistent with all AIs, and previously reported cumulative analyses have indicated that there is an overall survival benefit to the already documented advantage in disease-free survival (DFS).7, 14 Despite these positive aspects, toxicity profile and the pharmacoeconomic aspects should always be taken into account when discussing new treatment strategies, particularly in the presence of small benefits. Previously presented or published data ‘alarmed’ medical oncologists because of a trend (sometimes statistically significant) toward increased differences in cardiac events.10, 15 Those concerns are enhanced by the lack of a clear mechanism by which these drugs are supposed to act.16, 17

The presumed pathogenesis of cardiac pathology induced by AIs remains unclear, although it has been proposed that some pathways (reduction of circulating estradiol and alterations in lipid metabolism) are involved.16, 17 In fact, in head-to-head comparisons of AI and tamoxifen, it is not possible to distinguish between this supposed higher risk and the proposed cardioprotective effect of tamoxifen (attributed to its favorable impact on lipid profile).18 It also is important to note that the only trial comparing an AI with placebo did not identify a difference in cardiac events.19 Moreover, tamoxifen's protective effect seems to be proportional to the duration of treatment and, in itself, remains controversial,18 thus, raising the question of whether AIs actually are cardiotoxic or tamoxifen is cardioprotective. Furthermore, the need to evaluate AI-versus-tamoxifen trials to determine whether there are any differences in cardiac events comes to the surface. To gain a better understanding of whether adjuvant AIs increase the cardiovascular risk in postmenopausal women with early breast cancer and to quantify this risk, we performed a meta-analysis based on the published data to date.

MATERIALS AND METHODS

Our comprehensive analysis was conducted in 4 steps: 1) definition of the endpoints (definition of the question which the analysis was designed to answer), 2) definition of the optimal search strategy, 3) definition of the criteria for the selection of the eligible trials, and 4) a detailed description of the statistical methods used.7, 20

Endpoints Definition

Analyses were conducted to determine significant differences in primary and secondary endpoints. The primary endpoint was grade 3 and 4 cardiovascular adverse events (CVAEs), as defined by National Cancer Institute of Canada-Common Toxicity Criteria (version 2.0). Secondary endpoints were 1) thromboembolic adverse events (TEAEs) and 2) cerebrovascular adverse events (CBVAEs).

Search Strategy

The trial publication deadline for inclusion in this analysis was February 2007. Publications of trials were searched first; then, congress websites and/or specific reports were screened for updates every month. In particular, updates of RCTs were gathered through searches of the Medline (PubMed, available at URL: www.ncbi.nlm.nih.gov/PubMed), American Society of Clinical Oncology (ASCO) (available at URL: www.asco.org), European Society for Medical Oncology (ESMO) (available at URL: www.esmo.org), Federation of European Cancer Societies (FECS) (available at URL: www.fecs.be), and San Antonio Breast Cancer Symposium (SABCS) (available at URL: www.sabcs.org) web sites (each site was last accessed on February 28, 2007). The following keywords were used for the searches: adjuvant hormonal therapy, aromatase inhibitors, AIs, tamoxifen, endocrine, exemestane, anastrozole, letrozole, aminoglutetimide, switch, early, breast cancer, review, cardiovascular, ischemic, myocardial infarction, cardiotoxicity, metanalysis, meta-analysis, pooled analysis, randomized, phase 3, comprehensive review, systematic review. In addition to the Internet, the reference chapters of reviews and original articles also were scanned to search for any missing update of trials. Furthermore, lectures of major meetings (ASCO, ESMO, European Congress on Clinical Oncology, and SABCS) that had ‘adjuvant therapy for breast cancer,’ ‘early breast cancer,’ or ‘cardiotoxicity of endocrine therapy’ as topics were checked. No language restrictions were applied.

Trial Identification Criteria

We gathered phase 3 prospective and randomized trials that were published as full articles in peer-reviewed journals or that were presented at the mentioned meetings up to February 2007 in which previously untreated patients had undergone curative surgical resection for early breast cancer and were randomized to receive hormonal therapy with either an AI (experimental arm) or tamoxifen (control arm) for a total of 5 years of treatment. For entry criteria, all trials that explored adjuvant AI compared with tamoxifen either as an ‘early-switch’ strategy (after 2–3 years of tamoxifen) or as an ‘up-front’ strategy (starting at the time of surgery and planned for 5 years) were considered eligible. Phase 2 randomized trials were excluded, because such trials are not designed to compare drugs or strategies. Phase 3 trials that adopted AIs with different design, such as ‘late-switch’ trials (which adopted a placebo in the control arm after 5 years of tamoxifen), were excluded from the analysis. The AI arm was considered the experimental group, and the tamoxifen arm was considered the control group.

Data Extraction

Within each study arm for all trials, the number of grade 3 and 4 CVAEs, TEAEs, and CBVAEs were obtained. All data were reviewed and computed separately by 2 different independent investigators (E.B. and D.G.).

Data Synthesis

The log of relative risk ratios (RRs) was estimated for each studied endpoint, and 95% confidence intervals (95% CIs) were derived.21 By using this method, it was possible to apply both a fixed-effects model and a random-effects model according to the inverse variance and Mantel-Haenzel methods. To test for heterogeneity between trials, the Q statistic was used. Patients were accounted for and filled in on a 2 × 2 tables using intention-to-treat assignment.7, 22 The significance of the heterogeneity test suggested a preference for using random-effects estimation as an appropriate evaluation of the results. Absolute differences (ADs) for each endpoint were calculated (ie, AD = exp [RR · log (control rate)] − control rate).23 The number of treated patients needed to harm 1 patient (NNH) was determined (NNH: 1/[(AD)/100]).24 The results are depicted in all figures as conventional meta-analysis Forest plots. The pooled analysis calculations were accomplished by using the Comprehensive Meta-Analysis Software, version 1.0.23 (CMA; Biostat, Englewood, NJ).7, 22 The project was funded by the Italian National Ministry of Health. To reduce heterogeneity across the different generations of AIs, the analysis was carried out in consideration of 2 different populations: 1) overall analysis (ie, considering all AIs regardless of generation), and 2) third-generation AIs (CVAE-New).

RESULTS

Selected Trials

Seven trials (19,818 patients) were identified that were eligible for inclusion in this analysis5, 6, 8, 10–13, 15, 25 (Fig. 1). Two trials with similar design were pooled together and published as a single report (Austrian Breast Cancer Study Group 8/German Adjuvant Breast Cancer Group trial [ABCSG8/ARNO]).13 All trials have been published fully, and 3 of them recently were updated with more mature follow-up.8, 9, 11, 12, 26 Two trials were designed with the up-front strategy (11,108 patients),8, 9 and 5 trials were designed with the early switch strategy (8974 patients).5, 6, 10, 11, 13, 25, 26 The median follow-up ranged from 28 months to 68 months. Trial characteristics are listed in Table 1. There were no disagreements between investigators at the time of data extraction and analysis.

Figure 1. Outline of the search: flow diagram. RCTs indicates randomized clinical trials; pts, patients.

Table 1. Trial Characteristics

RCTs		Previous treatment with tamoxifen for 2–3 Years	no. of patients	Median FU, mo	AI	CVAE definition
Reference	Acronym
RCTs indicates randomized clinical trials; FU, follow-up; AI, aromatase inhibitor; CVAE, cardiovascular adverse event; Up-front strategy, AI started at time the of surgery and planned for 5 years; ATAC, Arimidex, Tamoxifen, Alone or in Combination trial; ANA, anastrozole; BIG-1-98, Breast International Group 1-98 trial; LET, letrozole; Early-switch strategy, switched to an AI after 2–3 years of tamoxifen; ITA, Intergruppo Tamoxifen Anastrozole trial; AGT, aminoglutetimide; CV, cardiovascular; GROCTA, Italian Breast Cancer Study Group; IES, Intergroup Exemestane Study; EXE, exemestane; ABCSG8/ARNO, Austrian Breast Cancer Study Group 8/German Adjuvant Breast Cancer Group trial.
Up-front strategy
Buzdar 20068; Hosell 200512	ATAC	No	6186	68 mo.	ANA	Ischemic cardiovascular
Coates 20079	BIG-1-98	No	4922	51 mo.	LET	Cardiac events including ischemic heart disease and cardiac failure
Early-switch strategy
Boccardo 200125	ITA	Yes	380	61 mo.	AGT	CV
Boccardo 20065	GROCTA	Yes	448	64 mo.	ANA	CV diseases
Coombes 200726	IES	Yes	4742	56 mo.	EXE	CV events
Jackesz 200513	ABCSG8/ARNO	Yes	3224	28 mo.	ANA	Myocardial infarction

Combined Analysis

All results obtained are displayed in Tables 2 through 5 and in Figures 2 through 4. RRs should be interpreted as follow: RRs >1.0 indicates more events in the AI arm (ie, higher risk), and RRs <1.0 indicate more events in the tamoxifen arm.

Figure 2. Primary endpoint: CVAE. CVAE indicates cardiovascular adverse events; NTotal, total number of patients; Early Switch, switched to an aromatase inhibitor (AI) after 2 to 3 years of tamoxifen; Fixed, fixed-effect model; Upfront, AI started at the time of surgery and planned for 5 years; Tam, tamoxifen.

Figure 3. Primary endpoint: CVAE-New. CVAE-New indicates cardiovascular adverse events/third-generation aromatase inhibitors; NTotal, total number of patients; Early Switch, switched to an aromatase inhibitor (AI) after 2 to 3 years of tamoxifen; Fixed, fixed-effect model; Upfront, AI started at the time of surgery and planned for 5 years; Tam, tamoxifen.

Figure 4. Secondary endpoint: TE. TE indicates thromboembolic adverse events; Early Switch, switched to an aromatase inhibitor (AI) after 2 to 3 years of tamoxifen; NTotal, total number of patients; Fixed, fixed-effect model; Upfront, AI started at the time of surgery and planned for 5 years; Tam, tamoxifen.

Table 2. Results: Primary Endpoint (Cardiovascular Adverse Events)

Strategy	no. of patients (No. of RCTs)	FEM/REM, RR [95% CI]	P	Het	AD, %	NNH
RCTs indicates randomized clinical trials; FEM, fixed-effect model; REM, random-effect model; RR, relative risk; 95% CI, 95% confidence interval; Het, heterogeneity; AD, absolute difference; NNH, number of patients needed to treat to harm 1 patient; Early switch, switched to an aromatase inhibitor (AI) after 2–3 years of tamoxifen; NS, not significant; Up front, AI started at the time of surgery and planned for 5 years.
Overall	19818 (7)	1.31 [1.07–1.60]	.007	.11	.52	189
Early switch	8710 (4)	0.97 [0.61–1.54]	.91	.47	NS
Up front	11,108 (2)	1.40 [1.12–1.74]/1.56 [0.83–2.94]	.002/.16	.034	NS

Primary endpoint

All trials (5 + 2 combined trials; 19,818 patients) were evaluable for the primary endpoint. When all AIs were considered, a significant AD of 0.52% in the overall population was observed in CVAEs for patients who received AIs (RR, 1.31; 95% CI, 1.07–1.60; P = .007) (Fig. 2), which corresponds to an NNH of 189 patients (Table 2) without significant heterogeneity. When evaluating only the trials with third-generation AIs (ie, anastrazole, exemestane, and letrozole), the difference between the 2 arms was more pronounced (RR, 1.34; 95% CI, 1.09–1.64; P = .038) (Fig. 3) without significant heterogeneity (P = .19) and with an AD of 0.57%, which corresponds to an NNH of 165 patients (Table 3). No significant differences were observed between the 2 arms for either the early-switch strategy or the up-front strategy (Tables 2 and 3).

Table 3. Results: Primary Endpoint (Third-generation Aromatase Inhibitors: Cardiovascular Adverse Events)

Strategy	no. of patients (No. of RCTs)	FEM/REM, RR [95% CI]	P	Het	AD, %	NNH
RCTs indicates randomized clinical trials; FEM, fixed-effect model; REM, random-effect model; RR, relative risk; 95% CI, 95% confidence interval; Het, heterogeneity; AD, absolute difference; NNH, number of patients needed to treat to harm 1 patient; Early switch, switched to an aromatase inhibitor (AI) after 2–3 years of tamoxifen; NS, not significant; Up front, AI started at the time of surgery and planned for 5 years.
Overall	19,438 (6)	1.34 [1.09–1.63]	.0038	.19	.57	165
Early switch	8330 (3)	1.07 [0.66–1.74]	.75	.76	NS
Up front	11,108 (2)	1.40 [1.12–1.74]/1.56 [0.83–2.94]	.002/.16	.034	NS

Secondary endpoints

The overall incidence of TEAEs, as expected, was higher in the tamoxifen arm (RR, 0.53; 95% CI, 0.42–0.65; P < 0.0001; heterogeneity test: P = .21) (Fig. 4). This risk translates into an AD of 1.17%, with an NNH of 85 patients (Table 4). Similar results were observed with both the early-switch strategy (RR, 0.33; 95% CI, 0.19–0.56; P = .00,004) (Fig. 4) and the up-front strategy (RR, 0.58; 95% CI, 0.46–0.73; P = .00,001) (Fig. 4), which translate into ADs of 0.92% and 1.39%, respectively. Thus, the NNH for TEAEs was 85 patients and 72 patients for the early-switch and up-front strategies, respectively. With regard to CBVAEs, although a strong trend in favor of AIs was observed particularly with the up-front strategy (RR, 0.76; 95% CI, 0.58–1.003; P = .053) in a homogeneous manner, the low number of events most likely was insufficient to attain statistical significant (Table 5).

Table 4. Results: Secondary Endpoint (Thromboembolic Adverse Events)

Strategy	no. of patients (No. of RCTs)	FEM, RR [95% CI]	P	Het	AD, %	NNH
RCTs indicates randomized clinical trials; FEM, fixed-effect model; RR, relative risk; 95% CI, 95% confidence interval; Het, heterogeneity; AD, absolute difference; NNH, number of patients needed to treat to harm 1 patient; Early switch, switched to an aromatase inhibitor (AI) after 2–3 years of tamoxifen; Up front, AI started at the time of surgery and planned for 5 years.
Overall	19,438 (6)	0.53 [0.42–0.65]	< .0001	.21	1.17	85
Early switch	8330 (3)	0.33 [0.19–0.56]	.00004	.58	0.92	107
Up front	11108 (2)	0.58 [0.46–0.73]	.00001	.29	1.39	72

DISCUSSION

The results from this meta-analysis indicated that previous reports with shorter follow-up overestimated the cardiovascular ischemic risk in patients who were receiving AIs because of the dilution of the events in a larger patient cohort and the updates of recent RCTs.10, 15 Indeed, the trial with the short median follow-up at the time of this analysis reported the least favorable RR for AI in terms of cardiac toxicity, as illustrated in Figures 2 and 3.13 It is interesting to note that, if this trial were excluded from our analysis, then the RR in the early-switch strategy in terms of CVAE would be less in favor of tamoxifen, although it always would remain nonsignificant (data not shown).

Although the number of CVAEs was greater for patients who were receiving AI, this translates into a low risk (approximately 0.50%), and the NNH would be >180 patients (Table 2). Although it is questionable to compare different strategies because of the different time of assignment during the natural history of cardiac disease, this issue is more controversial in survival outcome, as discussed previously.7

The main bias in interpreting this analysis (and all RCT cross-sectional reading) is the difference in reporting and defining cardiovascular events and cardiac risk factors among these RCTs (Table 1). A variety of cardiovascular endpoints or predefined parameters have been used in the majority of AI therapy trials, and there are differences in how they are reported. Indeed, in the Breast International Group 1–98 trial, the events are defined as adverse cardiac events (ischemic cardiac disease and cardiac failure) plus all cardiovascular events9, 15; whereas, in the Arimidex, Tamoxifen, Alone or in Combination trial, the events are defined only as ischemic cardiovascular events.12 In the latest Italian study, cardiac toxicity was reported as cardiovascular disease; whereas, in the previous Italian trial, it was reported as cardiovascular events.5, 25 In the Intergroup Exemestane Study trial, events were defined as cardiovascular events (ie, ischemic cardiac; such as myocardial infarction and angina) plus other cardiac events, whereas only myocardial infarction was reported in the pooled analysis of ABCSG8 and ARNO.10, 13, 26 Because of these and other differences across RCTs, any conclusion should be interpreted with caution.

To gain a better understanding of the issue of overall vascular toxicity, TEAEs and CVAEs also were investigated (Tables 4 and 5). Tamoxifen, as expected and as well documented, is associated with a significant increase in thromboembolic risk by from 1% to 1.4%, as demonstrated previously in patients with early and advanced disease.3, 27 Although they were not addressed (and perhaps not powered) for subgroup analyses, differences appear to exist between strategies, as shown in data and plots. Indeed, this risk appears to be time-dependent, ie, the longer the patients are exposed to tamoxifen (such as the up-front strategy), the higher the incidence. This length-of-tamoxifen-exposure effect is present across all endpoints (Tables 2–5, Figs. 2–4). Although there was a trend toward significance (more evident in the up-front strategy), CVAEs were not increased in patients who were receiving tamoxifen (Table 5); the lack of difference (especially in the up-front strategy) appears to be related to the small number of events. What is relevant for clinicians and patients is whether these events lead to deaths.28 Similarly, the overall outcome of those patients who experience a cardiac event is not actually reported across the examined RCTs; thus, a cumulative analysis from a literature-based approach is not possible.

Table 5. Results: Secondary Endpoint (Cerebrovascular Adverse Events)

Strategy	no. of patients (No. of RCTs)	FEM, RR [95% CI]	P	Het
RCTs indicates randomized clinical trials; FEM, fixed-effect model; RR, relative risk; 95% CI, 95% confidence interval; Het, heterogeneity; Early switch, switched to an aromatase inhibitor (AI) after 2–3 years of tamoxifen; Up front, AI started at the time of surgery and planned for 5 years.
Overall	16,146 (4)	0.84 [0.68–1.05]	.13	0.25
Early switch	5038 (2)	1.01 [0.70–1.45]	.94	0.19
Up front	11,108 (2)	0.76 [0.58–1.003]	.053	0.34

The introduction of new drugs or new treatment strategies into clinical practice should take into account the long-term safety profile data with updated and mature follow-up. A recent systematic review that approached this topic suggested that it is necessary to interpret early data (which can lead to stopping a trial prematurely) with care and skepticism.29 Many studies over the past 50 years have demonstrated that negative alteration in lipid profile is a risk factor for the development of cardiovascular disease: The lipid alterations may include increases in cholesterol, triglycerides, lipoprotein a, and low-density lipoprotein cholesterol and a decrease in high-density lipoprotein cholesterol. It has been implied that the negative to null effect of AI on lipid metabolism plays a role in the cardiotoxicity of AI, although no definitive evidence is available.13–15 In addition, steroidal AI actually may improve triglycerides and apolipoprotien A. Therefore, do AIs damage cardiovascular health directly compared with to tamoxifen, or are the differences observed caused by the proposed protective effects of tamoxifen (mainly because of positive lipidic effects)? This still is debated and has been explored in other meta-analyses.17, 30 The lack of increased CVAEs with letrozole compared with placebo in the MA.17 trial suggests that the differences between AIs and tamoxifen in other trials are caused by tamoxifen-based protection rather than because AIs actually increase cardiac events.17, 19 Because placebo-controlled trials in the prevention setting previously demonstrated the exact safety profile of tamoxifen without confounding factors, the ongoing RCTs in the very same setting will clarify definitively the real safety outline of AIs without biases provided by the disease itself or by comparative treatment.17

In analyzing these data, the difference in cardiac events between AI and tamoxifen does seem to exist, although it appears to be minimal. This needs to be balanced against the relevant toxicities of tamoxifen, such as thromboembolism, stroke, and endometrial carcinoma,7, 27 The effects of these toxicities on overall survival also must be considered. In addition, regarding the bone toxicity of AIs, promising data regarding the protective role of bisphosphonates to mitigate fracture risk in AI therapy are emerging in the literature.31, 32

Although the impact on clinical practice and research given by systematic reviews and meta-analyses33 is relevant, there are many drawbacks to literature-based approach.34 The difference in median follow-up between trials, given the hypotheses regarding its role in late toxicity profiles of new unknown drugs, is a crucial issue for this topic. A patient-based meta-analysis gathering the database from each individual trial is able to answer this specific question, but it is time-consuming, costly, and generally requires years to complete.34, 35 Given these general biases of this meta-analytic approach, if we consider the specific mentioned biases, such as differences in the definition of cardiac toxicity (Table 1), and the lack of an eventual link between these data and deaths, then our results should be carefully evaluated.

Notwithstanding all of the limitations of a literature-based meta-analysis, our results provide updated data regarding the risk of grade 3 and 4 cardiac events with adjuvant AI therapy in patients with breast cancer. Although the difference is present, it appears lower than was suspected previously. Hence, clinicians need to balance these differences in cardiac toxicity with the improved toxicity observed with AI regarding the thromboembolic risk, and the possible cerebrovascular risk, and improved disease-free survival when discussing and making treatment decisions for their patients.

Acknowledgements

We thank Paola Muti, MD (Scientiphic Director, ‘Regina Elena’ National Cancer Institute, Rome, Italy) for financial support, and Mariangela Ciccarese, MD (‘Vito Fazzi’ Hospital, Lecce, Italy) for critically reviewing and editing the article.