Adjuvant aromatase inhibitor therapy and early markers for cardiovascular disease in breast cancer survivors

Aromatase inhibitors (AIs) are an important component of the adjuvant treatment of hormone receptor positive breast cancer (BC) but concerns regarding their cardiovascular safety remain. In this cross-sectional study nested in a breast cancer cohort, we investigated the association between AI exposure and early markers for cardiovascular disease in BC survivors. The study population consisted of 569 women, who were 5–7 years (n = 277) or 10–12 years (n = 292) after BC diagnosis. All participants underwent carotid ultrasound, skin autofluorescence measurement and laboratory evaluation. To quantify AI exposure, we obtained the AI ratio by dividing the duration of AI use by the total duration of endocrine therapy (ET). Patients were classified according to their AI ratio into low (no ET or AI ratio < 0.40), intermediate (0.40 ≤ AI ratio ≤ 0.60) or high AI exposure (AI ratio > 0.60). The association between AI ratio and carotid intima media thickness (cIMT), advanced glycation end products (AGEs) and the presence of dyslipidemia was assessed using linear and logistic regression. Median age at study visit was 55.5 years (range 45.2–63.8). Forty percent (n = 231) of the study population had used AIs, of whom the majority sequentially with tamoxifen; median duration of AI use was 3.0 years. Mean cIMT and mean AGEs did not differ across AI exposure groups in univariable and multivariable analysis. The occurrence of dyslipidemia did not vary across AI exposure groups. Intermediate AI exposure was associated with more frequent occurrence of the combined endpoint (elevated cIMT, elevated AGEs and/or dyslipidemia). This association, however, was not present in the group with highest AI exposure. AI exposure was not associated with cIMT, AGEs or the presence of dyslipidemia. These results do not prompt a change in current clinical practice, although further research is warranted to validate our findings over time and in different BC populations. Trial registration number (clinicaltrials.gov): NCT02485626, June 30, 2015.


Introduction
Endocrine therapy (ET) is a key component in the adjuvant treatment of hormone receptor positive early breast cancer (BC). Tamoxifen was one of the first endocrine agents that became available, followed by aromatase inhibitors (AIs) [1]. Both tamoxifen and AIs negate the proliferative effects of estrogen on breast cancer cells, but their working mechanism differs [2]. Tamoxifen competitively antagonizes estrogen at its receptor site, but also has partial estrogen-agonist effects. AIs inhibit the enzyme aromatase, thereby inhibiting estrogen synthesis in peripheral tissue. Oncologic outcomes in early BC improved substantially as a result of the addition of AIs to the ET armamentarium [3]. Furthermore, AIs are not associated with an increased risk of thromboembolic events or endometrial cancer, whereas tamoxifen is [4,5].
An important unresolved issue, however, is the cardiovascular safety of AIs. Several clinical trials raised concern about higher rates of cardiovascular events in patients treated with AIs compared to tamoxifen [6,7]. Whether or not this reflects a true detrimental effect of AIs on cardiovascular disease (CVD) risk or a cardioprotective effect of tamoxifen (mainly attributed to the favorable effect on lipid profile [8]) is a matter of controversy. Several systematic reviews and meta-analyses have addressed the issue, with varying outcomes [9][10][11][12][13][14][15][16]. Given the oncological relevance of AIs, it is important to gain more insight into the cardiovascular safety of AIs and identify potential mechanisms for AI-induced CVD.
Previous studies on AI-induced CVD have mainly focused on clinically apparent CVD. Whilst clinical CVD is the most important outcome from a patient's perspective, these studies often require long follow-up and large patient numbers. In our study, we focus on subclinical measures for CVD, which are more readily available. Subclinical measures can increase knowledge on the underlying biology of the potential relation between endocrine therapy and CVD, and provide early clues for increased CVD risk, thereby enabling early interventions.
Intima media thickness (IMT) functions as a surrogate measure for atherosclerosis, and carotid-wall IMT (cIMT) predicts CVD and CVD mortality [17]. Advanced glycation end products (AGEs) are metabolic or oxidative stressderived end products of sugars, usually protein-bound. The presence of AGEs in the skin is independently associated with adverse cardiovascular events [18]. Cholesterol is a widely acknowledged independent risk factor for CVD, and is often mentioned as an important intermediate factor in the relation between AI use and CVD [19].
In this cross-sectional study nested in an established BC cohort, we aimed to investigate the association between AI use and subclinical measures for CVD (cIMT, AGEs and cholesterol) in BC survivors.

Study design
We performed a cross-sectional study in an established cohort of women treated for invasive BC or ductal carcinoma in situ (DCIS) [20,21]. Eligible patients had received treatment for invasive BC (TNM stage I-III) or DCIS at age 40-50 years in the Netherlands Cancer Institute-Antoni van Leeuwenhoek (NKI-AVL) or University Medical Center Groningen (UMCG) between 2002 and 2012. They were either 5-7 or 10-12 years after initial treatment. Patients could not participate if they had previously received radiotherapy or chemotherapy unrelated to BC/DCIS. Patients with a locoregional BC/DCIS recurrence or second BC/ DCIS after initial diagnosis could participate if there was no ongoing therapy for recurrent or second disease. Patients with a history of overt CVD (defined as heart failure, acute coronary syndrome, coronary revascularization intervention, symptomatic valvular dysfunction or cardiomyopathy) before BC/DCIS diagnosis were excluded; patients who developed overt CVD after BC/DCIS diagnosis were included. The institutional review board of the NKI-AVL approved the study and it is registered with ClinicalTrials. gov, identifier NCT02485626.

Procedures
A written invitation describing the study objectives and procedures was sent to all eligible women. Non-responders received up to two reminders. Of 911 invited women, 569 provided informed consent and completed the study visit. Participants filled out a baseline questionnaire, including items on current and past lifestyle factors, the presence of cardiovascular risk factors, family history of CVD, and the use of and compliance with ET. Detailed data on tumor and treatment characteristics, including the use and duration of ET, medical history, cardiovascular risk factors and medication use were abstracted from medical records or obtained from hospital registries and the participants' general practitioner. At study visit, sociodemographic variables, recent medical history and current medication use were recorded. Participants underwent standardized physical examination, blood sampling, electrocardiography, ultrasound of the common carotid and femoral arteries for IMT measurement, and skin autofluorescence to measure AGEs.

Measurements
Mean cIMT was measured in millimeters (mm) at the far wall of the left and right common carotid arteries using the Logiq E9 (GE Healthcare) ultrasound system at the NKI-AVL and the MyLab One (Esaote) ultrasound system at the UMCG. For our analysis, we used the average of the three left-sided and three right-sided mean cIMT measurements.
Skin autofluorescence was measured at the volar side of the lower arm left and right, three times at each side using the AGE reader mu (Diagnoptics). AGEs are expressed in arbitrary units (AU), with higher values indicating higher CVD risk. The mean value of all six measurements (three left, three right) was used in the analysis.
We used fasting serum blood samples to determine a complete cholesterol profile, including high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, total cholesterol and triglycerides.

Statistical analysis
Baseline characteristics were compared between patients who received no ET, tamoxifen only, AI only or both tamoxifen and AI using Kruskal-Wallis tests for continuous variables and chi squared tests for categorical variables. To quantify AI exposure, we calculated the AI ratio by dividing the duration of AI use by the total duration of ET: the higher the ratio, the longer the (relative) AI exposure [22]. For patients who did not receive any ET, the AI ratio was set to zero. The use of the AI ratio enabled us to account for accessory tamoxifen use and to (at least partially) overcome the effect of missing data: even if absolute ET duration was missing, the AI ratio could be determined for patients who used tamoxifen or AI only (0 and 1, respectively). We categorized patients into three groups according to their AI ratio. The first group (low AI exposure) consisted of patients who had used no ET or only/predominantly tamoxifen (AI ratio < 0.40). The second group (intermediate AI exposure) consisted of patients who had used both tamoxifen and AI for approximately equal durations (0.40 ≤ AI ratio ≤ 0.60), and the third group (high AI exposure) consisted of patients who had used only/predominantly AI (AI ratio > 0.60). To assess the robustness of the chosen categories, we performed sensitivity analyses with different AI ratio categorizations, absolute AI duration and long versus short AI use (AI use ≥ 5 years versus < 5 years) as independent variables.
We examined the association between AI ratio and cIMT and AGEs as continuous variables in linear regression models. We also assessed association with high cIMT and high AGEs in logistic regression models. High cIMT was defined as a cIMT value above the 90th percentile threshold per institute. High AGEs were defined as an AGE value above 1 standard deviation of the age-adjusted mean reference value [23]. For cholesterol, we tested the association between AI ratio and the presence of dyslipidemia at study visit, defined as fasting LDL cholesterol > 4.0 mmol/L, HDL cholesterol < 1.2 mmol/L, triglycerides > 4.0 mmol/L, or current lipid lowering treatment [23]. In search of any signal for association, we also assessed associations of the AI ratio with a combined endpoint, which consisted of either high cIMT, high AGEs or dyslipidemia, or a combination of these. Participants with dyslipidemia at time of breast cancer diagnosis were excluded from all analyses that included dyslipidemia as endpoint.
Potential patient-related confounders included age at study visit, body mass index (BMI) at study visit, diagnosis of hypertension at BC diagnosis (defined as a systolic blood pressure ≥ 140 mmHg or diastolic blood pressure ≥ 90 mmHg, the use of antihypertensive drugs or as specified in the medical record), the presence of other circulatory disease at BC diagnosis (defined as any diagnosis from category I00-I99, except I10-I15, according to the International Classification of Diseases, 10th revision (ICD10)) or endocrine disease at BC diagnosis (defined as any diagnosis from category E00-E90 according to ICD10), smoking habits and menopausal status at study visit (based on medical records and patient questionnaires at BC diagnosis, and estradiol and FSH levels at study visit). Because oophorectomy status and luteinizing hormone releasinghormone (LHRH) therapy were both strongly correlated with menopausal status, we excluded these parameters. Treatment-related confounders included the use of chemotherapy, trastuzumab and radiotherapy. Each potential independent predictor was tested in a univariable analysis first and included in multivariable analysis if the p-value 1 3 in univariable analysis was ≤ 0.1. If relevant, we tested for interaction between variables.
Significance tests were two-sided, and a p value of < 0.05 was considered statistically significant. All analyses were performed with IBM SPSS Statistics 27.

Study population
The total study population consisted of 569 patients. Median age at BC diagnosis differed significantly between patients who did not receive ET (46.   44. 5-49.5). Menopausal status at BC diagnosis did not differ between these groups. Patients with a more recent diagnosis more often received tamoxifen only, compared to patients diagnosed in earlier years. Cardiovascular risk factors at BC diagnosis were equally distributed among groups (Table 1). Almost all patients who had received ET had estrogen receptor positive and/or progesterone receptor positive BC (99.1%). The prevalence of human epidermal growth factor receptor 2 (HER2)-positive disease was unknown in 17.5% of patients, and was relatively high (40.4%) among patients who had received treatment with an AI only. Patients who had received ET in general had higher risk disease (higher TNM-stage, higher tumor grade) compared to those not treated with ET. Patients who had received ET more often underwent mastectomy (versus lumpectomy) than patients who had not received ET. Approximately 75% of patients treated with ET had also received chemotherapy, whereas only 28% of the no ET-group had received chemotherapy. The vast majority of chemotherapy regimens contained anthracyclines. Table 2 summarizes breast cancer and treatment characteristics. cIMT cIMT did not differ between AI ratio groups; median cIMT was 0.63 mm (IQR 0.56-0.71 mm) among patients with low AI exposure, 0.66 mm (IQR 0.59-0.75 mm) among patients with intermediate AI exposure and 0.64 mm (IQR 0.59-0.73 mm) among patients with high AI exposure (Table 3 and Fig. 2a). Each year increase in age at study visit was associated with an increase in cIMT of 0.01 mm (95% confidence interval (CI) 0.01 − 0.01). Overweight and obese patients had higher cIMT (0.02 mm (95% CI 0.00-0.04) and 0.04 mm (95% CI 0.02-0.06) respectively) than patients with a BMI < 25 kg/m 2 . In UMCG-patients, cIMT was 0.12 mm (95% CI 0.10-0.14) higher than in NKI-AVL-patients. Although AI ratio in itself was not significantly associated with cIMT, we did observe a significant interaction between institute and AI ratio. UMCG-patients with intermediate AI exposure had a 0.02 mm (95% CI − 0.07 − 0.03) lower cIMT than NKI-AVL-patients, and UMCG-patients with high AI exposure had a 0.05 mm (95% CI − 0.10 − 0.01) lower cIMT than NKI-AVL-patients.

Endocrine therapy exposure
Fifty patients had a cIMT above their institute specific 90th percentile cut-off. Only age at study visit (odds ratio (OR) 1.2, 95% CI 1.12-1.34) was associated with a cIMT value above the institute specific cut-off (Table 4).

AGEs
AI ratio was associated with AGEs on a continuous scale neither in univariable analysis nor after adjusting for potential confounders in multivariable analysis (  (Fig. 2b). AGEs increased by 0.01 AU (95% CI 0.01-0.02) per year increase in age at study visit, and patients with a history of endocrine disease (compared to those without endocrine disease history) and current smokers (compared to never smokers) had higher AGEs (0.15 AU (95% CI 0.01-0.30) and 0.37 AU (95% CI 0.28-0.46), respectively). In UMCG-patients, AGEs were 0.11 AU (95% CI 0.04-0.18) higher than in NKI-AVL patients. Overweight patients had lower AGEs (− 0.08 AU, 95% CI − 0.16 − 0.00) than those with a BMI < 25 kg/m 2 .
In all, 91 patients had elevated AGEs based on agespecific reference values. Patients with a history of endocrine disease (OR 2.35, 95% CI 1.10-5.03) and those who  (Table 4).

Dyslipidemia
At study visit, 195 (34.2%) patients had dyslipidemia. In univariable logistic regression analysis, women with intermediate AI exposure had higher odds of dyslipidemia at study  Table 4). For all outcomes, sensitivity analyses with absolute AI duration, a different AI ratio categorization or long AI use (≥ 5 years versus < 5 years) provided similar results (Supplementary Data).

Discussion
In this study, we investigated the association between exposure to AIs and early markers for CVD. We observed no statistically significant association between AI exposure and cIMT, AGEs or the presence of dyslipidemia; results were robust across several sensitivity analyses. Current treatment guidelines do not provide specific recommendations on AI use and CVD risk; our results do not call for a change in these guidelines [24].
Three previous studies showed cIMT results similar to those in our study. Blondeaux et al. found no significant difference in cIMT between AI users (median duration of use 53 months) and healthy controls [25]. Gallicchio et al. investigated several vascular parameters including cIMT in a small group (n = 112) of breast cancer patients and found no significant changes after 1 year of AI use [26]. An even smaller study (n = 85) also found no difference in median cIMT when comparing BC patients treated with AI (mean duration of use 34 months) to those not receiving endocrine treatment [27]. Carotid plaques, however, were seen more frequently among AI users in this study. Two other small studies suggested a detrimental effect of AI-use on endothelial function, although these effects were most pronounced in patients with additional CVD risk factors [28,29].
Several randomized controlled trials measured lipid spectrum in a subset of their trial participants. The ATENA study randomized patients to receive either 5 years of exemestane or no treatment after 5-7 years of tamoxifen and found no detrimental effect on lipid profile after two years follow-up [30]. In a Japanese substudy of the TEAM study, in which patients received exemestane, anastrozole or tamoxifen as adjuvant therapy, the lipid profile of tamoxifen users changed favorably, but in AI users, no significant effect on lipids was seen at 3 months and 1 year on treatment [31]. Atalay et al. found no detrimental effects of exemestane on cholesterol levels at 8, 24 and 48 weeks of treatment in patients with metastatic breast cancer who received either exemestane or tamoxifen as first-line therapy in a substudy of the EORTC trial 10951 [32]. To our knowledge, no other studies have evaluated the association between the use of AIs and AGEs in the skin.
Our results do not explain why several large cohort studies suggest an association between the use of AIs and a higher risk of overt CVD, such as myocardial infarction (MI) and heart failure (HF). Abdel-Qadir et al. observed a higher risk of hospitalization for MI in AI users compared to tamoxifen users in a cohort of 9350 BC patients after a mean follow-up of 3.2 years [33]. The cohort study by Khosrow-Kavar et al. included 23,525 patients and had similar results with a higher risk of HF and cardiovascular mortality in AI users (median follow-up 1.4 years) compared to   [36]. When comparing our study with the abovementioned cohorts, it should first be noted that the majority of AI users in our study had used AI as well as tamoxifen whereas AI users in the cohort studies typically had used AI only. It is possible that favorable effects of tamoxifen counterbalanced potential negative effects of AIs in our study. The fact that sensitivity analysis with absolute AI duration provided the same results, however, is reassuring. Secondly, our study mainly included pre-/perimenopausal women, in contrast to the abovementioned cohort studies which consisted of postmenopausal women only. We cannot rule out that the effect of AIs on cardiovascular risk differs across age groups or menopausal status.
Our study has several strengths compared to previous studies. First, we not only examined AI exposure as a dichotomous variable, but also quantified AI exposure by using the AI ratio. Second, we had detailed information on important potential confounders (comorbidity, smoking habits and body composition for example). Third, followup in our study (5-12 years) was notably longer than in previous studies. Fourth, we systematically collected valid subclinical CVD measures and were thus able to reliably capture early signs of cardiotoxicity.
Some limitations of our study require consideration. A first limitation is that the exact ET duration was unknown in 18% of the study population. We tried to overcome this problem by using the AI ratio. A second potential drawback of our study is the risk of survival bias. It is possible that patients who developed (severe) CVD after BC diagnosis had already died at the time of study start and were therefore underrepresented in the study population. However, prevalence of symptomatic CVD in this young cohort was low [20,21]. Although our study had longer follow-up than previous studies, 5-12 years might still be too short to develop vascular or metabolic abnormalities, in particular in a relatively young and healthy population such as ours. We are therefore planning a repeat study visit for all participants after an additional 5-8 years of followup. A last issue to consider is the fact that we recruited and assessed study participants in two different institutes, and that we observed significant effect modification of institute on the association between AI exposure and cIMT. Perhaps this interaction can be explained by differences in cardiovascular health between institutes (although baseline cardiovascular parameters did not significantly differ), but we cannot rule out that interinstitute variability in outcome measurements played a role as well.
In conclusion, our study did not show a clear association between exposure to AIs and early signs of cardiovascular damage in breast cancer survivors. Our results do not prompt a change in current clinical practice. Future studies should validate our findings over time (additional follow-up of this cohort is planned) and in different BC populations.