The Temporal Risk of Heart Failure Associated With Adjuvant Trastuzumab in Breast Cancer Patients: A Population Study

Abstract

Background:

The late cardiac effect of adjuvant trastuzumab and its potential interaction with anthracycline have not been well-studied on a population level.

Methods:

In this retrospective population-based cohort study, female breast cancer patients in Ontario, diagnosed between 2003 and 2009, were identified by the Ontario Cancer Registry and linked to administrative databases to ascertain demographics, cardiac risk factors, comorbidities, and use of adjuvant trastuzumab and other chemotherapy. Patients with pre-existing heart failure (HF) were excluded. The main endpoint was new diagnosis of HF. Analyses included Kaplan-Meier (KM) survival analysis, multivariable piecewise Cox regression, and competing risk and propensity score analyses. All statistical tests were two-sided.

Results:

Nineteen thousand seventy-four women with breast cancer treated with adjuvant chemotherapy were identified, of whom 3371 (17.7%) also received adjuvant trastuzumab. Anthracycline use was 84.9% overall. After a median follow-up of 5.9 years, patients treated with trastuzumab and chemotherapy were more likely to develop HF than patients on chemotherapy alone (5-year cumulative incidences of 5.2% vs 2.5%; log-rank P < .001). After adjusting for confounders, adjuvant trastuzumab remained independently associated with incident HF in the first 1.5 years (HR = 5.77, 95% CI = 4.38 to 7.62, P < .001), but not thereafter (HR = 0.87, 95% CI = 0.57 to 1.33, P = .53). Anthracycline use did not increase the risk of HF with trastuzumab synergistically, neither within (P_interaction = .92) nor beyond 1.5 years (P_interaction = .23).

Conclusion:

Adjuvant trastuzumab was associated with increased risk of new incidence of HF in breast cancer survivors during the period of adjuvant treatment but not thereafter. Routine intensive monitoring may not be necessary after completing adjuvant therapy.

Trastuzumab (Herceptin) is the standard-of-care treatment for human epidermal growth factor receptor 2–positive (HER2+) breast cancer (1–6), which accounts for approximately 25% of breast cancers (7,8). However, early clinical trials showed that trastuzumab produced an asymptomatic decrease in left ventricular ejection fraction (LVEF) and/or symptomatic (possibly severe) heart failure (HF) (2,3,9,10), leading to early discontinuation rates ranging from 4.3% to 19% in different trials (3,11,12,13). In the real world, the rates are likely greater (14–17).

There is substantial evidence from cohort studies and clinical trials demonstrating trastuzumab-associated cardiac dysfunction in the short term, months after therapy (2,10,12,13,15,18–23). However, there is less evidence depicting the risk of major cardiovascular adverse events in the long term, years after completion of therapy. Nearly all such evidence is derived from clinical trials, with median follow-ups ranging between two and eight years (3,9,12,24,25). The limitation of these trials is real-world applicability: Patients were excluded if they had underlying cardiac disease or LVEFs below a certain cutoff (2,10), and patients tended to be younger and received closer monitoring than what is found in the real world (19).

Additionally, the population studies to-date pay little attention to the late risks of HF; while they do portray the well-known risk of HF with trastuzumab, they fail to demonstrate whether this risk is confined to the period during or shortly after treatment or whether a substantial risk carries forward into survivorship. Long comparative follow-up studies are needed to further determine when cardiac events predominantly occur, and whether there is lingering damage that precipitates adverse events or death. A confounding consideration is that anthracycline-induced HF can manifest four to 20 years subsequent to use (26), and the annual incidences of HF continue to increase 10 to 16 years postanthracycline, potentially at progressively faster rates than other (nonanthracycline) chemotherapy regimens (27,28). Thus, there is an additional need to investigate whether trastuzumab exacerbates anthracycline-induced cardiotoxicity in a synergistic manner, which is suggested by preclinical data (29,30).

The purpose of this retrospective study is to assess the long-term risk of HF associated with the use of adjuvant trastuzumab plus chemotherapy in comparison with chemotherapy alone on a population level. A secondary purpose is to determine whether anthracycline use potentiates the risk of HF associated with trastuzumab in the general breast cancer population for patients treated by their own physicians.

Methods

Study Design and Data Collection

This was a population-based retrospective cohort study. Women with first diagnosis of breast cancer between 2003 and 2009 in Ontario, the most populous province in Canada, were identified by the Ontario Cancer Registry. Patients were linked to other Institute for Clinical Evaluative Sciences (ICES) databases to obtain patient demographics, comorbidities, cardiac risk factors, and other parameters. The databases used include the following: Registered Persons Database (RPDB) for demographic information, Ontario Cancer Registry (OCR) for cancer diagnostic details, Canadian Institute of Health Information Discharge Abstract Database (CIHI-DAD) for admission and discharge data, Ontario Health Insurance Plan (OHIP) for cardiac risk factors and comorbidities, Ontario Drug Benefits (ODB) for medication details, and the National Ambulatory Care Reporting System (NACRS) for emergency visits. Breast cancer patients in Ontario who received adjuvant chemotherapy were identified through the New Drug Funding Program (NDFP), a publicly funded drug program administered by Cancer Care Ontario (CCO). NDFP funds many chemotherapy medications, including adjuvant trastuzumab, epirubicin, paclitaxel, and docetaxel. Adjuvant trastuzumab has been funded since August 2005, and patients diagnosed in 2003 and 2004 were eligible if they completed adjuvant treatment within one year prior to August 2005. Further information on specific databases is included in Supplementary Table 1 (available online).

Patients were included in the study if they were diagnosed with breast cancer between 2003 and 2009, underwent lumpectomy or mastectomy with or without radiotherapy, and were prescribed an adjuvant chemotherapy regimen involving anthracyclines, taxanes (via NDFP), or other chemotherapy (via OHIP). Because NDFP funds adjuvant epirubicin and taxanes but not doxorubicin, doxorubicin use was determined by exclusion: a G345 OHIP claim for chemotherapy (anthracycline, taxane, or trastuzumab) in the absence of an NDFP record for epirubicin or taxanes.

Considering that NDFP funding of trastuzumab only started in August 2005, we composed a separate cohort of patients diagnosed with breast cancer between 2005 and 2009, and we repeated the statistical analysis with this cohort (Supplementary Tables 2–5 and Supplementary Figures 1–2, available online). This explores the potential for bias in selection of patients receiving treatment pre-2005.

Patients were excluded for the following conditions: diagnosis of HF prior to breast cancer (algorithm in next section), surgery performed after initiating chemotherapy, diagnosis of stage IV breast cancer, a time interval longer than 180 days between surgery and chemotherapy, and other previous cancer diagnoses (aside from breast).

Outcome and Explanatory Variables

The main outcome measured was a new diagnosis of HF. We employed an algorithm that was recently tested and validated for accurate identification of HF cases, with 85% sensitivity and 97% specificity; the criteria include one hospital admission diagnosis (CIHI-DAD) or one ambulatory care diagnosis (OHIP) followed by a second diagnosis (from either source) within one year (31). The International Classification of Diseases (ICD) diagnostic codes used to gather HF cases were the following: ICD-9 428 (“congestive heart failure”), ICD-10 150.0 (“congestive heart failure”), 150.1 (“left ventricular failure”), and 150.9 (“heart failure, unspecified”). The algorithm criteria we use to define a case of HF are similar to those previously used, and diagnostic codes we employed to identify HF are consistent with but narrower than those used in previous studies (16,17,20,21).

Various patient-level explanatory variables were collected from the ICES databases, including age, socioeconomic status, rural status, chemotherapy regimen, cardiac risk factors, cardiac comorbidities, other chemotherapy drug use, and stage of cancer at diagnosis. Socioeconomic status was determined by income quintile of neighborhood by postal code through RPDB. Comorbidities were measured using the Charlson Deyo Index (32,33). The follow-up period was up to the end of 2012.

Statistical Analyses

The baseline patient characteristics were compared between the group receiving adjuvant trastuzumab plus chemotherapy and the group receiving adjuvant chemotherapy alone. Categorical variables are presented as frequency (percentage) of nonmissing cases and compared using the chi-square test. Time-to-events was analyzed using the Kaplan-Meier (KM) method and compared by log-rank test. This time was calculated from the date of first chemotherapy administration until date of HF diagnosis, last physician contact, or death (censored)—whichever came first.

We performed multivariable Cox regression models to adjust for confounding factors and assess the effect in development of HF. Observation of preliminary data from our cumulative incidence plot demonstrated a noncongruent inflection point at approximately the 1.5-year mark. Therefore, a piecewise Cox regression model was conducted to evaluate the differential hazard ratios (HRs) for developing HF in a timeframe within 1.5 years since initiating treatment and a timeframe starting 1.5 years subsequent to initiating treatment. This was accomplished by adding a time-dependent covariate, which could demonstrate the HR of developing HF when taking trastuzumab, in the periods before and after the 1.5-year point. We evaluated the proportional hazards assumption for the time periods before and after 1.5 years.

We conducted subgroup analysis and testing for interaction between anthracycline and trastuzumab in mediating HF. After dividing our cohort into one group that used anthracyclines and another that did not, interaction between anthracycline and trastuzumab could be assessed. Interaction factors were ascertained from the regression model, both before and subsequent to the 1.5-year point. In addition, we conducted a sensitivity analysis in a cohort of patients age 65 years and older.

To check for robustness in the presence of the competing risk of death, the cumulative incidence function was also calculated in each group and an unadjusted comparison of groups was made using Fine and Gray’s method (34).

Because our comparison groups were not randomized, we performed propensity score (PS) analysis. We tested using PS as an additional covariate as well as a matched pair analysis using the greedy matching techniques (35) (details in the Supplementary Methods, available online). Linearity and interaction of covariates were assessed in the models. All statistical comparisons were two-sided with statistical significance defined at a P value of less than .05. Analyses were performed using SAS (version 9.3, SAS Institute Inc., Cary, NC).

Results

Population Characteristics

After excluding 584 women with previous HF diagnosis, a total of 19 074 women were identified as having been diagnosed with breast cancer (stage I-III) between 2003 and 2009 and received an adjuvant chemotherapy regimen. Of these women, 15 703 received chemotherapy only and 3371 (17.7%) received trastuzumab in addition to chemotherapy (Table 1). Fourteen point six percent of women were age 65 years or older. There were no statistically significant differences between the two groups with respect to comorbidities, cardiovascular risk factors, income quintile, and rural status. Anthracycline use was overall 84.9%, slightly more common in the chemotherapy-alone group. The two groups displayed statistically significant difference in diagnostic stage of breast cancer, in that the group receiving trastuzumab was more likely to have stage I and III cancer, but less of stage II.

Table 1.

Demographic and clinical characteristics of breast cancer patients according to adjuvant treatment

Characteristics	Total No. (%)	Chemotherapy alone No. (%)	Trastuzumab and chemotherapy No. (%)	P
N	19 074	15 703 (83.3)	3371 (17.7)
Median age, y	52	52	53
Age group, y
<45	4222 (22.1)	3467 (22.1)	755 (22.4)	.021
45–64	12 070 (63.3)	9994 (63.6)	2076 (61.6)
>64	2782 (14.6)	2242 (14.3)	540 (16.0)
Two or more comorbidities*	803 (4.2)	659 (4.2)	144 (4.3)	.84
Income quintile
Q1	3029 (15.9)	2465 (15.8)	564 (16.8)	.57
Q2	3541 (18.6)	2916 (18.6)	625 (18.6)
Q3	3863 (20.3)	3196 (20.4)	667 (19.9)
Q4	4191 (22.0)	3469 (22.2)	722 (21.5)
Q5	4385 (23.1)	3604 (23.0)	781 (23.3)
Rural	2345 (12.3)	1927 (12.3)	418 (12.4)	.84
Prior diabetes mellitus	1832 (9.6)	1520 (9.7)	312 (9.3)	.45
Prior hypertension	5601 (29.4)	4600 (29.3)	1001 (29.7)	.64
Prior AMI	77 (0.4)	65 (0.4)	12 (0.4)	.63
Cancer stage
I	2963 (22.2)	2273 (21.4)	690 (25.1)	<.001
II	7606 (57.0)	6174 (58.2)	1432 (52.1)
III	2783 (20.8)	2158 (20.3)	625 (22.8)
Anthracycline use	16 201 (84.9)	13 467 (85.8)	2734 (81.1)	<.001
Taxane use	10 686 (56.0)	8229 (52.4)	2457 (72.9)	<.001
Year of breast cancer diagnosis
2003	2384 (12.5)	2362 (15.0)	22 (0.7)	<.001
2004	2592 (13.6)	2306 (14.7)	286 (8.5)
2005	2749 (14.4)	2171 (13.8)	578 (17.1)
2006	2745 (14.4)	2110 (13.4)	635 (18.8)
2007	2856 (15.0)	2207 (14.1)	649 (19.3)
2008	2836 (14.9)	2261 (14.4)	575 (17.1)
2009	2912 (15.3)	2286 (14.6)	626 (18.6)

* Based on Charlson score. AMI = acute myocardial infarction.

Cardiac Events

The median follow-up time for the breast cancer survivors was 5.9 years (interquartile range = 4.2–7.6 years). Over the 10-year period of follow-up, the cumulative incidence of HF was statistically significantly higher in the group receiving adjuvant trastuzumab with chemotherapy compared with chemotherapy alone (Figure 1). According to KM estimates, the one-year incidence of HF was 3.0% in the trastuzumab group and 0.8% in the standard chemotherapy group and the five-year incidence was 5.3% vs 2.6% (P < .001, log-rank) (Table 2). As the HF incidences increased over time, the annual incidences were similar between groups from two to five years (Table 2). Competing risk analysis demonstrated similar results, with five-year cumulative incidences of 5.2% vs 2.5%, log-rank P < .001 (Table 2).

Table 2.

Cumulative incidences of HF over time since initiating chemotherapy +/- trastuzumab*

Analysis	Cumulative incidence of HF after:	Chemotherapy alone, %	Trastuzumab and chemotherapy, %	Absolute difference
Kaplan-Meier estimates	1 y	0.8	3.0	2.2
	1.5 y	1.3	4.7	3.4
	2 y	1.3	4.7	3.4
	3 y	1.7	4.9	3.2
	4 y	2.1	5.2	3.1
	5 y	2.6	5.3	2.7
Cumulative incidence of competing risks	1 y	0.7	3.0	2.3
	1.5 y	1.0	4.3	3.3
	2 y	1.2	4.7	3.5
	3 y	1.6	4.8	3.2
	4 y	2.0	5.1	3.1
	5 y	2.5	5.2	2.7

* Incidences were calculated by Kaplan-Meier estimates as well as by competing risk analysis for comparison. Time ‘zero’ is the start of adjuvant chemotherapy. Absolute difference was calculated by subtracting the incidence in the ‘Chemotherapy alone’ group from the incidence in the ‘Trastuzumab and chemotherapy’ group. HF = heart failure.

Figure 1.

Kaplan-Meier cumulative incidence of heart failure over time of adjuvant therapy. The numbers of patients at risk for each year are below the figure. Curves were derived from Kaplan-Meier survival estimates. When cumulative incidence of competing risks was graphed, Gray’s test P value was < .001 as well (see Table 2 for values).

The cumulative incidence curve for the trastuzumab group depicted an inflection point at approximately the 1.5-year mark, where the slope decreases sharply (Figure 1). This difference was quantified by a piecewise adjusted Cox regression model (Table 3), which revealed an adjusted hazard ratio of 5.77 (95% confidence interval [CI] = 4.38 to 7.62, P < .001) of experiencing a new onset of HF in the first 1.5 years of chemotherapy compared with conventional chemotherapy. However, after 1.5 years the risks of new incidence of HF were similar, with an adjusted hazard ratio of 0.87 (95% CI = 0.57 to 1.33, P = .53). Additionally, these two hazard ratios are statistically different (P < .001). Hazards became nonproportional towards the end of the first 1.5 years.

Table 3.

Adjusted HRs for development of HF*

Parameter	HR (95% CI)	P
Chemotherapy regimen
Chemotherapy alone	Ref
Trastuzumab plus chemo, within 1.5 y†	5.77 (4.38 to 7.62)	<.001
Trastuzumab plus chemo, after 1.5 y†	0.87 (0.57 to 1.33)	.53
Age, y (per 5-y increase)	1.28 (1.22 to 1.36)	<.001
Two or more comorbidities	1.76 (1.22 to 2.53)	.003
Income quintile
Q1	Ref
Q2	0.72 (0.52 to 0.99)	.045
Q3	0.71 (0.51 to 0.98)	.040
Q4	0.75 (0.54 to 1.03)	.074
Q5	0.80 (0.59 to 1.09)	.16
Rural status	0.77 (0.55 to 1.08)	.12
Diabetes mellitus	1.47 (1.11 to 1.95)	.008
Hypertension	1.03 (0.82 to 1.31)	.80
Previous myocardial infarction	0.51 (0.07 to 3.69)	.51
Anthracycline use	1.22 (0.89 to 1.67)	.21
Taxane use	0.70 (0.55 to 0.89)	<.001
Cancer stage
I	Ref
II	1.38 (1.04 to 1.83)	.027
III	1.65 (1.20 to 2.28)	.002
Year of diagnosis (per 1-y increase)	0.88 (0.81 to 0.94)	<.001

* Hazard ratios were adjusted for variables appearing in the table. CI = confidence interval; HF = heart failure; HR = hazard ratio.

† HRs provided for risk of HF within 1.5 years since initiating adjuvant chemotherapy, as well as subsequent to the first 1.5 years since the start of adjuvant chemotherapy. HRs within and after 1.5 years are statistically significantly different (P < .001). All statistical tests were two-sided.

PS analysis revealed similar results: The hazard ratio for risk of HF within 1.5 years was 5.83 (95% CI = 4.42 to 7.69, P < .001) and after 1.5 years was 0.88 (95% CI = 0.58 to 1.35, P = .55). By matched pair analysis, the values were similar both before 1.5 years (HR = 6.17, 95% CI = 3.85 to 9.90, P = .006) and after 1.5 years (HR = 0.94, 95% CI = 0.55 to 1.61, P = .83).

Aside from trastuzumab use, various explanatory variables were associated with statistically significantly increased HF risks (Table 3). They include increased age, underlying diabetes mellitus, two or more comorbidities, and having a stage II or III cancer. Anthracycline use was associated with HF nonsignificantly. More recent year of diagnosis and taxane use were associated with lower risk of HF.

Auxiliary Analyses

Subgroup analysis by use or nonuse of anthracyclines demonstrated that trastuzumab use was associated with statistically significantly higher cumulative incidences of HF in either case (Figure 2), with cumulative incidences slightly higher in nonanthracycline patients amongst those who did not use trastuzumab, but slightly higher in anthracycline patients amongst those who did use trastuzumab (Table 4). Interaction analysis in the Cox regression model revealed that use of anthracyclines did not increase the risk of HF associated with trastuzumab, neither within (P = .92) nor beyond 1.5 years (P = .23).

Table 4.

Kaplan-Meier cumulative incidences of HF over time since initiating chemotherapy +/- trastuzumab, segregated by anthracycline use*

Anthracycline use	Cumulative incidence of HF after:	Chemotherapy alone, %	Trastuzumab and chemotherapy, %	Absolute difference
Yes	1 y	0.7	3.0	2.3
	1.5 y	0.9	4.3	3.4
	2 y	1.2	4.8	3.6
	3 y	1.6	5.0	3.4
	4 y	2.0	5.2	3.2
	5 y	2.5	5.3	2.8
No	1 y	1.0	3.0	2.0
	1.5 y	1.3	3.9	2.6
	2 y	1.5	4.3	2.8
	3 y	1.9	4.4	2.5
	4 y	2.6	4.9	2.3
	5 y	3.2	5.1	1.9

* HF = heart failure.

Figure 2.

Kaplan-Meier cumulative incidence of heart failure divided into cohorts according to trastuzumab and anthracycline use. Sixteen thousand two hundred and one patients (84.9%) used anthracyclines, while 2873 (15.1%) used other chemotherapies. The numbers of patients at risk for the end point at each given year are included in the table.

A sensitivity analysis was conducted on a cohort age 65 years and older. By KM estimates, the five-year cumulative incidence of HF was 8.5% in the trastuzumab group and 5.9% in the chemotherapy-only group (Supplementary Figure 3, available online). In the adjusted Cox model, the risk of HF had a hazard ratio 5.34 (95% CI = 3.11 to 9.17, P = .014) within 1.5 years and a hazard ratio of 0.71 (95% CI = 0.32 to 1.60, P = .41) after 1.5 years.

Results were similar in an analysis limited to patients diagnosed in 2005 or later, when trastuzumab was available in Canada (Supplementary Tables 2–5 and Supplementary Figures 1–2, available online).

Discussion

This study retrospectively followed breast cancer patients for a median of 5.9 years and demonstrated an increased cumulative incidence of HF in patients receiving trastuzumab with adjuvant chemotherapy compared with conventional chemotherapy alone. The risk of new incidence of HF was 5.77-fold higher in patients who received trastuzumab within 1.5 years from the start of adjuvant therapy, but no difference in risk was observed subsequently. No interaction between trastuzumab and anthracycline was detected. To our knowledge, this is the largest population-level analysis of long-term trastuzumab cardiotoxicity.

This study demonstrated one-, three-, and five-year risks of HF of 3.0%, 4.8%, and 5.2%, respectively, for patients treated with trastuzumab plus chemotherapy. This compares with respective risks of 0.7%, 1.6%, and 2.5% risks with conventional chemotherapy alone. These cumulative incidence values are slightly higher, yet consistent with previous findings from clinical trials: The B-31 and the N9831 trials reported three-year cardiac event incidences of 4.1% and 2.9%, respectively, in the trastuzumab group, in comparison with 0.8% and 0% in the control groups (2,13). Similarly, a median 3.6-year follow-up of the HERA trial featured a symptomatic HF incidence of 1.9% in patients receiving trastuzumab, compared with 0.1% in the control group (9). The discrepancy can be explained by exclusion of patients from these clinical trials who had underlying heart disease or displayed cardiac symptoms or low LVEF after anthracycline treatment, before beginning taxanes +/-trastuzumab (2,10), producing cohorts with younger age distributions and fewer comorbidities than seen in the general population. Additionally, the time delay before adjuvant trastuzumab in the HERA trial was longer than seen in practice.

Conversely, the incidence values from our elderly cohort sensitivity analysis (8.5% vs 5.9% in respective groups) are lower than those from studies involving elderly patients from the SEER database (16,17,21), including one study which featured a 41.9% three-year incidence of HF in 431 patients receiving trastuzumab plus chemotherapy, compared with 20.2% in 5257 patients receiving anthracyclines alone (16). A retrospective cohort study involving breast cancer patients of all ages gathered from the Cancer Research Network demonstrated a 20.1% five-year incidence of “HF and/or cardiomyopathy” in 442 patients receiving trastuzumab and anthracyclines, compared with 4.3% in 3697 patients receiving anthracyclines only (20). The increased values observed in both these studies might be explained by the inclusivity and imprecision of the outcome variable “HF or cardiomyopathy” (acknowledged as a limitation by both studies), compared with our more stringent algorithm for “HF” diagnosis. Use of different administrative databases might also explain discrepancies.

Adjustment for confounding variables enabled determination of hazard ratios for development of HF. The risk of developing new HF within 1.5 years of initiating adjuvant chemotherapy was almost six-fold higher when taking trastuzumab plus chemotherapy compared with chemotherapy alone. However, for women who did not develop HF in the first 1.5 years, the risk of new incidence of HF subsequent to this time was not statistically different between groups. Given that the duration of adjuvant therapy is roughly 1.5 years, the results imply that the trastuzumab-associated risk of HF does not further increase beyond the approximate period of adjuvant treatment.

This striking temporal trend is consistent with previous analyses of randomized controlled trial data. In a seven-year follow-up of the B-31 trial, only two of 37 cardiac events occurred after two years among 944 patients who received trastuzumab plus chemotherapy (2). A median follow-up of eight years of the HERA trial demonstrated that the cumulative incidence of any type of cardiac end point increased during the period of trastuzumab treatment but was comparable with control patients thereafter (9,11,25). The aforementioned retrospective population-level cohort studies did not compare early vs late cardiac risk associated with trastuzumab (16,17,20,21). Overall, consistent evidence exists in the literature, but the long-term temporal trend of trastuzumab cardiotoxicity has not been quantitatively explored previously. This may be partially because of the absence of long-term follow-up data on trastuzumab patients until recently, as the drug has only been in public use for approximately 10 years.

As expected, increased age, underlying diabetes mellitus, and two or more comorbidities were associated with risk of HF. Explanation of the increased risk of HF with a stage II or III cancer might be justified by the greater leniency in dose and duration of chemotherapy in patients with more aggressive, life-threatening cancer. The reduced risk of HF with more recent year of diagnosis may be related to an increasing familiarity with cardiac monitoring and earlier interruption of trastuzumab prior to development of clinical HF (36). In contrast to other studies, anthracycline use was not associated with a statistically significant HF risk. This can be explained by the markedly lower baseline incidence of cardiac risk factors of patients prescribed anthracyclines in our study (Supplementary Table 6, available online), as is commonly observed (28).

In this study, we did not find any synergistic interaction between the clinical cardiotoxic effects of anthracyclines and trastuzumab within or beyond 1.5 years. The original phase III clinical trial (1) involving patients with metastatic breast cancer demonstrated cardiac dysfunction in 27% of patients taking trastuzumab plus anthracyclines, 8% anthracyclines alone, 13% trastuzumab plus paclitaxel, and 1% paclitaxel alone (37). However, interaction cannot be inferred because baseline cardiac screening and exclusion of high-risk patients were not performed in this early study, and simultaneous trastuzumab and anthracycline is no longer done clinically. The later trastuzumab clinical trials were also not insightful because they did not include all necessary treatment arms. One retrospective study with elderly patients analyzed HF risks in many treatment arms, and they did not observe any anthracycline-trastuzumab interaction (21). Despite the evidence refuting an interaction, the proposed molecular mechanisms of action of these drugs might suggest otherwise. Anthracyclines cause oxygen radical-mediated irreversible cell damage, whereas trastuzumab may block HER2 signaling pathways to disrupt cardiomyocyte contractility, block survival signals, and impair response to stressors (such as anthracyclines) reversibly (19,30). The latter mechanism presents a possibility for a trastuzumab-anthracycline interaction, whereby trastuzumab exacerbates anthracycline-induced cardiotoxicity, but this may be confined to concurrent drug administration, which is not done in practice (30). Further elucidation of the mechanism of action of trastuzumab is needed to clarify this issue.

There are several limitations to this study. We evaluated clinical HF exclusively, not other trastuzumab-related cardiac end points such as HF severity, LVEF, cardiomyopathies, ischemic heart disease, or cardiac-related death. HF is clinically important and more reliably measured retrospectively, while the other end points are not regularly obtainable from administrative databases, but some were explored previously in a small sample size (21). Depending on the comprehensiveness of administrative databases, future studies might investigate these end points in a long-term population-level design. Another limitation was our 85% sensitivity for detecting HF, but we are confident that this does not bias results. Additionally, various potential modifiers of HF risk could not be ascertained. Among adjuvant regimens of trastuzumab and/or chemotherapy, different doses, frequency of administration, and premature cessation were not identified, nor were use of radiation therapy (likely inconsequential [13,19,24]), cardio-protective medications, smoking, family history, body mass index, or baseline LVEF. Exclusion of these factors likely does not compromise the major implications of the study, but may affect the cumulative incidence and HR values. Another limitation is that interaction terms often lack power; our study may not completely rule out a clinically important synergy of anthracycline and trastuzumab on cardiotoxicity. Finally, this study demonstrates associations only and not causality.

This study has several strengths. It included the largest number of breast cancer patients receiving adjuvant trastuzumab in “real world” clinical practice, to our knowledge. Three thousand three hundred seventy-one patients represented 17.7% of the entire cohort; this resembles the expected percentage of breast cancers that are HER2+ (approximately 25%) (7,8). Second, the relatively long duration of patient follow-up, median of 5.9 years, enabled assessment of the late manifestations and temporal pattern of cardiotoxicity. Third, this study design provides a broad population-level representation of the long-term trastuzumab-associated HF risk. Most other studies on this topic are clinical trials or retrospective studies involving elderly patients, which are limited in their generalizability.

Our findings may have implications for clinical practice. No clear universal cardiac monitoring guideline exists for patients receiving adjuvant trastuzumab. Various guidelines have been proposed but are largely based on expert opinion and not validated (38–42); they involve monitoring every three, four, or six months over the 12 months of treatment, and possibly beyond. For instance, the Canadian Trastuzumab Working Group recommends that carefully selected patients receive a clinical and LVEF assessment at baseline and every three months during treatment (38). Yet, the need for post-treatment monitoring is also disagreed upon in the literature (38,39,41). This study suggests that routine intensive monitoring for HF is not needed after completion of trastuzumab because the risk of developing new HF after completion of trastuzumab is no different from taking anthracyclines alone.

In conclusion, use of adjuvant trastuzumab with chemotherapy is associated with a statistically significantly increased risk of HF compared with chemotherapy alone, with the heightened risk largely confined to the duration of adjuvant therapy. The late risk of HF after adjuvant trastuzumab is not statistically significant, and routine intensive monitoring may not be necessary for patients who did not develop HF during adjuvant treatment.

Funding

This work is supported by the Canadian Cancer Society (grant # 700839). This research was supported through provision of data by the Institute for Clinical Evaluative Sciences (ICES) and Cancer Care Ontario (CCO) and through funding support to ICES from an annual grant by the Ministry of Health and Long-Term Care (MOHLTC) and the Ontario Institute for Cancer Research (OICR). The opinions, results, and conclusions reported in this paper are those of the authors and are independent from the funding sources. No endorsement by ICES, CCO, OICR, or the Government of Ontario is intended or should be inferred.

This study was orally presented at the 50^th ASCO Annual Meeting (Abstract ID 9504) on June 2, 2014.