Participants

Recruitment was from 12 centres (see Acknowledgments) through the UK’s National Cancer Research Network (NCRN). This research forms part of a prospective gene-environment interaction study seeking the association of gene variants with cardiotoxicity (results of which will soon be submitted for publication). The power of such studies relies on cohort homogeneity, which amplifies the relative effect of remaining variables (genetic and non-genetic) [9]. Susceptibility to cAC is influenced by gender, race, age, cardiovascular disease and risk factors, cardiac medications, and anticancer regimen [1, 10]. Entry criteria (Table 1) balanced the desired homogeneity against feasibility of recruitment. Eligible were anthracycline naïve women aged >18 years without pre-existing cardiac disease, and with planned anthracycline chemotherapy for early breast cancer. Excluded were those of non-European ethnicity, or with potentially confounding comorbidities such as diagnosed hypertension, diabetes, BMI ≥35 kg/m² and renal impairment. For practical and ethical reasons, eligible women attending for cardiovascular magnetic resonance (CMR) continued in the study, even were confounding factors later recognised. Treatment regimens were determined by the attending clinicians at the recruiting centres, uninfluenced by study participation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Eligibility Criteria.

https://doi.org/10.1371/journal.pone.0165262.t001

Study Size and Timelines

Subjects were recruited between June 2005 and May 2009. A target of 276 subjects was based upon the requirements for study scale of the funding body and experience-informed power estimates for genetic study [9, 11]. Recruitment was slower than anticipated, and the study thus closed with 196 patients enrolled.

Cardiovascular Magnetic Resonance (CMR)

CMR was undertaken pre- and ≥12 months post-anthracycline chemotherapy (≥3 months after completing Trastuzumab). Scans were performed on 1.5 Tesla scanners (Siemens Medical Systems, Erlangen, Germany) at 3 centres (see acknowledgments). Cine images were acquired, for assessment of left ventricular ejection fraction and mass, using a steady state free precession sequence with retrospective ECG gating [12]. Late gadolinium-enhanced images were obtained at baseline and routinely at follow-up until October 2008 (following interim review of results). Gadolinium diethylenetriamine penta-acetic acid (Gd-DPTA, Magnevist, Schering) was administered as a 0.1mmol/Kg bolus dose via a peripheral cannula. Contrast-enhanced images were acquired 10 minutes post-injection using an inversion-recovery segmented gradient echo (Turbo-Flash) sequence[13]. Analysis of paired CMR scans (images side-by-side) was performed by a single investigator (PK, who was blind to identity, clinical, temporal and research data) using semi-automated PC-based software (CMRtools Cardiovascular Imaging Solutions, London, UK) [12, 13]

The primary outcome measure was change in left ventricular ejection fraction (LVEF), an absolute fall ≥5% defining (a priori) the ‘subclinical cardiotoxicity’ group. All other participants comprised the ‘minimally/unaffected control’ group. We defined overt anthracycline cardiotoxicity as the diagnosis of heart-failure by the clinical team, or a subclinical fall in LVEF ≥10% to below normal using age-gender-CMR specific reference ranges [12]. This being more appropriate than using 53% (the recommended lower limit of normal using 2D echocardiography[14]) because normal values differ between imaging modalities[15]. This issue is recognised in the 2014 expert consensus report on multimodality imaging in adult cancer patients, which also endorses CMR as the ‘reference standard’ for measuring LVEF[14].

Clinical Data

The data collection protocol is outlined in Fig 1. Baseline demographic and anthropometric data were recorded. Seated blood pressure was measured using automated sphygmomanometers, after 5 minutes rest. Baseline blood pressure (BP) was the mean of measurements at enrolment (oncology unit, 1 or 2 averaged readings) and CMR (1 reading). Percentage body fat was measured at the CMR Unit (BC531 Fitness Bioimpedence Innerscan Body Composition monitor, Tanita Corporation, Tokyo, Japan); equipment failure meant this was only recorded in a proportion of subjects. Body mass index (BMI: weight in kg/ (height in metres)²) and body surface area (BSA by Dubois formula: 0.007184 × (height in cm)^0.725 × (weight in kg)^0.425)) were calculated. Estimated glomerular filtration rate (eGFR) was calculated using the modified diet in renal disease (MDRD) formula [16]

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Summary of study design.

https://doi.org/10.1371/journal.pone.0165262.g001

Statistical Analysis

Analysis was undertaken using Stata version 12 (StataCorp Texas). Differences in mean values were assessed using unpaired t-tests or with the Mann-Whitney U test for non-normally distributed data. Categorical data were compared using Chi Squared and Fisher’s exact tests. Best discriminator values were defined by receiver operator curve analysis. Odds ratios and p-values were obtained from logistic regression models; where numbers were small, exact logistic regression was used. To determine independent predictors of cardiotoxicity, variables were selected using stepwise multiple logistic regression with backwards selection, removing terms with p ≥0.1 and adding those with p <0.05. Patients with missing data were excluded. The models were validated with bootstrap re-sampling; all terms retained in the model were selected in at least 65% of the bootstrap samples.

Observed anthracycline cardiotoxicity

CMR is the gold standard for serial measurement of LVEF, being superior to 2-dimensional echocardiography (better inter-study reproducibility allowing a 5% fall in LVEF to be reliably detected, while increasing study power) and radionuclide ventriculography (detection of structural heart disease and avoidance of ionising radiation) [11, 14, 17]. CMR studies have shown that LVEF normally rises by 0.1% per year, because of age-related cardiac remodelling [12, 18]. This relationship was observed in our cohort pre-chemotherapy, LVEF being 0.14% higher per year of age (linear regression, p<0.001). However, following treatment, LVEF deviated from this trajectory, instead falling by a mean of 2.1% over an average period of 21.7 months (p<0.0001). The majority of this cardiotoxicity was accounted for by the cAC group (mean fall 7.7%), while the ‘minimally/unaffected’ control group suffered no substantial impact (mean fall 0.6%).

The fact that no patients in our cohort developed overt anthracycline cardiotoxicity likely reflects the very low prevalence of comorbidities in the cohort, their relative youth, and the fact that the epirubicin doses used in early breast cancer are well below the 900mg/m² empiric dose cap.

Of note, it is not standard UK practice to routinely undertake cardiac imaging in patients receiving adjuvant anthracycline chemotherapy. Thus, no such data are available to augment those which we present. The 31 patients who received sequential Trastuzumab did have cardiac imaging (a mixture of non-standardised echocardiography and radionuclide ventriculography) at their different centres before and during Trastuzumab treatment (see below). Scanning during Trastuzumab therapy primarily detects type II (reversible) cardiotoxicity, not the type I cardiotoxicity that was the focus of our study.

Influence of cumulative anthracycline dose and time from treatment on cardiotoxicity risk

The identification of the characteristic cumulative dose-dependency of cAC (OR 2.21 [1.01–4.82] at ≥450mg/m² epirubicin) confirms the discriminative power of the study model. Although retention of number of anthracycline cycles on multivariate analysis (rather than cumulative dose) might signal that additional insults and/or the total duration of injury also contribute to cAC risk, the relationship was primarily driven by cumulative dose (Spearman correlation, r = 0.90, p<0.001). Evidence for the progressive nature of cAC was also observed, the OR for cardiotoxicity in those who did not receive Trastuzumab being 1.06 [1.00–1.12] per month post-treatment. The dataset and the spread of follow-up are, however, insufficient to accurately project the long-term course of decline.

Influence of anthropometric factors on cardiotoxicity risk

Increasing height, weight and BSA were all associated with increasing cAC risk on univariate analysis. On multivariate analysis, BSA was the best discriminator amongst the anthropometric factors when considered as a continuous variable (OR for cardiotoxicity of 2.08 [1.36–3.20] per standard deviation (0.16m²) increase). We are the first to report this association, which may be of clinical relevance suggesting as it does that BSA-adjusted dosing of anthracyclines (accepted standard practice) may increase cardiotoxicity risk. Such an effect is biologically plausible, were cardiomyocytes exposed to higher peak and/or cumulative anthracycline doses. Such a possibility has previously been suggested [19], while historically (unproven) concerns about increased risk of toxicity led some oncologists to apply an empirical dose cap at a BSA of 2m² [20]. Our observation that BSA >2.00m² was associated with subclinical cardiotoxicity (OR 3.59 [1.02–12.58] on univariate analysis) is thus noteworthy. No alternative explanation to BSA-indexing is apparent for the observed association between cAC risk and increasing height. Meanwhile, higher weight and obesity (which have previously been associated with cardiotoxicity[21, 22] might have more direct influence on cardiotoxicity risk given that body composition (specifically the proportion of lean and adipose tissues) affects the pharmacokinetics/dynamics of anticancer drugs [20, 23, 24]. In addition, non-pharmacokinetic mechanisms might also play a role, given that being overweight or obese are risk factors for developing heart failure [25]. Although BMI was not associated with cardiotoxicity in our study, it is a poor predictor of body composition [26], and it is noteworthy in this regard that those with body fat ≥31% were more likely to be in the cAC group (p = 0.02). In addition, the selected nature of our cohort, which included very few women with BMI >35kg/m², may have biased against a signal. Nonetheless, the relationship between body morphology and composition with anthracycline/Trastuzumab cardiotoxicity has become a matter of increasing interest, with a detailed systematic review and meta-analysis being only recently published [27]. In keeping with our findings, this concludes that ‘overweight and obesity are risk factors for cardiotoxicity from anthracyclines and sequential anthracyclines and trastuzumab’. Such associations are postulated to be driven in part by reduced levels of the adipocytokine (adipocyte-derived hormone) adiponectin in obese patients: adiponectin-knockout mice appear more susceptible to doxorubicin-induced cardiac contractile dysfunction, whilst exogenously-administered adiponectin protects such mice, and enhances cardiac function in wild-type mice treated with doxorubicin (reviewed in [27]).

The American Society of Clinical Oncology’s (ASCO’s) 2012 practice guideline endorses BSA-adjusted dosing of anthracyclines, and advise against empiric dose capping or using ideal body weight for calculations in obese patients [20]. It should be noted, however, that a paucity of toxicity and efficacy data means that these recommendations are not underpinned by a strong evidence base, but rather by concerns that alternative methods result in under-dosing and worse cancer outcomes [20]. Our observation of associations between anthropometric factors and cardiotoxicity, in a mechanistic study, do not change that assessment, but they do warrant further investigation and underscore the need for additional research into optimal dosing strategies.

Influence of elevated blood pressure on cardiotoxicity risk

Hypertension was first suggested to increase cardiac susceptibility to anthracyclines in the 1970s [28]. However, there have been conflicting reports [22, 29] and it is only in the past decade that studies using large datasets from the Surveillance, Epidemiology and End Results (SEERS)-Medicare database have established hypertension as an independent risk factor, hazard ratios of 1.8 and 1.5 being reported in treatment of lymphoma [10] and early breast cancer [30] respectively. Despite this, interpretation of the literature remains complicated because few data, if any, fully discriminate the contribution of hypertension per se to cardiotoxicity, from the influence of associated comorbidities (including coronary disease, atrial fibrillation, diabetes, renal dysfunction, obesity [10, 29, 31, 32]) and cardiovascular medications (such as ACE inhibitors, mineralocorticoid receptor antagonists and beta-blockers) which may be cardioprotective [33, 34]. Our selected cohort allowed prospective study of the influence of blood pressure, largely unhindered by these considerations. Women with elevated BP (≥140/90mmHg) pre-chemotherapy were substantially more likely (OR 5.36 [1.73–17.61]) to be in the cAC group, despite mean blood pressure (147.3/86.1mmHg) being only modestly above normal. Although, confirmed clinic measurements ≥140/90mmHg define hypertension [35], blood pressure is a continuous rather than categorical variable. Increasing values from within the normal range correlate with risk of both vascular mortality [36] and heart failure [37]. It thus seems likely that blood pressure also acts as continuous variable in respect of cAC risk, albeit that our dataset was too small to be able to investigate this effect. Our results, taken with the wider published literature, suggest that close attention be paid to blood pressure measurement before (and after) chemotherapy, with early treatment initiated when hypertension is confirmed–an approach advocated by expert panels of the European Societies of Cardiology and Medical Oncology–albeit that benefits remain to be proven [1, 6].

Trastuzumab and cardiotoxicity

We sought clinical features which might contribute to risk of anthracycline-induced (Type 1, permanent) cardiotoxicity. Trastuzumab, whilst not the primary focus of attention, can itself cause cardiotoxicity by both potentiating anthracycline-induced (type I) cardiotoxicity and, more commonly, by causing reversible falls in cardiac contractility (type II cardiotoxicity) [3]. In trials of adjuvant Trastuzumab, LVEF was depressed by ≥10 or ≥15% (depending on study) in 3–16%, while cardiotoxicity led to treatment discontinuation in 4.3–15.6% [1, 6]. Non-trial data suggest an incidence of heart failure/cardiomyopathy as high as 20% 3–5 years after adjuvant Trastuzumab[31].

In our study, all 31 patients commencing Trastuzumab completed the full 12-month course, with one (3.2%) identified as experiencing a transient fall in LVEF (type II cardiotoxicity) on routine scanning by clinical teams. None exhibited overt anthracycline cardiotoxicity at follow-up CMR. Trastuzumab was identified as a risk factor for the subclinical cAC group on multivariate analysis, albeit that the magnitude of associated risk (OR 3.35 [1.18–9.51]) also reflected the longer interval to follow-up. This relatively muted effect of Trastuzumab is likely a consequence of the low prevalence of comorbidities which are believed to underlie the frequent cardiotoxicity observed in registries [31]. The interval between completing anthracycline and commencing Trastuzumab therapy (median of 3.7 months in our study) is also believed to be important, cardiotoxicity rates being highest when these drugs are co-administered, rather than sequentially administered as was the case for our cohort [6].

Age, radiotherapy and cardiotoxicity risk

The reported association between older age and cAC [10, 32] was absent in our cohort, likely due to its relative youth (only 0.13% were ≥65 years old), and possibly the low prevalence of age-related comorbidity. The use of radiotherapy was also not associated with a fall in LVEF, consistent with other contemporary reports and probably reflecting the low cardiac-doses received with modern breast cancer treatment [30, 31, 38].

Risk prediction modelling

Risk prediction models for cAC might help guide individual management [3, 8]. Although models have been proposed for cAC in metastatic breast cancer and for Trastuzumab cardiotoxicity [21, 31, 32], these are not in widespread clinical use. Meanwhile, none yet exist for cAC prediction in patients with early breast cancer. In our study, multivariate analysis revealed the number of anthracycline cycles, body surface area, BP ≥140/90mmHg and Trastuzumab treatment to be the independent discriminators of being in the cAC group. The model based on these variables had an area under the ROC curve of 0.78. Predictive tests with an area of >0.7 are generally considered to offer fair discriminatory ability. However, even within our highly selected cohort, the model had a limited positive predictive value (PPV) for cardiotoxicity (62.5% at a 5% false positive rate). To inform therapeutic strategy, models would need to perform significantly better. Despite this, a basic predictive model such as ours might have utility in stratifying patients post-chemotherapy where ongoing cardiac surveillance is considered [8]. While such surveillance is not routine practice (or validated for cost-efficacy), such follow-up has been suggested by both European and American experts [1, 3].

Subclinical cardiotoxicity and its potential clinical relevance

Given that the contractile reserves of the heart are high, significant cardiotoxicity can be present before LVEF falls by as much as 10% to below normal[5]; furthermore, once such dysfunction has developed it may not be fully reversible [14]. This has led to a search for techniques to allow detection of cardiotoxicity at an earlier stage.

We sought subclinical cardiotoxicity within a selected cohort as a means to identify risk factors and explore disease mechanism. Overt cardiotoxicity is an unsuitable endpoint for our purposes, given scarcity of early dramatic declines, that would have demanded a study of massive scale, and inclusion of more diverse population. The definition of ‘subclinical cardiotoxicity’ has no consensus. However, this does not impact on the validity of our findings as the threshold set will depend on the power to reliably detect cardiotoxicity-related-changes, and thus vary by the phenotype measured (be that LVEF, tissue strain, or biomarkers) and the modality used to do so. We used a fall in LVEF≥5% on CMR to define our subclinical cardiotoxicity group. Although this cut-off is below the threshold recommended for use in clinical practice and in oncology trials (i.e. falls in LVEF of at least 10% to below normal) [1, 14], that threshold reflects the need to diagnose cardiotoxicity in individuals of sufficient severity that interruption of anticancer treatment is warranted–an issue entirely different from that which we addressed. Further, a 10% fall in LVEF is at the limit of what can be reliably detected using 2D echocardiography, the most widely available imaging modality [14]. Our use of CMR (the ‘reference standard’ for LVEF evaluation[14]), allowed us to reliably detect smaller changes in LVEF, given its superior inter-study reproducibility [11, 17]. Meanwhile, others have suggested a fall in LVEF≥5% be used to define anthracycline cardiotoxicity[39]–that report being based on multi-gated acquisition (MUGA), which has high inter-study reproducibility, comparable to CMR (and superior to 2D echocardiography) [14, 15, 39]).

By showing associations with several established cardiotoxicity risk factors, the results from our study validate a fall in LVEF ≥5% as a definition of subclinical cardiotoxicity. Furthermore, we have also validated a study-model which can be applied to larger and more disparate cohorts. Future studies of similar design might be used to study the influence of other factors on cAC risk (such as obesity, physical activity or blood pressure as a continuous variable), or to investigate the efficacy of potentially cardioprotective drugs.

In the short and medium term, changes in left ventricular ejection fraction of the magnitude we observed are not of clinical importance and the benefits of anthracycline/Trastuzumab therapy undoubtedly outweigh the risks in patients such as ours. However, whether and to what extent therapy might impact on cardiac health in the true long-term (potentially several decades hence) is uncertain. In the most comprehensive study in adults to date, Cardinale and colleagues reported a 9% incidence of anthracycline cardiotoxicity (LVEF fall >10% to below 50%) in breast cancer and lymphoma patients undergoing serial echocardiography after a median 5.2 years [38]. Half the patients crossing their diagnostic threshold for cardiotoxicity did so within 3.5 months of completing chemotherapy, and 98% within a year. Nevertheless, a small number of cases of cardiotoxicity developed between 5 and 10 years post chemotherapy, consistent with the progressive nature of anthracycline cardiotoxicity (evidence for which we also identified—above)[38]. This effect, at least in part, explains the increasing prevalence of cardiac dysfunction and heart failure in survivors of childhood cancer over subsequent decades [2, 38, 40, 41]. The mechanisms underlying this phenomenon remain to be fully elucidated given the multiple toxic effects of anthracyclines, but mitochondrial dysfunction (DNA damage and impaired bioenergetics [42]) and possibly cellular senescence (including shortening of telomere length and loss of cardiac progenitor cells [43]) may play a role. Meanwhile, declines in cardiac function after chemotherapy may also occur as a result of “multiple hits”: exposure to subsequent insults, such as coronary ischaemia, hypertension and impaired glucose homeostasis, the prevalence of which is high in cancer survivors [5, 41, 44]. Given these observations, and that falls in resting LVEF imply significant cardiac injury [5], it is quite plausible that those who have experienced falls in LVEF ≥5% following anthracyclines will have an increased lifetime risk of multifactorial heart failure. Even were individual risk low, the number of long-term survivors from breast cancer is large and expanding [8, 38]. Consequently, any small increase in population risk may result in a significant number of additional heart failure cases.

We do not advocate using an acute fall in LVEF≥5% clinically to guide chemotherapy, given that doing so would risk withholding potentially lifesaving chemotherapy.

Late gadolinium enhancement

Using LVEF alone at any-cut off to diagnose cardiotoxicity in individuals has well-documented limitations [5]. The ability to detect interstitial myocardial fibrosis (a hallmark of advanced anthracycline cardiomyopathy [45]) with CMR might offer additional prognostic value [46]. We observed no late gadolinium enhancement in the 120 cases imaged, confirming this technique is not useful for detection of early cAC [46]. However, it does not imply an absence of fibrosis: enhancement is only observed when focal areas of fibrosis are present. Newer CMR T1-mapping techniques (not available in the study period) may more accurately detect diffuse fibrosis [46].

Limitations

Our study does have limitations. Firstly, the cohort is of modest size. This is partially mitigated by prospective assessment using CMR (our study of 165 equates to >1000 subjects had 2-dimensional echocardiography been used [11, 17]), and discriminative power provided by the highly selected (homogenous) cohort. Even so, it is not possible to fit complex models and interactions of covariates using the data from studies of this size. Secondly, we use a subclinical endpoint, and assume that studying risk factors for subclinical cAC will give insights into the pathogenesis of more overt cardiotoxicity. Our ability to detect associations with established risk factors support this assumption. Thirdly, being selected by sex, race and for low incidence of cardiovascular co-morbidities (including known coronary artery disease or formally-diagnosed hypertension), the cohort does not accurately reflect the larger population of cancer patients, in whom the absolute risk of cAC is likely to be greater, and the relative contribution of the risk factors we identified remains to be confirmed. Fourthly, while we used the standard definition for elevated blood pressure, our measurements were insufficient to formally diagnose hypertension (for which repeated clinic or ambulatory measurements are required)[35]. Indeed, in seeking a homogeneous cohort with low prevalence of cardiovascular comorbidities (above) we had actively sought to exclude such patients. Fifthly, some datasets were not entirely complete: percentage body fat was only available in 118; blood pressure was unavailable for 1 subject and measured at a single time-point in 14%; and radiotherapy details were unavailable in 2%. In addition, the interval between chemotherapy and follow-up varied (largely due to difficulties in patients prioritising return). However, we view none of these factors as systemic confounders to the validity of our findings. Finally, our model was developed and tested on the same population and the ROC curve area may therefore be overestimated. Validation on a separate sample is required to confirm the predictive ability.