Three electronic databases were searched (MEDLINE, Scopus, and ISI Web of Science) to identify potentially eligible articles using a pre-defined search strategy (Appendix A).

The search encompassed all articles from inception to October 25, 2020.

This process resulted in 857 articles in the MEDLINE database, 1048 articles in ISI Web of Science, and 840 in Scopus (Figure 1). A further 19 articles were later identified, mainly through manual searches and citations.

Figure 1.

Flowchart of study selection using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram to illustrate the study selection process.

(0,46MB).

We considered only human studies assessing the effects of HER-2 directed agents in breast cancer patients undergoing chemotherapy and reporting on independent risk factors. This strategy ensured that articles that did not mention dyslipidemia in their title or abstract due to non-significant results were included. The included studies assessed the impact of risk factors for cardiotoxicity, analyzing at least two study arms, comparing either patients with cardiotoxicity to those without or patients with a given factor to those without. Subsequently, the studies were screened for the role of dyslipidemia in the development of TIC.

Our primary outcome was cardiotoxicity, defined according to the criteria used in the HERA trial17 as symptomatic (e.g. HF and/or dyspnea, and/or referral to a cardiologist) or asymptomatic (e.g. decline in left ventricular ejection fraction [LVEF] >10% from baseline or LVEF <50%). We did not, however, exclude articles that diverged slightly from this definition for the qualitative and quantitative synthesis.

Secondary outcomes consisted of symptomatic cardiotoxicity, discontinuation of trastuzumab due to cardiac causes, recovery of cardiac function after a cardiac event and reintroduction of therapy after discontinuation.

We excluded (1) studies that mainly focused on anthracycline effects rather than trastuzumab; (2) studies that analyzed outcomes in a pediatric population; (3) non-human studies; (4) studies that followed patients for less than six months; and (5) guidelines, systematic reviews and meta-analyses, case reports, editorials, letters, and/or review articles with no original data.

No articles were excluded based on population size, publication date, or language.

After removal of duplicates, two reviewers (JFP and MMC) independently screened the articles at title/abstract level according to the predefined inclusion and exclusion criteria. Afterward, the two reviewers (JFP and MMC) independently analyzed the full texts of studies not previously excluded using the same inclusion and exclusion criteria. Disagreements were resolved by consensus with a third reviewer (CDS) serving as final arbitrator. Efforts were made to contact investigators in order to obtain publications not accessible by other means.

Two reviewers (JFP and MMC) independently analyzed the full texts that had met the inclusion criteria and extracted data into a pre-established spreadsheet.

The full text of short-listed articles was systematically appraised for the following items: first author, year of publication, nationality, study setting, study design, number of patients with breast cancer, number of patients treated with anthracyclines and/or trastuzumab, duration of follow-up, patients’ median age and age at diagnosis, method of LVEF assessment, and number of patients developing or not developing cardiotoxicity (or number of patients with and without a given cardiovascular risk factor).

The risk of bias was assessed at the study level according to the method used by Haffar et al.,18 given that for the most part, the studies included are observational. This method was also used to classify sub-analyses of randomized controlled trials (RCTs) since the primary subject of these studies was not the effect of dyslipidemia. Two reviewers (JFP and MMC) independently analyzed the studies. Any disagreements were resolved by consensus.

Review Manager® (version 5.4.1)19 was used for statistical analysis and to derive forest plots.

The frequency of a given risk factor was assessed using absolute and relative frequencies. If these were not present, they were calculated.

Odds ratios (ORs) were used as a summary measure. The precision of effect sizes was measured using 95% confidence intervals (CIs) and corresponding p-values for both. We chose the OR since relative estimates are more comparable than absolute effects between studies with different designs, populations, and lengths of follow-up.20

The Cochran Q test (I2)21,22 was used to assess heterogeneity and statistical inconsistency. Small study effects and reporting bias were explored with a visual inspection of asymmetry in funnel plots.23 We classified heterogeneity as 0%, signifying absence of detected heterogeneity; 0–10%, indicating low heterogeneity; 10–50%, indicating moderate heterogeneity; and over 50%, indicating high heterogeneity.

A random-effects model was used to pool data owing to the anticipated heterogeneity in the included trials. In fact, adjusted indirect comparisons that use a fixed-effects model tend to underestimate the standard errors of pooled estimates.24

To better understand the specific effect of dyslipidemia alone as a risk factor for cardiotoxicity, we also performed a pre-planned subgroup analysis involving only studies reporting the most adjusted measures.

The search returned 2745 records, of which 1407 remained after removal of duplicates. After title and abstract screening, followed by full-text appraisal, 39 articles matched our eligibility criteria and were included in our systematic review.25–63 Afterward, 18 records were excluded from the meta-analysis because of data duplication (n=5), missing data (n=8) or being sub-analyses of large RCTs (n=5), and were thus not suitable for use in pooled estimates in conjunction with observational studies. Hence, 21 studies were further examined by meta-analysis (Figure 1). Table 1 summarizes the characteristics of the included studies.

Of the 39 articles included in the qualitative analysis, most were Italian31–36,49–51,57,58 (n=9) or American29,38,41,48,54,55,63 (n=7), and all were published between 200555 and 2020.25,30,33,35 The shortest follow-up time was six months47,61 and the longest was 9.5 years.33 Seven studies were prospective and observational,30,43,45–47,59 27 were retrospective and observational,29,31–34,36,37,39,41,42,44,49–53,56–58,60–63 and five were subanalyses of large trials35,38,48,54,55 (of the ALTTO trial,35 NSABP B-31,38,48,55 and HERA.54 The studies included a total of 21079 patients (17998 excluding those based on the same population32,38,46,48,50,51,57), with ages ranging from 2027 to 9253 years, and one study focused on the elderly population.53 No mentions of male patients were found. The median sample size was 179 patients, ranging from 4553 to 4190.35

Patients were treated in both (neo)adjuvant and palliative settings. Only 11 studies did not analyze HER-2 confirmed breast cancer.26,34,39,40,42,43,49,51,52,58,62 Some studies had higher rates of anthracycline use and lower rates of trastuzumab therapy. This was most evident in Kaboré et al.43 (43% patients treated with trastuzumab) and Yoon et al.32 (15.7%). We tried to keep this bias to a minimum by excluding three articles in which trastuzumab use was less than 5% and were therefore judged to have inadequate intervention. These characteristics and other several variations in the treatment for adjuvant/neoadjuvant therapy, dosing, and regimen are further described in Table 1.

The definition of dyslipidemia was similar between studies. Alghafar et al.25 and Farolfi et al.36 defined it as total plasma cholesterol >5.2 mmol/l or use of lipid-lowering medications, while Ganz et al.,38 Romond et al.48 and Tan-Chiu et al.55 used only lipid medications as the marker for dyslipidemia. Matos et al.45 defined dyslipidemia as a combination of low-density lipoprotein cholesterol >3 mmol or total cholesterol >5 mmol/l. Piotrowski et al.,47 Russo et al.50,51 and Tarantini et al.57 defined dyslipidemia as total serum cholesterol >190 mg/dl or lipid-lowering therapy or triglycerides >150 mg/dl. Bonifazi et al.34 used International Classification of Diseases, Ninth Revision (ICD-9) codes to identify patients diagnosed with dyslipidemia.

In our systematic review, after excluding articles dealing with the same populations,32,38,46,48,50,51,57 lacking information on dyslipidemia prevalence,60,62 and large subanalyses of RCTs,35,38,48,54,55 we found the overall prevalence of dyslipidemia to be 11.32%, ranging between studies from 1.75%39 to 43.9%.33 In the meta-analysis this figure was 15.84% and in RCTs only, it was 7.01%.

Most studies included in this systematic review defined cardiotoxicity as a ≤10–16% decline in LVEF, a decline in LVEF to <50–55%, or patients exhibiting signs and symptoms of heart failure.25–33,35–38,40–46,48–63 All these studies used either echocardiography or multigated acquisition (MUGA) scan as a tool to serially assess LVEF. However, outcome definitions and assessment of results differed widely between different studies, with a vast array of descriptive terminology being used. Bonifazi et al.34 defined the primary outcome as ICD-9 code reports referring to possibly drug-induced cardiac hospitalization in the main diagnosis, while Gong et al.39 and Rossi et al.49 defined it as hospitalizations due to cardiac events, and it was defined as use of medication in Ganz et al.,38 Romond et al.,48 and Tan-Chiu et al.55 We also included other studies in which there were slight variations in the percentages of LVEF decrease29,36,44,46,62,63 (Table 1), such as asymptomatic cardiotoxicity defined as LVEF drop ≥16%29,63 or LVEF drop ≥15%,36,46 or for symptomatic cardiotoxicity, LVEF drop ≥10% to LVEF <55%62,63 or LVEF <53%.44

A detailed characterization of cardiotoxicity-related parameters is presented in Table 2. In this systematic review, excluding RCTs,35,38,48,54,55 we found the overall incidence of TIC to be 11.94%, with 5.59% of patients presenting symptoms of heart failure. Performing the same calculation only for RCTs, we found these values to be 9.18% and 2.4%, respectively. In the primary meta-analysis, the cardiotoxicity rate was 13.55%.

In observational studies that provided data concerning discontinuation of trastuzumab, 11.13% of 3808 patients discontinued treatment temporarily or permanently due to cardiac complications. The figure was 6.61% when only RCTs were considered.

Furthermore, in studies that provided data concerning recovery of cardiac function after trastuzumab discontinuation, recovery was observed in 72.12% of patients discontinuing trastuzumab due to cardiac reasons.

Also, in studies reporting reintroduction of trastuzumab after discontinuation, we observed that 43.65% presented significant recovery of cardiac function that enabled continuation of therapy.

Six of the included studies28,31,36,43,52,53 presented univariate estimations of ORs concerning the role of dyslipidemia in trastuzumab-induced cardiotoxicity, while another six presented multivariate data.26,28,35,38,47,53 In all other studies, we calculated ORs based on demographic data found in the reports.

Of the latter studies, only Bonifazi et al.34 found a significant association (OR=2.28, 95% CI 1.22–4.26, p=0.01). The study's original summary measure was the hazard ratio (HR). This study identified the rate of severe cardiac adverse events among 2046 women treated with trastuzumab for early breast cancer, through a record linkage between health care databases and searches for records concerning ICD-9 codes referring to cardiac events and cardiovascular risk factors. A cumulative risk of cardiotoxicity was then estimated using the Kaplan-Meier method over a follow-up of three years. The predictors found were age and history of cardiac disease.

Other studies were close to the significance threshold. Gunaldi et al.42 retrospectively assessed a sample of 111 women, however, a significant association between hyperlipidemia and LVEF decrease was not found (OR=2.65, 95% CI 0.90–7.79, p=0.08). Kaboré et al.43 used prospective data in a French national multicenter study aiming to examine the association of body mass index (BMI) and cardiotoxicity defined as a decrease in LVEF in a total of 929 patients. In multivariate analysis, obesity was independently associated with cardiotoxicity. However, no significant association was found for dyslipidemia on univariate analysis (OR=2.46, 95% CI 0.97–6.23, p=0.05). This proximity to significance may also be in part explained by the low rates of trastuzumab use (43%).

Furthermore, studies that performed subanalyses of large RCTs,35,38,48,54,55 including studies reporting multivariate measures,35,38 did not observe such an association either: OR=0.90, 95% CI 0.60–1.36, p=0.63 as observed in Eiger et al.35 and OR=1.92, 95% CI 0.85–4.36, p=0.12 for the association between baseline elevated lipid profile medication use and low Duke Activity Index Status score in Ganz et al.38

Other studies were not included in the meta-analysis because of unavailability of data to calculate OR. These studies used other summary measures such as risk ratio (RR)41,55,62 or HR,39,40,48 however, they also failed to find an association. Of note that one of these studies performed rigorous adjustments for HR.39

Our random-effects meta-analysis encompassed 21 studies5,25,26,28–31,33,34,36,40–43,46,53,56,58,59,61,63 and included 6135 patients. The prevalence of dyslipidemia was 15.84%. The pooled estimate for the OR of cardiotoxicity for individuals with dyslipidemia undergoing trastuzumab treatment for breast cancer was 1.25 (95% CI 1.01–1.53, p=0.04, I2=0%) (Figure 2).

Figure 2.

Forest plot representing the effect on cardiotoxicity of dyslipidemia compared with the absence of a diagnosis of dyslipidemia in trastuzumab-based breast cancer treatment (odds ratio with 95% confidence interval, random effects meta-analysis). CI: confidence interval; OR: odds ratio; SE: standard error.

(0,47MB).

We also performed a subgroup analysis by selecting only studies reporting the most adjusted results,26,28,30 which resulted in a pooled estimate for the OR of cardiotoxicity for individuals with dyslipidemia undergoing trastuzumab treatment for breast cancer of 0.89 (95% CI 0.73–1.10, p=0.28, I2=0%) (Figure 3). However, one study30 demonstrated a disproportionate weight in the pooled result, but it did cause significant heterogeneity.

Figure 3.

Forest plot representing the effect on cardiotoxicity of dyslipidemia compared with the absence of a diagnosis of dyslipidemia in trastuzumab-based breast cancer treatment only in studies reporting adjusted data (odds ratio with 95% confidence interval, random effects meta-analysis). CI: confidence interval; OR: odds ratio; SE: standard error.

(0,16MB).

There were no significant differences between individual studies in the magnitude of the association between dyslipidemia and cardiotoxicity in our primary meta-analysis, as indicated by the statistical test for heterogeneity (tau2=0.00; chi-square=17.21, df=20 [p=0.64]; I2=0%).

However, across studies, the treatment was heterogeneous due to the use of different regimens (Table 1), which makes it difficult to define the effect specifically caused by dyslipidemia in the setting of trastuzumab therapy.

The quality of the studies and the risk of bias were assessed at the study level using the method of Haffar et al.18 Overall, we judged 13 studies to be of moderate quality, 1227,29,33,36,40–42,44,52,53,59,61 due to incomplete exclusion of pre-existing LVEF impairment and one due to lack of specification of the tool used for outcome assessment.34 One study was deemed to be of low quality49 because of failure to collect diagnosis details and unclear diagnostic method. The results of this assessment are shown in Table 3.

Publication bias and small-study effects were assessed for all collected variables, as demonstrated by the funnel plot of data from the 21 studies included in the meta-analysis of raw ORs, which was asymmetrical (Figure 4). These data indicate that included studies with small sample sizes appear to underestimate the effect of dyslipidemia.

Figure 4.

Funnel plot of included studies. OR: odds ratio; SE: standard error.

(0,09MB).

Despite the existence of a plethora of studies on the mechanisms and characteristics of TIC, there remains ambiguity concerning robust clinical predictors for this complication. The systematic review by Jawa et al.13 identified hypertension, diabetes, age, and previous anthracycline use as risk factors for a cardiac event, but failed to prove other associations with known risk factors, including dyslipidemia. Attempts have also been made to develop risk scores that will predict rates of heart failure and cardiotoxicity to enable appropriate monitoring in this group,48,64 considering such factors as baseline LVEF,48 age,48,64 adjuvant chemotherapy, coronary artery disease, atrial fibrillation or flutter, diabetes, hypertension, and renal failure.65

Dyslipidemia is linked with heart failure and coronary heart disease in the general population.14 There have been several animal and human studies that point to the beneficial effect of statins in the context of TIC,66–68 and a study has shown that rats fed a high-lipid diet are more sensitive to anthracycline-induced cardiotoxicity.65 We therefore hypothesized that dyslipidemia may play a role in TIC. A previous meta-analysis13 failed to show such an association; however, most of the literature had been based on small samples and there was a lack of studies presenting formal multivariate adjustment, hence the need for an up-to-date and comprehensive systematic review on this topic.

This systematic review was planned and designed to assess the association between dyslipidemia and TIC in breast cancer patients and included 39 studies found by a systematic search regarding this topic.

We found the overall prevalence in this systematic review of dyslipidemia in observational studies to be 11.32%, ranging between studies from 1.75%39 to 43.9%.33 This wide range may be explained because a large proportion of patients excluded in our primary analysis were in subanalyses of data from large RCTs,35,38,48,54,55 in which the dyslipidemia prevalence was 7.01%. These trials included relatively healthier and younger patients. Consequently, there is a risk that it may inadequately represent the real-world breast cancer population, who may present with a higher prevalence of risk factors, including dyslipidemia. Indeed, a major research concern in oncology is the lack of information on elderly populations.69 This question was addressed in a study included in our review that included a population with a median age of 75.9 years53 and that presented a higher prevalence of dyslipidemia (28.9%) than most included studies.

We documented a significant rate of TIC in this population, around 12.0% in the overall review of observation studies, which is consistent with previously reported data.13,70 A recent pooled analysis of adjuvant trials investigated the incidence of TIC and its impact on treatment completion.70 The incidence of symptomatic heart failure in our meta-analysis was 3.18%, which was slightly higher than the figure reported by these authors (2.3%). This suggests that cardiotoxicity may be more frequent outside clinical trials, as observed in recent retrospective cohorts.4

In the qualitative and quantitative synthesis of the data, only one study, Bonifazi et al.,34 showed a clear association between dyslipidemia and trastuzumab-induced cardiotoxicity (OR=2.28, 95% CI 1.22–4.26, p=0.01). This study had the third largest sample size (2046 patients); however, ICD codes were used for both identification of patients diagnosed with dyslipidemia and the incidence of cardiac events (defined as hospitalization). Hence, we judged this study to be at moderate risk of bias for our meta-analysis, and thus the results in the meta-analysis of unadjusted measures may be overestimated due to the weight of this study. Other studies appear to suggest an association; however, in none did it achieve statistical significance.33,42,43 These findings are further supported by the absence of association found in our subgroup analysis, which only took into consideration adjusted measures. However, it is of note that studies were also unclear about their methods of adjustment, with only Ayres et al.28 specifying that their adjustments controlled for age and BMI.

In our meta-analysis, an association of dyslipidemia with TIC was observed in the pooled data from raw ORs in observational studies. This association was not supported by a subgroup analysis of studies reporting the most adjusted measures. This suggests that the role of dyslipidemia may merit further attention, since it is a condition that often exists in interplay with other comorbidities in a synergistic interaction. Indeed, obesity,35,42,43,59,71 diabetes,13,35,59,65 and hypertension13,30,38,42,47,49 are factors reported to be associated with TIC, which suggests that metabolic syndrome as a whole could be linked to this complication. A study by Kosalka et al.44 found that, compared with any risk factor alone, the combination of two or three comorbidities (such as diabetes, dyslipidemia, and obesity) is associated with a significant increase in the incidence of symptomatic cancer therapy-related cardiotoxicity.

This systematic review was conducted according to the PRISMA guidelines.15,16 Data selection was rigorous, and the analysis was thorough. We were also conservative in our analysis, as undefined data were not considered and the most precise and adjusted measures were extracted for a subgroup analysis.

However, this study has all the limitations inherent to systematic reviews and meta-analyses, particularly of observational studies. Overall, we judged thirteen studies to be of moderate quality27,29,33,34,36,40–42,52,53,59,61 and one of low quality.49 There is considerable heterogeneity between the included studies concerning sample size, therapeutic regimens, cancer stages and molecular profiles, patient demographics, and inclusion/exclusion criteria and follow-up in each study.

The fact that advanced cancer predisposes to further doses of treatment should be taken into consideration, since it may enhance the risk for cardiac harm and consequently overestimate the incidence of cardiotoxicity. Furthermore, different imaging modalities were used, particularly MUGA scans as opposed to echocardiography, which is currently considered the preferred imaging modality for surveillance.72 We should also bear in mind that there is still no clear consensus on the correct definition of cardiotoxicity, hence our outcome aggregates subclinical and clinical cardiotoxicity as a single endpoint. This highlights the need to create a universally acceptable definition of cardiotoxicity in this setting that could be uniformly used in future trials.

Although we tried to minimize the risk of selection bias by performing a wider initial search for all possible risk factors for TIC, the studies included were not specifically designed to address this question and the definitions of dyslipidemia were not always stated and differed between studies. This led to bias in data collection, as we were only able to extract data from studies in which the authors considered dyslipidemia to be a relevant factor.

Although our systematic review and meta-analysis found an association between dyslipidemia and cardiotoxicity in a pooled estimation of unadjusted data, the same did not apply for the meta-analysis of multivariate measures. Hence, in the absence of other relevant cardiovascular risk factors, routine review of these patients’ lipid profile may not be as important as previously thought. Also, breast cancer patients who are candidates for trastuzumab therapy and who present isolated dyslipidemia should probably not be considered at high risk for the development of cardiotoxicity and can be managed in the same way as patients with no dyslipidemia, without referral for cardio-oncology assessment.