Methods

This systematic review was conducted by 2 independent reviewers (ML, GBZ) solving divergences after consensus, with piloting of search strategies and abstraction forms. It was reported in keeping with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines and was prospectively registered on metcardio.org .

Google Scholar, MEDLINE/PubMed, and SciVerse were searched for pertinent studies. Specifically, MEDLINE/PubMed was queried according to the following highly sensitive strategy: (doxorubicin OR anthracycline* OR epirubicin OR adriamycin) AND (cardiotoxic* OR (cardiac AND toxicit*)) AND (predict* OR independent* OR Prognosis/Broad[filter] OR Therapy/Broad[filter]) NOT (animal [mh] NOT human [mh]) NOT (comment[pt] OR editorial[pt] OR meta-analysis[pt] OR practice-guideline[pt] OR review[pt]).

Studies were first screened at the title/abstract level, and then retrieved in full text if considered potentially pertinent. They were then included if reporting on at least 200 patients of any age group, treated with any type of anthracycline, for any type of cancer, following patients for at least 6 months, and reporting on independent risk factors of clinical or subclinical cardiotoxicity. Thus, we excluded studies reporting on <200 patients, lacking multivariable analysis for predictors of cardiotoxicity, or duplicating previous reports. Notably, we chose a priori to discard studies that included <200 subjects because they were unlikely to provide valid and robust incidence and prognostic estimates, given established assumptions of multivariable analysis methods.

Full text of short-listed articles were systematically appraised abstracting the following items: first author, journal, year of publication, location, number of patients with cancer, number of patients with breast cancer, breakdown of cancer, age at diagnosis, number of patients aged <18 years, female gender, hypertension, previous myocardial infarction, diabetes mellitus, number of patients treated with anthracyclines, breakdown of anthracyclines, cumulative anthracycline dose, concomitant radiotherapy, use of liposomal doxorubicin, and use of trastuzumab. Clinical and subclinical cardiotoxicity were defined in keeping with the individual study definitions, but in general as clinical signs or symptoms of cardiac disease and systolic dysfunction at echocardiographic or nuclear scan, respectively. In addition, the rate of cardiovascular events or admissions, cardiac death, and all-cause death were appraised. The internal validity of the studies was appraised with the Newcastle-Ottawa Scale.

Continuous data are reported as median (first to third quartiles) and categorical variables as n (%). In keeping with established methods, incidence estimates were pooled with a random-effects generic inverse variance approach, computing point summary estimates and 75% confidence intervals for statistical inference. Meta-regression was used to explore moderators of incidence estimates, with an inverse variance weighting method. Statistical inconsistency was appraised with I ² , and small study effects were explored with visual inspection of funnel plots. Predictors were not subjected to formal meta-analysis because they were reported in incomplete fashion in most articles. Instead, we relied on descriptive statistical analysis only for them. Computations were performed with RevMan 5 (Cochrane Collaboration, Copenhagen, Denmark) and SPSS 19 (IBM, Armonk, New York).

Results

The review flow is depicted in Figure 1 . From a total of 16,054 citations, 16,017 studies were excluded after initial screening, and 19 after retrieval as full text, leaving 18 studies ( Table 1 ). Included studies were published from 1979 to 2011, and enrolled a total of 49,017 patients with cancer, of whom 23,764 had breast cancer, and 22,815 receiving anthracyclines ( Tables 1 and 2 ). Overall study quality was satisfactory, although most studies stemmed from institutional databases and were retrospective ( Table 3 ).

Review profile.

After a median follow-up of 9 years the incidence of subclinical cardiotoxicity was 17.9% (11.6% to 24.2%, I ² = 99%), whereas the incidence of clinically overt cardiotoxicity was 6.3% (3.2% to 9.3%, I ² = 99%). Overall cardiovascular events occurred in 10.6% (3.5% to 17.6%, I ² = 100%), with cardiac death in 0.4% (0% to 0.9%, I ² = 72%), and all cause death, including cancer death, in 31.9% (3.7% to 60.2%, I ² = 100%). When subclinical cardiotoxicity occurred, the most common approach was to envision other nonanthracycline agents unless clearly not indicated to avoid further cardiac injury, whereas when overt cardiotoxicity occurred, chemotherapy and radiotherapy were discontinued altogether, and symptomatic therapy for heart failure was initiated.

Meta-regression analysis disclosed that the risk of subclinical cardiotoxicity was positively associated with the year of publication (β = 1.118, p = 0.032), and follow-up duration (β = 1.814, p = 0.001), whereas it was negatively associated with the number of included patients (β = −0.003, p = 0.002) and number of included patients treated with anthracyclines (β = 0.003, p = 0.002). No statistically significant moderator of clinically overt cardiotoxicity was identified, but these and the previous findings should be viewed in light of the low statistical power and risk of ecological fallacy typical of meta-regression. Nonetheless, funnel plot inspection confirmed the presence of small study effects for the risk of both subclinical and clinical cardiotoxicity.

Comprehensive appraisal of independent risk factors of cardiotoxicity showed that cumulative anthracycline dose was most consistently reported as an accurate and robust predictor of cardiotoxicity, with an acceptable prognostic role for chest radiotherapy, African-American ethnicity, very young or very old age, diabetes, hypertension, very high or very low body weight, or severe co-morbidities as well. Conversely, the prognostic impact of other covariates, such as gender, appeared less consistent. Given multiplicative assumptions of odds ratios (i.e., additive assumption of logistic regression coefficients), it can be hypothesized that patients with ≥3 risk factors face a five- to sixfold increased risk of cardiotoxicity compared with patients with no risk factor.