Cardiotoxicity of anthracycline agents for the treatment of cancer: systematic review and meta-analysis of randomised controlled trials

Abstract

Background: We conducted a systematic review and meta-analysis to clarify the risk of early and late cardiotoxicity of anthracycline agents in patients treated for breast or ovarian cancer, lymphoma, myeloma or sarcoma.

Methods: Randomized controlled trials were sought using comprehensive searches of electronic databases in June 2008. Reference lists of retrieved articles were also scanned for additional articles. Outcomes investigated were early or late clinical and sub-clinical cardiotoxicity. Trial quality was assessed, and data were pooled through meta-analysis where appropriate.

Results: Fifty-five published RCTs were included; the majority were on women with advanced breast cancer. A significantly greater risk of clinical cardiotoxicity was found with anthracycline compared with non-anthracycline regimens (OR 5.43 95% confidence interval: 2.34, 12.62), anthracycline versus mitoxantrone (OR 2.88 95% confidence interval: 1.29, 6.44), and bolus versus continuous anthracycline infusions (OR 4.13 95% confidence interval: 1.75, 9.72). Risk of clinical cardiotoxicity was significantly lower with epirubicin versus doxorubicin (OR 0.39 95% confidence interval: 0.20, 0.78), liposomal versus non-liposomal doxorubicin (OR 0.18 95% confidence interval: 0.08, 0.38) and with a concomitant cardioprotective agent (OR 0.21 95% confidence interval: 0.13, 0.33). No statistical heterogeneity was found for these pooled analyses. A similar pattern of results were found for subclinical cardiotoxicity; with risk significantly greater with anthracycline containing regimens and bolus administration; and significantly lower risk with epirubicin, liposomal doxorubicin versus doxorubicin but not epirubicin, and with concomitant use of a cardioprotective agent. Low to moderate statistical heterogeneity was found for two of the five pooled analyses, perhaps due to the different criteria used for reduction in Left Ventricular Ejection Fraction. Meta-analyses of any cardiotoxicity (clinical and subclinical) showed moderate to high statistical heterogeneity for four of five pooled analyses; criteria for any cardiotoxic event differed between studies. Nonetheless the pattern of results was similar to those for clinical or subclinical cardiotoxicity described above.

Conclusions: Evidence is not sufficiently robust to support clear evidence-based recommendations on different anthracycline treatment regimens, or for routine use of cardiac protective agents or liposomal formulations. There is a need to improve cardiac monitoring in oncology trials.

Figure 1

Results of literature search and selection of randomised controlled trials for systematic review.

Figure 2

Clinical cardiotoxicity defined as incidence of CHF in RCTs comparing different treatment regimens. The open diamond represents the pooled Peto Odds Ratio and 95% confidence interval (CI) for each treatment comparison. I-squared represents the proportion of variability between studies in excess of that expected due to chance, and p = probability that differences between study estimates are due to chance.

Figure 3

Sub clinical cardiotoxicity defined as reduction LVEF in RCTs comparing different treatment regimens. The open diamond represents the pooled Peto Odds Ratio and 95% CI for treatment comparisons 1-4, and relative risk (RR) with 95% CI for comparisons 5-7. I-squared represents the proportion of variability between studies in excess of that expected due to chance, and p = probability that the differences between study estimates are due to chance.

Figure 4

Clinical and subclinical cardiotoxicity in RCTs where cardiotoxicity outcomes could not be categorised as one or the other. The open diamond represents the pooled Peto Odds Ratio and 95% CI for treatment comparisons 1, 2 and 4, and relative risk (RR) with 95% CI for comparisons 5-7. I-squared represents the proportion of variability between studies in excess of that expected due to chance, and p = probability that differences between study estimates are due to chance.