Statistical analysis
The incidence and corresponding 95% confidence intervals (CIs) was calculated by the number of patients experiencing cardiac toxicities (LVEF decline or CHF) and total number of breast cancer patients treated with anti-HER2 agents were extracted from the safety profiles of included trials. Odds ratio (OR) and corresponding 95% CIs were used to assessed the risk of cardiac toxicities associated with dual HER-2 blockade versus anti-HER-2 monotherapy. We used the Peto method to calculate the ORs and 95% CIs because this method provided the best CI coverage and was more appropriate method for meta-analysis when dealing with low event rates(<5%)[15]. To determine the risk of cardiac toxicities associated with dual anti-HER2 blockade within particular groups, we also conducted the following prespecified subgroup analyses: the treatment setting (neoadjuvant, adjuvant and metastatic), the type of anti-HER2 therapies (dual HER2 blockade vs. trastuzumab; dual HER2 blockade vs. lapatinib), concomitant treatment (none, chemotherapy and hormonal therapy). If Peto OR>1, it reflected that more risk in dual HER2 blockade group, and vice versa. A statistical test with a p -value less than 0.05 was considered significant. To measure overall heterogeneity across the included cohorts, we calculated theI 2 statistic, withI 2greater than 50% indicating high heterogeneity. Sensitivity analysis using different statistical models was performed to assess the stability of results. We assessed potential publication bias by visual inspection of the symmetry of funnel plots and with the Egger regression asymmetry test. All statistical analyses were performed by using Version 2 of the Comprehensive MetaAnalysis program (Biostat, Englewood, NJ) and Open Meta-Analyst software version 4.16.12 (Tufts University).