Characteristics of the included studies
After an initial literature search on PubMed, Web of Science, Embase, ScienceDirect, Google Scholar and clinicaltrials.gov databases, 189 articles were identified. After excluding duplicates and irrelevant studies, 59 potentially relevant articles remained. Of the 59 articles, 32 were further excluded due to mis-matching contents (21 studies did not report the comparison between sequential and concurrent regimens; 8 reviews and 3 case reports). For the remaining 27 articles, another 21 studies were excluded for the following reasons: no available data (n=7); no comparison between sequential and concurrent regimens (n=4); regimens did not contain an anthracycline and taxane (n=8); and duplicates (n=2). Finally, 6 articles were included in this meta-analysis [9-14] (Figure 1).
Among the 6 studies, a total of 6866 breast cancer patients after surgery were given the sequential regimen of an anthracycline and taxane as adjuvant chemotherapy, while 6847 patients received concurrent treatment (Table 1). The publication years ranged from 2010 to 2017. All of the studies were phase III randomized control trials.
Quality assessment
The detail of the risk of bias summary is outlined in (Figure 2). All studies were considered to have a median risk of bias. Randomized sequence generation was implemented in all 6 studies, and 4 studies implemented allocation concealment. All studies were conducted on the intention-to-treat principle. None of the 6 studies were blinded to the participants or the outcome assessment.
Meta-analysis for DFS
Significant heterogeneity among studies (I2=59%, P=0.03; Figure 3) was demonstrated in analysis DFS between the sequential and concurrent regimens, thus we used the randomized effects model to pool the RR. The meta-analysis showed that sequential regimens of anthracycline and taxane appeared not to add significant improvement in DFS over the concurrent regimen (RR: 1.05; 95% CI: 0.97-1.14; P=0.22, Figure 3).
Meta-analysis for OS
Significant heterogeneity (I2 =55%, P =0.05, Figure 4) was observed among studies for OS in comparison between the sequential and concurrent regimens, thus the randomized effects model was used. The pooled estimate showed that there was no significantly improved OS between sequential regimens and concurrent regimens (RR: 1.03, 95% CI: 0.94 to 1.13, P =0.51, Figure 4).
Sub-analysis in node status for DFS and OS
The eligible patients in the HORG trial [12] were early breast cancer patients at high risk and axillary lymph node-negative status, while the other trials included patients with node-positive status. We conducted a sub-analysis according to the axillary lymph node status. The pooled estimate showed that there was a significantly better DFS in patients with node-positive status who were administrated a sequential regimen (RR: 1.08; 95% CI: 1.02-1.14, P =0.004, Figure 5A), yet the OS was similar for both regimens (RR: 1.07; 95% CI: 0.96-1.19, P =0.24, Figure 5B).
The choice of anthracyclines may be another reason causing heterogeneity. Epirubicin was selected in the HE10/00 trial [14], while the other four trials used doxorubicin among patients with node-positive status. The sub-analysis data according to the choice of anthracycline showed that there was no significant heterogeneity, however, a better DFS (RR: 0.91; 95% CI: 0.86-0.97, P=0.002, Figure 6A) and OS (RR: 0.92; 95% CI: 0.84-0.99, P=0.03, Figure 6B) were achieved in patients with doxorubicin and a taxane in the sequential group.
The cycles of concurrent regimen with doxorubicin and a taxane also appeared to affect heterogeneity. The patients in the Big02-98[11] and NSABP B-30 trials [10]received 4 cycles of doxorubicin and taxanes, while patients in the other two trials [9, 13] were treated for six cycles. Therefore, we then conducted another sub-analysis. The pooled estimate showed that fewer cycles (4 cycles) of concurrent treatment had worse DFS (RR:1.16; 95% CI:1.06-1.27, P=0.0009, Figure 7A) and OS (RR:1.18; 95% CI:1.05-1.33, P=0.007, Figure 7B) compared to sequential regimen, whereas more cycles(6 cycles) rescued the loss.
Sensitivity analysis and publication bias
Sensitivity analysis showed that there was no significantly different incidence through omitting each study. No significant publication bias was found based on the Egger’s and Begg’s test (P > 0.05, Figure 8).