Optimal alternative therapy for metastatic HER2-positive breast cancer progression who previously received trastuzumab-based therapy: a systematic review and network meta-analysis

Background: Trastuzumab remains the standard protocol recommended for human epidermal growth factor receptor 2 (HER2)-positive metastatic breast cancer (MBC). After disease progression, the decision to continue trastuzumab treatment in combination with another type of chemotherapy, or to replace it with another HER2-targeted therapy, has not been conclusive. Methods: We systematically searched databases and extracted data from randomized controlled trials (RCTs) on patients who experienced disease progression after receiving trastuzumab-based therapies. Paired meta-analyses were performed to provide a comparison between patients continuously received trastuzumab-based therapy and those receiving other targeted therapies. Bayesian network meta-analysis was used to synthesize available evidence of direct or indirect comparison. Results: The 10 selected articles that included 3,158 patients were analyzed. The paired meta-analysis showed a 29% (hazard ratio (HR): 0.71; 95% confidence interval (CI): 0.60–0.85) reduction in mortality risk for patients continued receiving trastuzumab-based therapy than those receiving other targeted regimen. Bayesian network meta-analysis suggested that pertuzumab and trastuzumab plus capecitabine (PHC) showed an improvement in the overall survival (OS), with the highest surface under the cumulative ranking (SUCRA) (0.962 (82.3%)). In terms of progression-free survival (PFS) and objective response rate (ORR), trastuzumab emtansine (T-DM1) showed the highest SUCRA (0.949 (73.2%) and 0.997 (98.5%), respectively).

option for patients who experience disease progression after receiving trastuzumabbased therapy. Patients treated with PHC had the highest OS. Meanwhile, patients treated with T-DM1 showed better PFS and ORR than those treated with other regimens.

Background
Human epidermal growth factor receptor-2 (HER2) amplification occurs in 15%-20% of patients with breast cancer and is associated with more aggressive biological processes and poor prognosis [1,2]. Since trastuzumab has been proven beneficial for both early-stage and metastatic HER2-positive breast cancer, trastuzumab plus paclitaxel or docetaxel has been used as the standard first-line treatment for those group of patients [3][4][5].
Trastuzumab combined with adjuvant chemotherapy significantly improved the prognosis of patients with HER-2 positive breast cancer [6]. The HERNATA study found that trastuzumab plus vinorelbine is equally effective as trastuzumab plus docetaxel in the treatment of metastatic breast cancer [7]. The NEfERTT trial indicated that the progression-free survival (PFS) of patients treated with trastuzumab plus paclitaxel was similar to that of patients treated with neratinib plus paclitaxel [8]. Indeed, a dual HER2 blockade of trastuzumab and pertuzumab provides median PFS benefits over trastuzumab plus docetaxel for HER2-positive metastatic breast cancer [9]. In the last decades, numerous HER2-directed therapies has changed the course of HER2-positive breast cancer.
Nevertheless, most patients experience progression within 1 year, because of intrinsic or acquired resistance to trastuzumab-based therapy. The reasons for this resistance have not yet been well explicated. Some trials have reported the effectiveness of recombinant anti-HER2 agents as treatment for patients with metastatic breast cancer who experienced progression after receiving trastuzumabbased therapy. In the German Breast Group 26 trial, continuous use of trastuzumab plus capecitabine was more effective than treatment with capecitabine alone in patients with progression who were heavily pretreated with trastuzumab [10]. The PHEREXA study indicated that the addition of pertuzumab to trastuzumab and capecitabine did not improve PFS [11]. After switching to other HER2-directed therapies, the results were inconsistency in the trails of EMILIA [12] and LUX-Breast 1 [13].
Therefore, the decision to continue trastuzumab treatment even after disease progression is a pivotal perplexity for clinicians. The use of trastuzumab in combination with a different chemotherapy, or the use of other HER2-targeted therapies, remains inconclusive. We therefore aimed to conduct systematic review and meta-analysis to determine the optimal therapy for HER2-positive metastatic breast cancer patients who experienced disease progression after receiving trastuzumab-based therapies. We also aimed to address the above issues in some terms and provide a reference for clinicians.

Literature and search strategy
PubMed, Embase, and Cochrane Library for randomized controlled trials (RCTs) of metastatic HER2-positive breast cancer targeted therapy were searched up to July 31, 2018. To avoid missing documents, we searched the abstracts of the European Society for Medical Oncology (ESMO) and American Society of Clinical Oncology (ASCO). We also manually searched the reference lists of the included bibliographies and related reviews to supplement the primary sources. The gray literatures were excluded. The Medical Subject Headings/Emtree terms combined with free-text words, "breast cancer," "breast neoplasms," "randomized controlled trial," and known HER2-targeted agents (e.g., "T-DM1," "pertuzumab," "trastuzumab," and "lapatinib"), were the keywords used during the search. To avoid the omission of relevant literatures, we searched for articles that included a discussion on the pretreatment regimens and the inclusion and exclusion criteria for rigorous screening during follow-up. The search strategy was made by two authors, one with a background in carcinoma (Fubin Feng) and the other in advanced statistics (Jinhui Tian). The meta-analysis was reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [14].

Selection criteria
The following publications were included: (1) prospective phase II or III RCTs, (2) RCTs that were conducted among patients with pathologically confirmed HERpositive breast cancer, (3) RCTs that were conducted among patients with metastatic or advanced stage breast cancer, (4) RCTs that were conducted among patients with disease progression during or after trastuzumab-based therapy, (5) RCTs that were conducted among patients who were permitted but not required to receive anthracyclines and taxanes, (6) RCTs that contained at least two arms, and (7) RCTs with extractable outcomes (e.g., PFS, overall survival (OS), objective response rate (ORR)). By contrast, studies that included patients treated with two or more HER2-directed regimens in an advanced setting, studies that included patients who received endocrine therapy, and studies that were performed in patients treated with immunotherapeutic agents were excluded. The independent investigators (Fubin Feng and Tingting Zhang) evaluated the titles and abstracts of all included studies. If discrepancies existed, a third reviewer (Jinhui Tian) was consulted until a consensus was reached.

Data extraction and quality assessments
The authors (Fubin Feng and Tingting Zhang) extracted the relevant data from the original text according to the pre-designed form. The extracted data were as follows: first author's name, publication year, total number of patients in each RCT, treatment setting (dose, frequency, and treatment duration), funding source, characteristics of participants, and primary outcomes (e.g., PFS, OS, and ORR). We used the Cochrane Collaboration's risk of bias assessment tool to assess the risk of bias of included studies and recorded the evidence in order to provide a conclusion [15]. All studies were rated as follows across seven specified domains: "low risk of bias," "high risk of bias," or "unclear risk of bias." If discrepancies existed, the two authors need to reach a consensus via discussion.

Statistical methods
In the traditional paired meta-analysis, the regimens with trastuzumab and those without trastuzumab were compared. Hazard ratios (HR) and 95% confidence intervals (CIs) in PFS and OS were extracted from the included studies. The natural logarithm of HR and standard error of PFS and OS were pooled for the traditional meta-analysis, and then the effect size of odds ratios (ORs) were also calculated.
The heterogeneity of the pairwise comparisons was assessed using I 2 and expressed as p values. If the values of I 2 ≥50% indicated a large amount of inconsistency, then a random-effect model was used in the analyses. When the values were below 50%, a fixed-effect model was used in the analyses [16]. Using the STATA 13.0 command, a network plot was drawn to provide a direct comparison between the two interventions [17].
A network meta-analysis (NMA) with a Bayesian statistical model was used to indirectly compare two or more interventions. We used the WinBUGS version 1.4.3 (MRC Biostatistics Unit, Cambridge, UK) in the analysis, which employs Markov chain Monte Carlo methods. The selection of fixed model or random model was based on the deviance information criteria (DIC) value. The surface under the cumulative ranking (SUCRA) was used to rank the effectiveness of regimens in the network analysis [18]. As only a closed loop was presented in the meta-analysis, the nodesplitting method was used to evaluate the inconsistencies between direct and indirect comparison [19]. A meta-package was used to identify the inconsistencies within a closed loop. A funnel plot was used to evaluate the presence of publication bias, and the Egger test was used to check for asymmetries in the funnel plot. The results of Begg's and Egger's test demonstrated no significant publication bias (p>0.05) [20]. The statistical analyses were performed using STATA. The risk of bias was evaluated using the Review Manager (Version 5.3); creation of a network plot and analysis of SUCRA were performed using STATA Version 13.0 (Stata Corporation, College Station, TX, USA).

Overview of the literature search
A total of 3,214 studies were identified by database search, of which 795 duplicate articles were excluded. Of the remaining studies (n=2,419), 2,375 articles were excluded as they did not meet the inclusion criteria after screening the titles and abstracts. A total of 44 publications were selected for full review. The following trials were excluded: single-arm trails (n=12), trials with conference abstract and no relevant data (n=9), non-RCTs (n=7), and trials comprised only of one study (n=6). At length, 10 articles of 9 RCTs met the inclusion criteria. The PRISMA diagram is depicted in Figure 1. were not included in the network analysis but were included in the paired metaanalysis were as follows: trastuzumab plus vinorelbine (HV regimen), afatinib plus vinorelbine (AV regimen), lapatinib plus trastuzumab (HL regimen), and lapatinib regimen [13,25,26]. The direct and indirect relationships of those regimens are shown in Figure 2. Seven included trials demonstrated the direct and indirect comparisons between LC regimen and other regimens, which was finally performed in the Bayesian analysis. All trials, except the EGF104900, that have explicitly provided the data on PFS, OS, and ORR were reported in one study. The EGF104900 trial reported the data on PFS and OS in two publications [25,26]. All trials had two arms and presented the hormone receptor status; patients who received continuous trastuzumab-based therapy were reported in five studies [10,11,13,21,25], of which one study compared the effects of trastuzumab mono anti-HER2 therapy with that of dual anti-HER2 regimens [11]. The characteristics of the included studies were depicted in Table 1, and the studies were considered reliable.

Risk of bias assessment
The variable qualities of nine included RCTs is shown in Figure 3. Five trials had an unclear risk for random sequence generation [10,22,23,24,25]. All trials were open label and two were thought exhibit attrition bias due to incomplete outcomes data [13,24]. Two articles reported the independent review facility-assessed PFS [11,12].
Meanwhile, one trial did not report the OS clinical data in the subgroup analysis and only mentioned the PFS at specified time points [21].

Publication bias
The funnel plots did not reveal any evidence of obvious asymmetry for PFS, OS, and ORR ( Figure 6). In the Egger's test, the p values for PFS, OS, and ORR were 0.741, 0.721, and 0.262, respectively, and no significant publication bias was observed in the included studies. To determine whether continuous use of trastuzumab is beneficial for HER2-positive metastatic breast cancer (MBC) patients previously treated with trastuzumab, a traditional meta-analysis was conducted and suggested that patients who received continuous trastuzumab-based therapy showed better PFS and OS than those who received non trastuzumab-based therapy. The HR for OS and PFS reduced (29% and 21%, respectively) and had lesser inconsistencies. However, for ORR, no significant difference was observed in our meta-analysis (HR: 1.16; 95% CI: 0.98-1.37). This suggests that trastuzumab combined with another type of chemotherapy remains a better option in this setting. However, the progress on prior trastuzumab-based therapy within 1 year sometimes suggested superior results of PFS in the group lapatinib plus capecitabine than trastuzumab plus capecitabine. Some previous studies predicted that those patients who developed intrinsic resistance or early resistance to trastuzumab might be not indicated for continuous trastuzumab therapy [21]. The definition of trastuzumab resistance is challenging; as indicated in some trials, a considerable proportion of patients retain sensitivity to trastuzumab after experiencing progression during prior trastuzumab-based therapy.  [12,22]. The inconsistency between modest PFS and increase OS effect was unclear and should not undermine the importance of OS result in the network analysis. Just as the ASCO suggested, the trastuzumab/pertuzumab combination remains the first-line treatment and T-DM1 as the second-line treatment because of its negative effects on PFS [28,29].
Our results showed that T-DM1 showed better PFS and ORR than other regimens. In the included EMILIA study, T-DM1 significantly prolonged the PFS and OS of HER2positive breast cancer patients previously treated with trastuzumab compared with lapatinib plus capecitabine [12]. T-DM1 should be used as an alternative therapy for patients who are heavily treated with trastuzumab as recommended by the National Comprehensive Cancer Network Panel [30]. In the indirect comparison, T-DM1 demonstrated superior PFS (HR: 0.65; 95% CI: 0.46-0.93) and ORR (JR: 3.85; 95% CI: 1.92-7.69) compared with HC and an increasing ORR compared with PHC. In the MARIANNE study, T-DM1 plus pertuzumab did not show any improvement in OS compared with trastuzumab plus taxane as first-line treatment for HER2-positive advanced breast cancer [31]. The PFS for T-DM1 alone was also non-inferior to that of trastuzumab plus taxane [31].
There were discrepancies in the results of previous studies that evaluated the effects of tyrosine kinase inhibitors (TKIs; neratinib and afatinib) as an alternative regimen. Neratinib is an oral irreversible pan-ErbB receptor TKI, while afatinib is an irreversible HER1 and HER2 inhibitor [32]. The EGF104900 trial compared trastuzumab plus lapatinib with lapatinib as treatment for MBC patients who previously received trastuzumab and indicated the synergistic effect of these regimens on the PFS and OS of both treatment groups [25,26]. Regrettably, the dual HER2 inhibition with trastuzumab plus lapatinib was not included in the network analysis, as no direct comparison was made between this regimen and other regimens. It may be feasible to perform an NMA; however, in a traditional metaanalysis of the efficiency of trastuzumab-based therapy versus non trastuzumabbased therapy, all regimens were included to increase the feasibility of the present NMA. In the latest ASCO Annual Meeting, a biomarker analysis from EGF104900 showed that subtype defined by PAM50 and HER2/neu mRNA expression had a higher response rate and longer PFS and OS [29,33]. The use of trastuzumab plus neratinib showed a 27% ORR in 33 patients with advanced breast cancer who were previously treated with trastuzumab [34,35]. Oral neratinib showed substantial clinical activity, with a significant clinical activity (ORR: 24%) and 59% 16-week PFS rates among patients who previously received trastuzumab treatment [36]. In the two-arm study, neratinib was neither inferior nor non-inferior to lapatinib plus capecitabine [23]. According to the network-analysis results, it is essential to identify which clinical subsets of patients are the best beneficiaries.
The study provides insight into the treatment of patients with metastatic HER2positive breast cancer progression who were previously treated with trastuzumab, using a Bayesian statistical model. Some limitations of the study should be acknowledged. First, all included studies were open-label studies; although no significant bias was detected by the Review Manager, this finding will not improve the reliability of our study. Second, the data were extracted from published studies, which may have resulted in publication and reporting bias. Third, the estrogen receptor status in some studies was indistinct, and the biological differences may affect the prognosis, although endocrine therapy and immunotherapeutic agent were excluded. Fourth, we mainly focused on patients previously treated trastuzumab-based therapy who experienced progression, omitting pertuzumab and other HER2-target regimens relevant to the progression, as some of them are used in the clinical practice. Hence, caution must be taken when extrapolating the results of patients previously treated with pertuzumab and other HER2-target regimens.
Finally, we had no access to patients' data on the mechanism of drug resistance and prognosis of molecular and subgroups. Thus, it was hard to identify which clinical subsets of patients are most likely to benefit from alternative treatment.

Conclusions
This NMA showed that the continued use of trastuzumab with another type of chemotherapy remains a better option for patients previously treated with trastuzumab-based therapy who experienced progression. The degree of benefit may depend on the time of progression and the mechanism of resistance, which should be further explored. Patients treated with PHC will more likely exhibit an improvement in OS. Moreover, patients treated with T-DM1 showed better PFS and ORR than those treated with other regimens. In general, these two therapeutic regimens are indistinguishable and require further evaluation using larger head-tohead clinical trial validation and elaborate subgroup analysis.

Availability of data and materials
All data generated or analysed during this study are included in this published article.

Competing interests
The authors declare that they have no competing interests

Funding
This study was supported by grants from National Science Foundation of China (nos. 81473513 and 81673799). The sponsors did not play a role in the article publication.

Authors Contributions
CGS, FBF and TTZ conceived and designed the study; FBF, FY, LYQ, XW and JHT performed data analysis; CL, JZ and FBF contributed analysis tools; FBF and CGS wrote the paper.     The surface under the cumulative ranking probability curve (SUCRA) of the comparisons