Skip to main content
NCBI home page
Cancer Medicine logoLink to Cancer Medicine
. 2018 Dec 7;8(1):383–399. doi: 10.1002/cam4.1892

A Bayesian network meta‐analysis of the efficacy of targeted therapies and chemotherapy for treatment of triple‐negative breast cancer

Huihui Chen 1, Wei Lu 1, Yixin Zhang 1,2, Xuan Zhu 1,3, Jiaojiao Zhou 1,3, Yiding Chen 1,3,✉
PMCID: PMC6346255  PMID: 30525293

Abstract

Triple‐negative breast cancer (TNBC) is a heterogeneous disease with poorer prognosis than other subtypes, yet effective therapies are still not available. We aimed to compare the efficacy of various targeted therapies with chemotherapy (CT) in TNBC patients using a network meta‐analysis. A systematic literature search was performed in PubMed, EMBASE, and the Cochrane Library. A total of 27 randomized controlled trials (RCTs), involving 6924 TNBC patients, were included. Olaparib significantly improved PFS (0.43, 0.29‐0.64) and ORR (2.57, 1.31‐5.09) in comparison with CT. As for bevacizumab + CT, it showed a significant improvement of PFS (0.66, 0.55‐0.80) and ORR (2.15, 1.16‐4.05) compared with CT + placebo. It was also superior to CT alone in PFS (0.48, 0.35‐0.65) and pCR (1.30, 1.13‐1.49 for breast and axillary nodes and 1.26, 1.11‐1.44 for breast). Other targeted agents like iniparib, sorafenib, cetuximab, and ipatasertib combined with CT showed significant superiority in PFS compared with CT alone, and the HRs were 0.75 (0.62‐0.90), 0.44 (0.21‐0.91), 0.67 (0.47‐0.96), and 0.44 (0.24‐0.81), respectively, while some other agents such as sunitinib and cetuximab had the lowest SUCRA in OS, PFS, or ORR without any benefits. In conclusion, our results indicated that the addition of bevacizumab to CT was beneficial for TNBC patients, and olaparib had a great effect in PFS and ORR, especially for those with BRCA mutations. When combined with CT, targeted agents including iniparib, sorafenib, cetuximab, and ipatasertib may have better efficacies for treating TNBC.

Keywords: network meta‐analysis, randomized controlled trials, targeted therapies, triple‐negative breast cancer

1. INTRODUCTION

Breast cancer is the most common malignancy in women, with an estimated 1.67 million new cases diagnosed and 522 000 deaths worldwide in 2012.1, 2 Triple‐negative breast cancer (TNBC) represents 15% of breast cancers and is defined by the absence of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2).3 Notorious for its aggressiveness, TNBC is associated with a higher mortality than other subtypes.4 Due to its molecular characteristics, TNBC patients can not benefit from endocrine and anti‐HER2 therapies.5 Some studies revealed that many TNBC patients were highly sensitive to chemotherapy and had better neoadjuvant response rates compared with others,6, 7 suggesting chemotherapy may be an alternative therapeutic strategy. However, there is a high risk of relapse and disease progression after chemotherapy, and TNBC patients with residual lesions have significantly worse survival compared with non‐TNBC patients.7 Consequently, no standard therapy existed for TNBC patients and discovering effective targeted agents for this patient population has been a particularly urgent need.

In recent years, increased knowledge of the molecular alterations in TNBC has led to several promising clinical approaches, including inhibitors of poly ADP‐ribose polymerase (PARP), angiogenesis, epidermal growth factor receptor (EGFR), etc, which are currently being evaluated and may shed new light on treatment strategies.5, 8 Bevacizumab, a directed vascular endothelial growth factor A (VEGF‐A) antagonist, was chosen as a candidate treatment.8 Addition of bevacizumab to chemotherapy was reported to improve response rates and time to progression among TNBC patients.9, 10 Preclinical and clinical studies have shown that BRCA‐mutated tumors had increased responses rates to PARP‐inhibitor therapy.11 Even with a lack of a BRCA mutation, a substantial proportion of TNBC patients showed biological similarities with BRCA‐associated breast cancers,5, 12 making it possible that PARP inhibitors could be applied to TNBC patients. In the neoadjuvant I‐SPY2 trial,13 veliparib combined with carboplatin increased pCR rates from 26% to 51% in the TNBC group. Apart from all the above, some other molecular targeted drugs such as perbrolizumab (a monoclonal antibody targeting PD‐1), glembatumumab vedotin (an antibody‐drug conjugate [ADC] targeting transmembrane glycoprotein NMB [gpNMB]), are attracting researchers’ attention, as well.5, 14, 15

Two previous traditional meta‐analyses were performed to explore the issue of whether the targeted therapy combined with chemotherapy (CT) was superior to CT alone.16, 17 However, there were many limitations in using the pairwise meta‐analysis. First, among all available treatments, a lack of head‐to‐head trials makes direct comparisons of certain treatments impossible. Second, study endpoints and targeted therapies are various among different trials. Bayesian network meta‐analysis enables indirect comparison using a common comparator when a head‐to‐head trial is not available, and it produces estimation of the relative effectiveness and rank ordering among all interventions.18, 19 Thus, we performed a network meta‐analysis to compare different study endpoints of various targeted agents in the treatment of TNBC.

2. MATERIALS AND METHODS

2.1. Literature search strategy

This study was designed in accordance with the instructions of the preferred reporting items for systematic reviews and the meta‐analysis (PRISMA) extension statement incorporating network meta‐analyses of healthcare interventions.20 PubMed, EMBASE, and the Cochrane Collaboration Central Register of Controlled Clinical Trials were searched, and the search strategy was provided in the Supporting Information. Bibliographies of the included trials and related reviews were also manually scanned for additional references.

2.2. Selection criteria

We included only randomized controlled clinical trials (RCTs) assessing the efficacy of targeted therapies alone or in combination with chemotherapy for TNBC patients. Study end outcomes included overall survival (OS), progression‐free survival (PFS), disease‐free survival (DFS), objective response rate (ORR) and pathological complete response (pCR), and hazard ratios (HRs) or relative risks (RRs) with their corresponding 95% confidence intervals (CIs) when available or could be calculated. Single arm studies, non‐English literature, duplicated studies, and studies without sufficient outcome measures were excluded.

2.3. Data extraction and assessment for risk of bias

Two abstractors (Huihui Chen and Wei Lu) independently extracted data from the included studies. Disagreements were resolved by discussion with a third reviewer until consensus was reached. We collected information about the first author, year of publication, the number of patients in each arm, age, TNM stage, ECOG performance status, neoadjuvant chemotherapy or not, treatment regimens, study endpoints, and their relevant HRs or RRs. If HRs and 95% CIs were indirectly reported in the form of Kaplan‐Meier survival curves, we estimated them according to Guyot's methods.21 For different follow‐up durations in the same trial, only data with the longest duration were adopted. Methodological risk of bias of individual studies was assessed by using the Cochrane risk of bias method, including random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessors, incomplete outcome data, selective reporting, and other types of bias,22 and each part was evaluated as low risk, unclear risk, and high risk of bias.

2.4. Data synthesis and analysis

In the first place, network plots showed direct comparisons between treatment arms and offered a brief description of their characteristics. The influence and contribution of each direct comparison to the whole network was then presented with contribution plots.23 In addition, we adopted comparison‐adjusted funnel plots to assess the potential small‐study effects in the network meta‐analysis.23

For Bayesian network meta‐analysis, we built the Bayesian framework based on the log‐hazard ratio scale and the posterior distribution of the treatment effect was estimated using Markov Chain Monte Carlo methods.24 Both fixed and random effects models were tested in the primary analysis, and the selection of these two models was based on deviance information criteria (DIC) and convergence of the model. DIC was used to quantify the fit of the fixed or random effects model,19 with differentials of 2 to 5 between the random effects model and fixed effects model being considered important.25 Convergence of the model was examined by using Brooks‐Gelman‐Rubin diagnostic methods, inspecting autocorrelation between iterations of the Markov chain and determining whether the error of effect size was less than 5% of the posterior standard deviation.26 All Bayesian network analyses were performed using two chains, and after discarding the results of a burn‐in period of 40 000 iterations, each chain had a sample of 200 000 iterations which were then thinned every 20th iteration to reduce autocorrelation. Vague priors, such as N(0, 106) for the study‐specific baseline and treatment effect coefficients used to insure estimates of effect sizes and precision, were informed by within‐study differences between treatments rather than by differences in absolute response between the studies.24 The median and 95% credible intervals (CrIs) of the posterior distribution were recorded as estimates for effect sizes, and credible intervals could be interpreted as conventional CIs. The probability and the cumulative probability of rank for each treatment were calculated, and the surface under the cumulative ranking curve (SUCRA) was adopted to compare the efficacy of each treatment.27

To evaluate whether there was inconsistency between indirect and direct comparisons, we compared the estimated effect sizes from Bayesian network meta‐analysis with the corresponding pooled effect sizes from traditional pairwise meta‐analysis. The extent of heterogeneity across studies was tested using the I 2 test, and I 2 > 50% together with P < 0.05 indicated significant heterogeneity.28 We adopted a random effects model if significant heterogeneity existed; otherwise, we adopted a fixed effects model. For sensitivity analysis, we conducted network meta‐analysis based on the frequentist graph‐theoretical model.29, 30 Like the SUCRA value in Bayesian network meta‐analysis, the P‐score was used for ranking treatments in the frequentist network meta‐analysis without resampling.31

The Bayesian network meta‐analysis was performed using the Markov Chain Monte Carlo engine WinBUGS software (MRC Biostatistics Unit, Cambridge, UK), RevMan 5.0 software (The Cochrane Collaboration, Copenhagen, Denmark), Stata 12.0 (StatCorp, College Station, TX, USA), and R 3.2.1 (R development Core Team, Vienna, http://www.R-project.org) with the “netmeta” package (V.0.9‐2).30 The statistical significance level was set to P < 0.05 unless specified.

3. RESULTS

3.1. Study selection work flow and summary of characteristics of the included trials

As shown in Figure 1, a total of 12 584 studies through electronic database searches and nine studies from relevant studies or reviews were identified. After removing duplicated studies and initial screening, we retrieved the full text of 948 potentially eligible articles for detailed assessment, and 920 studies were excluded according to the inclusion and exclusion criteria. The treatment arms of the screened 28 studies were recorded, yet Dieras’ study, which compared the efficacy between onartuzumab + bevacizumab + CT, bevacizumab + CT + placebo, and onartuzumab + CT + placebo for TNBC patients, was excluded because the treatments did not share common nodes with the others.32 Finally, 27 RCTs involving 6924 TNBC patients were included and the characteristics of the included trials are displayed in Table 1.9, 10, 15, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56

Figure 1.

Figure 1

Open in a new tab

Flow diagram of literature search and study selection. A total of 27 randomized controlled trials that met the inclusion criteria were included in the network meta‐analysis

Table 1.

Characteristics of the included studies

Author‐year Trial identifier Median age (y) TNM Stage ECOG Neoadjuvant chemotherapy Number in arm 1 Number in arm 2 Arm 1 Arm 2 End points HR for OS HR for PFS RR for ORR RR for pCR breast and axillary nodes RR for pCR breast
Baselga‐2012 2007‐000290‐32 (EudraCT) NA IIIB, IIIC, IV 0‐1 No 20 33 Sorafenib and capecitabine Placebo and capecitabine PFS NA 0.60 (0.31‐1.14) NA NA NA
Baselga‐2013 NCT00463788 Arm 1:53
Arm 2:52 		IV 0‐2 No 115 58 Cetuximab and cisplatin Cisplatin OS, PFS, ORR 0.82 (0.56‐1.20) 0.67 (0.47‐0.97) 1.93 (0.83‐4.48) NA NA
Bell‐2016 NCT00528567 Arm 1:50
Arm 2:50 		T1b‐T3 0‐2 No 1301 1290 Bevacizumab and standard chemotherapy Standard chemotherapy OS, DFS 0.93 (0.74‐1.17) NA NA NA NA
Bergh‐2012 NCT00393939 NA Locally recurrent, IV 0‐2 No 58 69 Sunitinib and docetaxel Docetaxel PFS NA 1.03 (0.65‐1.63) NA NA NA
Brufsky‐2012 NCT00281697 Arm 1:55
Arm 2:49 		IV 0‐1 No 112 47 Bevacizumab plus taxane or gemcitabine or capecitabine or vinorelbine Placebo plus taxane or gemcitabine or capecitabine or vinorelbine OS, PFS, ORR 0.62 (0.39‐1.01) 0.49 (0.33‐0.74) 2.14 (1.14‐4.02) NA NA
Carey‐2012 NCT00232505 Arm 1:52
Arm 2:49 		IV 0‐2 and unknown No 71 31 Cetuximab and carboplatin Cetuximab OS, PFS, ORR 0.93 (0.57‐1.50) 0.49 (0.23‐0.64) 2.62 (0.62‐11.02) NA NA
Curigliano‐2013 NCT00246571 Arm 1:52
Arm 2:52 		Locally recurrent, IV 0‐1 and ≥2 No 113 104 Sunitinib Capecitabine or vinorelbine or docetaxel or paclitaxel or gemcitabine OS, PFS, ORR 1.16 (0.86‐1.56) 1.20 (0.89‐1.63) 0.39 (0.10‐1.49) NA NA
Finn‐2009 NCT00075270 NA III, IV 0‐2 and unknown No 71 60 Lapatinib and paclitaxel Placebo and paclitaxel PFS NA 1.25 (0.85‐1.83) NA NA NA
Forero‐2015 NCT01307891 Arm 1:51 Arm 2:50 IV NA No 39 21 Tigatuzumab and nab‐paclitaxel Nab‐paclitaxel PFS, ORR NA 1.27 (0.76‐2.19) 0.74 (0.35‐1.55) NA NA
Gonzalez‐2014 NCT00499603 Arm 1:46
Arm 2:52 		IIA‐IIIC NA Yes 23 27 Everolimus plus paclitaxel, 5‐fluorouracil, epirubicin, and cyclophosphamide Paclitaxel, 5‐fluorouracil, epirubicin, and cyclophosphamide ORR, pCR NA NA 0.76 (0.50‐1.16) 1.17 (0.48‐2.85) NA
Gray‐2009 NCT00028990 NA IV NA No arm1+arm2: 232 arm1+arm2: 232 Bevacizumab and paclitaxel Paclitaxel PFS NA 0.49 (0.34‐0.70) NA NA NA
Han‐2018 NCT01506609 NA Locally recurrent, IV 0‐2 No 40 42 Veliparib plus carboplatin and paclitaxel Placebo plus carboplatin and paclitaxel PFS NA 0.82 (0.47‐1.40) NA NA NA
Jovanovic‐2017 NCT00930930 Arm 1:52
Arm 2:52 		II, III NA Yes 96 49 Everolimus plus cisplatin and paclitaxel Placebo plus cisplatin and paclitaxel DFS, ORR, pCR NA NA 0.89 (0.77‐1.04) 0.74 (0.50‐1.10) NA
Kim‐2017 NCT02162719 Arm 1:54
Arm 2:53 		Locally recurrent, IV 0‐1 No 62 62 Ipatasertib and paclitaxel Placebo and paclitaxel PFS,ORR NA 0.60 (0.37‐0.98) 1.25 (0.78‐2.00) NA NA
Kummar‐2016 NCT01306032 54 IV 0‐1 No 21 18 Veliparib and cyclophosphamide Cyclophosphamide PFS, ORR NA 0.57 (0.21‐0.81) 1.71 (0.17‐17.38) NA NA
Llombart‐2015 NCT01204125 NA II‐IIIA 0‐1 Yes 94 47 Iniparib and paclitaxel Paclitaxel ORR, pCR NA NA 1.04 (0.78‐1.38) 0.85 (0.42‐1.71) 0.95 (0.48‐1.88)
Minckwitz‐2012 NCT00567554 Arm 1:49
Arm 2:48 		T1c‐T4d NA Yes 323 340 Bevacizumab plus epirubicin, cyclophosphamide, and docetaxel Epirubicin, cyclophosphamide, and docetaxel pCR NA NA NA 1.32 (1.08‐1.60) 1.28 (1.07‐1.54)
Nahleh‐2016 NCT00856492 NA IIB‐IIIC NA Yes 32 35 Bevacizumab plus nab‐paclitaxel, doxorubicin, cyclophosphamide, and pegfilgrastim Nab‐paclitaxel, doxorubicin, cyclophosphamide, and pegfilgrastim OS, PFS, pCR 0.49 (0.19‐1.29) 0.46 (0.20‐1.05) NA 2.08 (1.14‐3.78) NA
O'Shaughnessy‐2011 NCT00540358 Arm 1:56
Arm 2:53 		IV 0‐1 and unknown No 61 62 Iniparib plus gemcitabine and carboplatin Gemcitabine and carboplatin OS, PFS, ORR 0.57 (0.36‐0.90) 0.59 (0.39‐0.90) 1.63 (1.06‐2.51) NA NA
O'Shaughnessy‐2014 NCT00938652 Arm 1:53
Arm 2:54 		Locally recurrent, IV 0‐2 No 261 258 Iniparib plus gemcitabine and carboplatin Gemcitabine and carboplatin OS, PFS, ORR 0.85 (0.69‐1.04) 0.79 (0.65‐0.98) 1.12 (0.87‐1.43) NA NA
Pivot‐2011 NCT00333775 NA IV 0‐1 No 60 43 Bevacizumab and docetaxel Placebo and docetaxel PFS NA 0.68 (0.46‐1.00) NA NA NA
Robert (Cape)‐2011a NCT00262067 NA Locally recurrent, IV NA No 87 50 Bevacizumab and capecitabine Placebo and capecitabine PFS NA 0.72 (0.49‐1.06) NA NA NA
Robert (Tax/Anthra)‐2011a NCT00262067 NA IV NA No 96 46 Bevacizumab and standard chemotherapy Placebo and standard chemotherapy PFS NA 0.78 (0.53‐1.15) NA NA NA
Robson‐2017 NCT02000622 NA IV 0‐1 No 102 48 Olaparib Standard therapy PFS, ORR NA 0.43 (0.29‐0.63) 2.58 (1.30‐5.11) NA NA
Sikov (No carbo)‐2015b NCT00861705 NA II, III NA Yes 105 107 Bevacizumab plus paclitaxel, doxorubicin, and cyclophosphamide Paclitaxel, doxorubicin, and cyclophosphamide pCR NA NA NA 1.02 (0.74‐1.40) 1.20 (0.90‐1.61)
Sikov (With carbo)‐2015b NCT00861705 NA II, III NA Yes 110 111 Bevacizumab plus paclitaxel, doxorubicin, cyclophosphamide, and carboplatin Paclitaxel, doxorubicin, cyclophosphamide, and carboplatin pCR NA NA NA 1.34 (1.05‐1.71) 1.27 (1.02‐1.57)
Tredan‐2015 NCT00633464 Arm 1:50
Arm 2:53 		Locally recurrent, IV NA No 39 40 Cetuximab and ixabepilone Ixabepilone ORR NA NA 1.20 (0.64‐2.25) NA NA
Yardley‐2015 NCT01156753 NA III, IV 0‐3 No 28 11 Glembatumumab vedotin Standard chemotherapy OS, PFS, ORR 0.65 (0.29‐1.45) 0.69 (0.32‐1.54) 4.55 (0.27‐76.05) NA NA
Yardley‐2016 NCT01427933 NA III, IV 0‐1 and ≥2 No 21 22 Ramucirumab and eribulin Eribulin PFS NA 0.66 (0.32‐1.35) NA NA NA
Open in a new tab

HR for OS or PFS was presented as arm1 vs arm2, and HR < 1 indicated patients in arm1 achieved better OS or PFS than patients in arm2. RR for ORR or pCR was presented as arm1 vs arm2, and RR > 1 indicated patients in arm1 achieved more ORR or pCR than patients in arm2.

CI, confidence interval; CT, chemotherapy; NA, not available; ORR, objective response rate; OS, overall survival; pCR, pathological complete response; PFS, progression‐free survival.

a

Study Robert (Cape)‐2011 and study Robert (Tax/Anthra)‐2011 were from the same trial, and their CT was different, as shown in the table above.

b

Study Sikov (No carbo)‐2015 and study Sikov (With carbo)‐2015 were from the same trial, and their CT was different, as shown in the table above.

For Robert's study, there were two intervention groups and each group was composed of two treatment arms: one was bevacizumab + CT, and another was CT alone. The chemotherapy regimens in these two groups were entirely different, and the analysis was performed separately; therefore, we considered it to be two independent trials, as was also the case for Sikov's study. Regimens in seven trials9, 10, 42, 44, 47, 48 were neoadjuvant chemotherapies. Study endpoints included OS, PFS, ORR, and pCR. We did not analyze DFS since only Bell's study took it as a study endpoint.35, 57 Overall, the included RCTs were of high quality, but the methodological risk of bias assessment suggested a high risk of performance might exist, which was due to the absence of a placebo (Figure 2). Therefore, we defined the treatment CT + placebo and CT as different regimens.

Figure 2.

Figure 2

Open in a new tab

Methodological risk of bias assessment of the included studies in this network meta‐analysis. Methodological risk of bias contained the following part: random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessors (detection bias), incomplete outcome data (attrition bias), selective reporting (reporting bias), and other bias. Each part was evaluated as low risk of bias, unclear risk of bias, and high risk of bias. The length of the bar showed the percentage of total studies

3.2. Evidence network of the comparisons and comparison‐adjusted funnel plots

The network plots (Figure 3) displayed the comparisons of the eight treatments for OS, 16 treatments for PFS, 13 treatments for ORR, five treatments for pCR breast and axillary nodes, and three treatments for pCR breast. Bevacizumab + CT vs CT, iniparib + CT vs CT, and cetuximab + CT vs CT were the most frequent treatment regimens being directly compared. The majority of target therapies were combined with CT, yet cetuximab, glembatumumab vedotin, sunitinib, and olaparib were used as monotherapies. There were no closed triangular or quadratic loops formed by direct comparisons; consequently, loop‐specific heterogeneity estimates could not be calculated. The influence and contribution of each direct comparison to the whole network were presented in Figure S1, using weighted squares along with the respective percentages. (The contribution plot of PFS was provided in a table format thanks to the large‐scale data.).

Figure 3.

Figure 3

Open in a new tab

Network of the comparisons included in the network meta‐analysis. A‐E, Network of the comparisons for OS (A), PFS (B), ORR (C), pCR breast and axillary nodes (D), and pCR breast (E). Each node corresponded to a regimen included in the analysis, and each line represented direct comparisons between regimens, with node size and line width proportional to the number of trials directly comparing the regimens. (CT, chemotherapy; OS, overall survival; PFS, progression‐free survival; ORR, objective response rate; pCR, pathological complete response)

To assess the potential small‐study effects, the comparison‐adjusted funnel plots were applied. Since there was not a single reference line against which symmetry could be judged, we made the specific assumption that target therapies alone or in combination with CT were favored in small studies. As we can see in Figure S2, symmetrical comparison‐adjusted funnel plots suggested no small‐study effects existed.

3.3. Bayesian network meta‐analysis

First, we evaluated the fit of the fixed or random effects model according to DIC. For OS, DIC was 6.985 in the fixed effects model, a little bit larger than 6.773 in the random effects model. However, DIC was smaller in the fixed effects model for other study endpoints, with differences of no more than 2. The Brooks‐Gelman‐Rubin diagnostic test indicated better convergence of the fixed effects model. Besides, the high level of uncertainty of the posterior standard deviation showed that there was little information to inform the random effects parameters.24 Therefore, we adopted the fixed effects model for network meta‐analysis.

As shown in Figure 4A and Table 2A, iniparib + CT significantly improved OS compared with CT (0.79, 0.66‐0.96) or CT + placebo (0.55, 0.31‐0.97), while bevacizumab + CT significantly improved OS compared with CT + placebo (0.62, 0.38‐1.00). In accordance with the results of PFS, we found several targeted agents were beneficial to TNBC patients. As Figure 4B and Table 2B showed, cetuximab + CT (0.67, 0.47‐0.96), sorafenib + CT (0.44, 0.21‐0.91), bevacizumab + CT (0.48, 0.35‐0.65), veliparib + CT (0.58, 0.36‐0.94), iniparib + CT (0.75, 0.62‐0.90), ipatasertib + CT (0.44, 0.24‐0.81), and olaparib (0.43, 0.29‐0.64) all significantly improved PFS compared with CT. Moreover, when in comparison with CT + placebo, bevacizumab + CT (0.66, 0.55‐0.80), ipatasertib + CT (0.60, 0.37‐0.98), and olaparib (0.59, 0.35‐0.99) showed a distinct improvement. However, it was noteworthy that iniparib + CT decreased PFS compared with bevacizumab + CT (1.55, 1.08‐2.22), suggesting that bevacizumab was superior. As for ORR, only bevacizumab + CT and olaparib showed benefits. Bevacizumab + CT significantly improved ORR compared with CT + placebo (2.15, 1.16‐4.05) and olaparib significantly improved ORR compared with CT (2.57, 1.31‐5.09; Figure 4C and Table 2C). Furthermore, bevacizumab + CT was also effective in both pCR breast and axillary nodes and pCR breast. As shown in Figure 4D,E and Table 2D‐E, bevacizumab + CT significantly improved pCR breast and axillary nodes (1.30, 1.13‐1.49) and pCR breast (1.26, 1.11‐1.44) when compared with CT.

Figure 4.

Figure 4

Open in a new tab

Forest plot of the estimated HR and RR for different target therapies compared with CT in the fixed effects network meta‐analysis. A and B, Forest plot of the estimated HR for different target therapies compared with CT in the fixed effects network meta‐analysis for OS (A) and PFS (B). HR and its 95% CrI <1 favored target therapies otherwise favored CT. C‐E, Forest plot of the estimated RR for different target therapies compared with CT in the fixed effects network meta‐analysis for ORR (C), pCR breast and axillary nodes (D), and pCR breast (E). RR and its 95% CrI>1 favored target therapies otherwise favored CT. (CT, chemotherapy; OS, overall survival; PFS, progression‐free survival; ORR, objective response rate; pCR, pathological complete response; CrI, credible interval)

Table 2.

(A) Estimated HR between all treatments for OS in the fixed effects network meta‐analysis. (B) Estimated HR between all treatments for PFS in the fixed effects network meta‐analysis. (C) Estimated RR between all treatments for ORR in the fixed effects network meta‐analysis. (D) Estimated RR between all treatments for pCR breast and axillary in the fixed effects network meta‐analysis. (E) Estimated RR between all treatments for pCR breast in the fixed effects network meta‐analysis

A
CT 1.45 (0.86‐2.45) 0.82 (0.56‐1.20) 0.90 (0.72‐1.13) 1.16 (0.86‐1.56) 0.79 (0.66‐0.96) 0.88 (0.48‐1.63) 0.65 (0.29‐1.46)
CT + Placebo 0.57 (0.30‐1.10) 0.62 (0.38‐1.00) 0.80 (0.43‐1.48) 0.55 (0.31‐0.97) 0.61 (0.27‐1.38) 0.45 (0.17‐1.18)
Cetuximab + CT 1.10 (0.70‐1.71) 1.42 (0.87‐2.31) 0.97 (0.64‐1.49) 1.08 (0.66‐1.74) 0.79 (0.33‐1.95)
Bevacizumab + CT 1.29 (0.89‐1.88) 0.89 (0.66‐1.19) 0.98 (0.51‐1.90) 0.72 (0.31‐1.69)
Sunitinib 0.69 (0.48‐0.98) 0.76 (0.38‐1.52) 0.56 (0.24‐1.34)
Iniparib + CT 1.11 (0.58‐2.11) 0.81 (0.36‐1.90)
Cetuximab 0.74 (0.27‐2.05)
Glembatumumab vedotin
B
CT 0.73 (0.51‐1.03) 0.67 (0.47‐0.96) 0.44 (0.21‐0.91) 1.03 (0.65‐1.65) 0.48 (0.35‐0.65) 0.91 (0.54‐1.53) 1.27 (0.75‐2.16) 0.58 (0.36‐0.94) 0.75 (0.62‐0.90) 0.66 (0.32‐1.37) 1.20 (0.88‐1.62) 0.69 (0.32‐1.51) 1.37 (0.73‐2.54) 0.44 (0.24‐0.81) 0.43 (0.29‐0.64)
CT + Placebo 0.92 (0.56‐1.51) 0.60 (0.31‐1.15) 1.41 (0.80‐2.54) 0.66 (0.55‐0.80) 1.25 (0.86‐1.84) 1.74 (0.92‐3.30) 0.80 (0.51‐1.25) 1.03 (0.69‐1.53) 0.90 (0.41‐2.05) 1.65 (1.04‐2.62) 0.95 (0.40‐2.24) 1.89 (0.91‐3.83) 0.60 (0.37‐0.98) 0.59 (0.35‐0.99)
Cetuximab + CT 0.65 (0.28‐1.47) 1.53 (0.86‐2.77) 0.72 (0.44‐1.15) 1.36 (0.72‐2.54) 1.89 (1.00‐3.61) 0.87 (0.48‐1.58) 1.11 (0.74‐1.67) 0.98 (0.45‐2.23) 1.78 (1.11‐2.85) 1.03 (0.44‐2.45) 2.04 (1.22‐3.39) 0.65 (0.32‐1.33) 0.64 (0.38‐1.09)
Sorafenib + CT 2.36 (0.97‐5.64) 1.10 (0.56‐2.17) 2.09 (0.97‐4.44) 2.91 (1.16‐7.27) 1.34 (0.60‐2.94) 1.71 (0.78‐3.68) 1.51 (0.54‐4.25) 2.74 (1.22‐6.12) 1.58 (0.53‐4.65) 3.14 (1.16‐8.23) 1.00 (0.44‐2.27) 0.99 (0.42‐2.29)
Sunitinib + CT 0.47 (0.27‐0.81) 0.88 (0.44‐1.76) 1.23 (0.61‐2.48) 0.57 (0.29‐1.10) 0.72 (0.44‐1.19) 0.64 (0.28‐1.51) 1.16 (0.66‐2.01) 0.67 (0.27‐1.68) 1.33 (0.60‐2.87) 0.42 (0.20‐0.91) 0.42 (0.23‐0.76)
Bevacizumab + CT 1.89 (1.24‐2.90) 2.63 (1.42‐4.86) 1.21 (0.77‐1.91) 1.55 (1.08‐2.22) 1.36 (0.63‐3.02) 2.49 (1.61‐3.82) 1.43 (0.62‐3.34) 2.85 (1.41‐5.67) 0.90 (0.54‐1.54) 0.89 (0.55‐1.46)
Lapatinib + CT 1.39 (0.67‐2.92) 0.64 (0.36‐1.16) 0.82 (0.47‐1.42) 0.72 (0.30‐1.79) 1.32 (0.72‐2.40) 0.76 (0.30‐1.96) 1.50 (0.65‐3.38) 0.48 (0.26‐0.90) 0.47 (0.25‐0.91)
Tigatuzumab + CT 0.46 (0.22‐0.94) 0.59 (0.33‐1.03) 0.52 (0.21‐1.28) 0.94 (0.51‐1.74) 0.54 (0.21‐1.41) 1.08 (0.47‐2.42) 0.34 (0.15‐0.77) 0.34 (0.17‐0.66)
Veliparib + CT 1.27 (0.76‐2.12) 1.12 (0.48‐2.69) 2.04 (1.16‐3.59) 1.18 (0.47‐2.98) 2.34 (1.05‐5.09) 0.75 (0.38‐1.46) 0.73 (0.40‐1.36)
Iniparib + CT 0.88 (0.42‐1.88) 1.60 (1.12‐2.29) 0.92 (0.42‐2.07) 1.83 (0.95‐3.51) 0.58 (0.31‐1.11) 0.58 (0.38‐0.89)
Ramucirumab + CT 1.81 (0.81‐3.98) 1.04 (0.35‐3.03) 2.07 (0.79‐5.38) 0.66 (0.26‐1.69) 0.65 (0.29‐1.48)
Sunitinib 0.57 (0.25‐1.34) 1.14 (0.57‐2.29) 0.36 (0.18‐0.72) 0.36 (0.22‐0.59)
Glembatumumab vedotin 1.98 (0.72‐5.45) 0.63 (0.23‐1.69) 0.62 (0.26‐1.51)
Cetuximab 0.32 (0.13‐0.76) 0.31 (0.15‐0.66)
Ipatasertib + CT 0.99 (0.48‐2.03)
Olaparib
C
CT 0.86 (0.55‐1.34) 1.42 (0.86‐2.36) 1.83 (0.85‐3.93) 0.74 (0.36‐1.57) 0.76 (0.50‐1.17) 1.72 (0.18‐17.57) 1.15 (0.97‐1.37) 0.40 (0.10‐1.50) 4.48 (0.27‐77.75) 0.54 (0.12‐2.48) 1.07 (0.55‐2.03) 2.57 (1.31‐5.09)
CT + Placebo 1.66 (0.85‐3.28) 2.15 (1.16‐4.05) 0.86 (0.36‐2.04) 0.89 (0.77‐1.04) 2.02 (0.18‐21.25) 1.35 (0.83‐2.18) 0.46 (0.11‐1.87) 5.29 (0.30‐93.78) 0.63 (0.13‐3.09) 1.25 (0.77‐2.01) 2.99 (1.34‐6.85)
Cetuximab + CT 1.29 (0.51‐3.25) 0.52 (0.21‐1.27) 0.54 (0.28‐1.04) 1.21 (0.11‐13.09) 0.81 (0.48‐1.38) 0.28 (0.07‐1.16) 3.18 (0.18‐55.95) 0.38 (0.09‐1.60) 0.75 (0.33‐1.72) 1.80 (0.78‐4.22)
Bevacizumab + CT 0.40 (0.14‐1.17) 0.42 (0.22‐0.80) 0.94 (0.08‐10.74) 0.63 (0.28‐1.39) 0.22 (0.05‐1.01) 2.49 (0.13‐46.65) 0.29 (0.05‐1.61) 0.58 (0.27‐1.28) 1.41 (0.50‐4.02)
Tigatuzumab + CT 1.03 (0.45‐2.41) 2.33 (0.19‐26.60) 1.56 (0.73‐3.34) 0.54 (0.12‐2.47) 6.19 (0.33‐114.40) 0.73 (0.13‐3.93) 1.45 (0.54‐3.86) 3.47 (1.27‐9.63)
Everolimus + CT 2.26 (0.20‐23.60) 1.51 (0.95‐2.39) 0.52 (0.13‐2.10) 5.94 (0.34‐104.10) 0.71 (0.14‐3.45) 1.40 (0.85‐2.30) 3.36 (1.52‐7.52)
Veliparib + CT 0.69 (0.07‐7.18) 0.24 (0.02‐3.55) 2.70 (0.07‐103.70) 0.32 (0.02‐5.36) 0.63 (0.06‐7.22) 1.53 (0.14‐17.69)
Iniparib + CT 0.34 (0.09‐1.31) 3.90 (0.23‐67.76) 0.47 (0.10‐2.16) 0.92 (0.47‐1.82) 2.23 (1.10‐4.51)
Sunitinib 11.74 (0.49‐256.70) 1.37 (0.18‐10.75) 2.74 (0.62‐12.02) 6.53 (1.47‐29.56)
Glembatumumab vedotin 0.12 (0.01‐3.23) 0.24 (0.01‐4.53) 0.58 (0.03‐10.93)
Cetuximab 1.99 (0.38‐10.44) 4.77 (0.89‐25.71)
Ipatasertib + CT 2.41 (0.94‐6.24)
Olaparib
D
CT 1.58 (0.60‐4.20) 1.18 (0.48‐2.86) 0.85 (0.42‐1.67) 1.30 (1.13‐1.49)
CT + Placebo 0.74 (0.50‐1.10) 0.54 (0.17‐1.80) 0.82 (0.31‐2.18)
Everolimus + CT 0.73 (0.24‐2.25) 1.10 (0.44‐2.68)
Iniparib + CT 1.52 (0.75‐3.09)
Bevacizumab + CT
E
CT 0.95 (0.48‐1.86) 1.26 (1.11‐1.44)
Iniparib + CT 1.33 (0.66‐2.70)
Bevacizumab + CT
Open in a new tab

Estimated HR (for OS and PFS) or RR (for ORR and pCR) and its 95% CrI between all treatments were shown in each cell. The column treatment was compared with the row treatment. A‐B. HR < 1 indicated patients in the column treatment group achieved better OS/PFS than patients in the row treatment group, and the numbers were blue if the Bayesian P value < 0.05. HR > 1 indicated patients in the column treatment group achieved worse OS/PFS than patients in the row group, and the numbers were red if the Bayesian P value < 0.05. C‐E. RR > 1 indicated patients in the column treatment group achieved better ORR/pCR than patients in the row treatment group, and the numbers were blue if the Bayesian P value < 0.05. RR < 1 indicated patients in the column treatment group achieved worse ORR/pCR than patients in the row group, and the numbers were red if the Bayesian P value < 0.05.

CI, confidence interval; CT, chemotherapy; ORR, objective response rate; OS, overall survival; pCR, pathological complete response; PFS, progression‐free survival.

The probability of rank for each treatment regimen was also calculated (Figure S3). Nevertheless, ranking of treatment regimens based merely on the probability of being the best should be avoided, since this does not account for the indeterminacy in the relative treatment effects and could illogically give higher ranks to treatment regimens for which little evidence was available.23 Therefore, we selected the cumulative probability rankograms (Figure S4) and calculated SUCRA, which was a plain transformation of the mean rank and accounted for both the location and the variance of the relative treatment regimen effects.27

In terms of OS, glembatumumab vedotin ranked first (SUCRA: 79.8%) and iniparib + CT ranked second (74.9%), yet CT + placebo had the lowest SUCRA of 8.0% and sunitinib had the second lowest SUCRA (Figure 5A and Table S1A). As for PFS, we found olaparib, ipatasertib + CT, and sorafenib + CT were in the top three with SUCRAs of 87.6%, 85.5%, and 83.8%, respectively, yet cetuximab, tigatuzumab + CT, and sunitinib had the lowest SUCRAs, even worse than CT alone (Figure 5B and Table S1B). In terms of ORR, olaparib showed the best efficacy again with SUCRA of 87.8% and glembatumumab vedotin (82.0%) and bevacizumab + CT (77.7%) came in second and third while sunitinib had the lowest mean rank (Figure 5C and Table S1C). Yet, for pCR breast and axillary nodes, targeted therapies did not show superiority against CT + placebo, which had the highest SUCRA of 81.3% while bevacizumab + CT only ranked second (Figure 5D and Table S1D). With regard to pCR breast, bevacizumab + CT ranked first (89.5%), and iniparib + CT, and CT ranked second and third sequentially (Figure 5E and Table S1E).

Figure 5.

Figure 5

Open in a new tab

SUCRA for all treatments in the network meta‐analysis. A‐E, SUCRA histogram of different treatments for OS (A), PFS (B), ORR (C), pCR breast and axillary nodes (D), and pCR breast (E). (CT, chemotherapy; OS, overall survival; PFS, progression‐free survival; ORR, objective response rate; pCR, pathological complete response; Pla, placebo; Cetu, cetuximab; Beva, bevacizumab; Suni, sunitinib; Ini, iniparib; Glemba, glembatumumab vedotin; Sora, sorafenib; Lapa, lapatinib; Tiga, tigatuzumab; Veli, veliparib; Ramu, ramucirumab; Ipata, ipatasertib; Ola, olaparib; everolimus)

3.4. Traditional pairwise meta‐analysis of direct comparisons

A consistency test was conducted by comparing the estimated effect sizes from the Bayesian network meta‐analysis with the pooled effect sizes from the traditional pairwise meta‐analysis. Only direct comparisons reported at least twice were considered. The fixed effects model was adopted, since no significant heterogeneity was observed according to the I 2 test. As we expected, the results of the traditional pairwise meta‐analysis were highly consistent with the results of direct comparisons in the Bayesian network meta‐analysis (Figure S5).

3.5. Sensitivity analysis based on the frequentist graph‐theoretical model

We also performed network meta‐analysis based on the frequentist graph‐theoretical model for sensitivity analysis.29 The fixed effects model was adopted, consistent with previous Bayesian network meta‐analysis. Forest plots of the efficacy of treatment regimens against CT in terms of OS, PFS, ORR, and pCR are presented in Figure S6, and the results are also highly consistent with the Bayesian network meta‐analyses in Figure 4.

Like the SUCRA value in the Bayesian network meta‐analysis, the P‐score was used for ranking treatments in the frequentist graph‐theoretical meta‐analysis.31 In terms of OS, glembatumumab vedotin had the highest P‐score of 0.7981 and iniparib + CT had the second highest P‐score of 0.7496, yet CT + placebo had the lowest P‐score of 0.0813 and sunitinib had the second lowest P‐score of 0.2035 (Table S2A). As for PFS, olaparib, ipatasertib + CT, and sorafenib + CT had P‐scores of 0.8764, 0.8564, and 0.8370, respectively, ranking as the top three, yet cetuximab had the lowest P‐score of 0.1083 and tigatuzumab + CT had the second lowest P‐score of 0.1309 (Table S2B). In terms of ORR, olaparib showed the highest P‐score of 0.8778 not surpirsingly, yet sunitinib had the lowest P‐score of 0.1175 (Table S2C). However, in terms of pCR breast and axillary nodes, targeted therapies or combined therapies did not show superiority against CT + placebo, and CT + placebo had the highest P‐score of 0.8130, similar to the results of SUCRA (Table S2D). As for pCR breast, bevacizumab + CT had the highest P‐score of 0.8944 with iniparib + CT and CT following behind (Table S2E). All rankings based on P‐scores in the frequentist graph‐theoretical meta‐analysis were consistent with SUCRA values in the Bayesian network meta‐analysis, which confirmed the stability of our results.

4. DISCUSSION

Triple‐negative breast cancer is quite a complex disease. Its relative aggressiveness and impressive heterogeneity have become challenges that clinicians face in making strides against this disease.58 Although several large‐scale RCTs have shown the clinical efficacies of targeted therapies, comparisons between those treatments remain uninvestigated. Therefore, we conducted a network meta‐analysis to compare the efficacy of various targeted agents in the treatment of all‐stage TNBC.

Our results suggested that the PARP inhibitor‐olaparib was the best for PFS and ORR and patients can benefit from bevacizumab + CT with respect to OS, PFS, ORR, and pCR compared with CT with or without placebo. In addition, our study also confirmed the role of some other targeted agents such as ipatasertib, cetuximab, iniparib, and sorafenib in improving survival outcomes. For example, iniparib + CT had a better effect in prolongation of OS than CT + placebo while sorafenib + CT, cetuximab + CT, and iniparib + CT acquired better PFS in comparison with CT alone. These results were quite distinct from the previous network meta‐analysis for targeted treatment of advanced TNBC.59 Ge et al59 found that only bevacizumab + CT effectively improved PFS while no significant differences in OS, PFS, and ORR were found between other agents and CT. Several distinctions existed between our studies. Ge et al focused on advanced/metastatic TNBC and categorized both CT and CT + placebo as CT, while we analyzed all strategies for all‐stage TNBC and made a distinction between CT alone and CT + placebo. Despite the placebo being considered ineffective, we found that the studies whose control group was treated with CT + placebo were double‐blinded with lower performance bias, while other studies were regarded as open‐label studies.

The rankings of treatments were made as well in our study. According to the results of treatment rank probabilities, we found that olaparib provided a best benefit over all other therapies. It is well‐known that breast cancer patients who carry germline mutations in either BRCA1 or BRCA2 are sensitive to PARP inhibitors.60 Although the OlympiAD trial only included patients with BRCA mutations, it was noteworthy that TNBC patients benefited more than those who were HR‐positive maybe due to the similarities in the gene‐expression profiles of BRCA1‐deficient breast cancers and sporadic TNBCs.49, 55 In addition, veliparib or iniparib + CT also showed superiority in PFS. Consequently, we consider that there is a strong rationale for treating TNBC with PARP inhibitors.

Bevacizumab is the most widely studied agent. A recent conventional meta‐analysis of additional bevacizumab as neoadjuvant therapy showed statistically significant improvements in pCR of TNBC patients,61 consistent with our results. Different chemotherapeutic agents all showed a greater benefit of survival outcomes or ORR when combined with chemotherapy.37, 43, 48, 51 Our results provided additional evidence that bevacizumab may be a viable treatment for TNBC. On the contrary, ramucirumab, another VEGFR‐2 antagonist, did not significantly improve the PFS of TNBC patients.54 It seems that antiangiogenic agents in the treatment of TNBC remain controversial.

Glembatumumab vedotin was found to have the highest SUCRA in OS, which meant it had greater probability of being effective. However, the statistical power was insufficient maybe due to the limited number of included studies and sample sizes. Glembatumumab vedotin is an ADC drug against the tumor‐associated antigen gpNMB. Previously, noteworthy activity was observed in glembatumumab vedotin‐treated TNBC patients, with a possible improvement in OS and PFS.15 The further subgroup analysis indicated that glembatumumab vedotin significantly improved OS and PFS for tumor gpNMB‐overexpressed TNBC patients.15 For those gpNMB‐overexpressed patients, glembatumumab vedotin may become a promising targeted therapy,62 and we expect more clinical trials carried out to confirm it.

Apart from the treatments mentioned above, immune checkpoint inhibition has received much attention in recent years. Previous studies reported that programmed cell death ligand 1 (PD‐L1) was expressed in 20% of TNBC63 and high PD‐L1 expression contributed to poor prognosis,64 suggesting the PD‐L1/PD‐1 pathway blockade as a highly promising therapy. Pembrolizumab is under study for treatment of TNBC nowadays. In the phase Ib KEYNOTE‐012 trial, pembrolizumab demonstrated an ORR of 18.5% with a median time to response of 17.9 weeks.14 In spite of the low response rate, a subset of patients showed long‐lasting responses,14 supporting further development of pembrolizumab for the treatment of TNBC.

The limitations of this study also need to be acknowledged. First of all, there were no closed loops in our network, thus, evaluation of inconsistency could not be conducted. Heterogeneity between included trials should not be ignored, since the disease stage of patients included in the present analysis was heterogenous and data did not allow us to perform subgroup analysis. Meanwhile, different study designs (double‐blinded or not) and treatment settings (neoadjuvant, adjuvant, or metastasis therapy) were both causes of heterogeneity. Second, the number of included studies was relatively small. For some agents, such as sorafenib and lapatinib, only one RCT was identified and the sample size was small. More than half of the studies were open‐label, allowing a high risk of performance bias. Maybe more blinded RCTs are needed since not entirely identical results were observed for CT + placebo and CT. Additionally, we did not distinguish the specific composition of chemotherapy across studies because different chemotherapy strategies were chosen in various trials. TNBC patients may show distinct sensitivity to different chemotherapeutic agents. For instance, a recent meta‐analysis reported significant pCR rates increase of platinum‐based neoadjuvant chemotherapy for TNBC65 while in our study, Baselga et al,33 Carey et al,38 and O'Shaughnessy et al49 used cisplatin or carboplatin as a regimen. Finally, although Bayesian network meta‐analysis is a potential solution enabling indirect comparison when a head‐to‐head trial is not available and making it realizable to rank several treatments by combining direct and indirect comparisons, it only pools the ratios rather than differences between treatments in each study. Perhaps that's why glembatumumab vedotin has great probabilities to rank first in OS and second in ORR but make no significant differences compared with CT or CT + Placebo. Although this is not the first network meta‐analysis of targeted agents for TNBC, our study may show different results, indicating that more targeted agents have potential efficacy for TNBC patients.

Nowadays, with improving understanding of biological complexity and diversity of TNBC, six different molecular subtypes were identified,66 which helped to develop molecular drivers that can be therapeutically targeted. In view of the limited clinical benefit with targeted agents in unselected TNBC patients of some trials, Bareche et al67 unraveled molecular heterogeneity of six subtypes by using an integrative genomic analysis. The researchers found specific differences in mutational and copy number profiles characterizing each subtype and offered novel therapeutic avenues for these patients.67 For example, basal‐like 1 (BL1) subset characterizing with high genomic instability for BRCA1/2 may be sensitive to PARP inhibitors while mesenchymal (M) subtype tumors with high level of EGFR mRNA expression may be potential candidates to anti‐VEGF inhibitors and EGFR inhibitors.67 Thus, the application of a comprehensive genomic characterization may help to select the most appropriate treatment in TNBC and individualized treatment for TNBC patients may be extremely necessary.67, 68 Combined with our findings, we assumed that bevacizumab may be more suitable for M subtype TNBC and olaparib, veliparib could be used in BL1 subset.

In conclusion, our network meta‐analysis suggested that addition of bevacizumab to chemotherapy could be recommended to treat TNBC patients. Olaparib had great effects in PFS and ORR especially for those with BRCA mutations. Glembatumumab vedotin ranked first in OS, without statistically significant differences compared with CT alone, but it may work well while gpNMB is overexpressed. When in combination with CT, some other targeted agents such as ipatasertib, cetuximab, sorafenib, veliparib, and iniparib showed significant superiority in PFS, exhibiting therapeutic value to some extent. Without any benefits from sunitinib, cetuximab, and tigatuzumab, we advise not to prescribe them to TNBC patients. Based on the impressive heterogeneity of TNBC, treatment personalization deserves to be promoted. Multiple targeted agents for TNBC are still at various stages of development and we expect more advances will be made in the near future.

CONFLICT OF INTEREST

None declared.

Supporting information

 

Click here for additional data file. (19.4MB, eps)

 

Click here for additional data file. (17.3MB, eps)

 

Click here for additional data file. (5.3MB, eps)

 

Click here for additional data file. (5.5MB, eps)

 

Click here for additional data file. (13.4MB, eps)

 

Click here for additional data file. (11.2MB, eps)

 

Click here for additional data file. (123.5KB, doc)

ACKNOWLEDGMENTS

We acknowledge Dr. Yunxian Yu of Department of Epidemiology & Health Statistics, Zhejiang University School of Medicine for helping to solve some statistical problems.

Chen H, Lu W, Zhang Y, Zhu X, Zhou J, Chen Y. A Bayesian network meta‐analysis of the efficacy of targeted therapies and chemotherapy for treatment of triple‐negative breast cancer. Cancer Med. 2019;8:383–399. 10.1002/cam4.1892

Huihui Chen and Wei Lu contributed equally to this work.

Funding information

This study was supported by the funding of the Key Program of the Natural Science Foundation of Zhejiang Province (Grant No. LZ16H160002) and the Zhejiang Provincial Program for the Cultivation of High‐level Innovative Health Talents.

REFERENCES

  • 1. Harbeck N, Gnant M. Breast cancer. Lancet. 2017;389(10074):1134‐1150. [DOI] [PubMed] [Google Scholar]
  • 2. Ferlay J, Soerjomataram I, Dikshit R, et al. Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012. Int J Cancer. 2015;136(5):E359‐E386. [DOI] [PubMed] [Google Scholar]
  • 3. Gluz O, Liedtke C, Gottschalk N, Pusztai L, Nitz U, Harbeck N. Triple‐negative breast cancer–current status and future directions. Ann Oncol. 2009;20(12):1913‐1927. [DOI] [PubMed] [Google Scholar]
  • 4. Sorlie T, Perou CM, Tibshirani R, et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci USA. 2001;98(19):10869‐10874. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5. Denkert C, Liedtke C, Tutt A, von Minckwitz G. Molecular alterations in triple‐negative breast cancer‐the road to new treatment strategies. Lancet. 2017;389(10087):2430‐2442. [DOI] [PubMed] [Google Scholar]
  • 6. von Minckwitz G, Untch M, Blohmer JU, et al. Definition and impact of pathologic complete response on prognosis after neoadjuvant chemotherapy in various intrinsic breast cancer subtypes. J Clin Oncol. 2012;30(15):1796‐1804. [DOI] [PubMed] [Google Scholar]
  • 7. Liedtke C, Mazouni C, Hess KR, et al. Response to neoadjuvant therapy and long‐term survival in patients with triple‐negative breast cancer. J Clin Oncol. 2008;26(8):1275‐1281. [DOI] [PubMed] [Google Scholar]
  • 8. Bayraktar S, Gluck S. Molecularly targeted therapies for metastatic triple‐negative breast cancer. Breast Cancer Res Treat. 2013;138(1):21‐35. [DOI] [PubMed] [Google Scholar]
  • 9. Minckwitz G, Eidtmann H, Rezai M, et al. Neoadjuvant chemotherapy and bevacizumab for HER2‐negative breast cancer. The New England Journal of Medicine. 2012;366:299‐309. [DOI] [PubMed] [Google Scholar]
  • 10. Sikov WM, Berry DA, Perou CM, et al. Impact of the addition of carboplatin and/or bevacizumab to neoadjuvant once‐per‐week paclitaxel followed by dose‐dense doxorubicin and cyclophosphamide on pathologic complete response rates in stage II to III triple‐negative breast cancer: CALGB 40603 (Alliance). J Clin Oncol. 2015;33(1):13‐21. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Lord CJ, Tutt AN, Ashworth A. Synthetic lethality and cancer therapy: lessons learned from the development of PARP inhibitors. Annu Rev Med. 2015;66:455‐470. [DOI] [PubMed] [Google Scholar]
  • 12. Foulkes WD, Stefansson IM, Chappuis PO, et al. Germline BRCA1 mutations and a basal epithelial phenotype in breast cancer. J Natl Cancer Inst. 2003;95(19):1482‐1485. [DOI] [PubMed] [Google Scholar]
  • 13. Rugo HS, Olopade OI, DeMichele A, et al. Adaptive randomization of veliparib‐carboplatin treatment in breast cancer. N Engl J Med. 2016;375(1):23‐34. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Nanda R, Chow LQ, Dees EC, et al. Pembrolizumab in patients with advanced triple‐negative breast cancer: phase Ib KEYNOTE‐012 study. J Clin Oncol. 2016;34(21):2460‐2467. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. Yardley DA, Weaver R, Melisko ME, et al. EMERGE: A randomized phase II study of the antibody‐drug conjugate glembatumumab vedotin in advanced glycoprotein NMB ‐ expressing breast cancer. J Clin Oncol. 2015;33(14):1609‐1619. [DOI] [PubMed] [Google Scholar]
  • 16. Clark O, Botrel TE, Paladini L, Ferreira MB. Targeted therapy in triple‐negative metastatic breast cancer: a systematic review and meta‐analysis. Core Evid. 2014;9:1‐11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Bramati A, Girelli S, Torri V, et al. Efficacy of biological agents in metastatic triple‐negative breast cancer. Cancer Treat Rev. 2014;40(5):605‐613. [DOI] [PubMed] [Google Scholar]
  • 18. Bafeta A, Trinquart L, Seror R, Ravaud P. Reporting of results from network meta‐analyses: methodological systematic review. BMJ. 2014;348:g1741. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Caldwell DM, Ades AE, Higgins JP. Simultaneous comparison of multiple treatments: combining direct and indirect evidence. BMJ. 2005;331(7521):897‐900. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Hutton B, Salanti G, Caldwell DM, et al. The PRISMA extension statement for reporting of systematic reviews incorporating network meta‐analyses of health care interventions: checklist and explanations. Ann Intern Med. 2015;162(11):777‐784. [DOI] [PubMed] [Google Scholar]
  • 21. Guyot P, Ades AE, Ouwens MJ, Welton NJ. Enhanced secondary analysis of survival data: reconstructing the data from published Kaplan‐Meier survival curves. BMC Med Res Methodol. 2012;12:9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Higgins J, Altman DG, Gotzsche PC, et al. The Cochrane Collaboration's tool for assessing risk of bias in randomised trials. BMJ. 2011;343:d5928. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Chaimani A, Higgins J, Mavridis D, Spyridonos P, Salanti G. Graphical tools for network meta‐analysis in STATA. PLoS ONE. 2013;8(10):e76654. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24. Woods BS, Hawkins N, Scott DA. Network meta‐analysis on the log‐hazard scale, combining count and hazard ratio statistics accounting for multi‐arm trials: a tutorial. BMC Med Res Methodol. 2010;10:54. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25. Spiegelhalter DJ, Best NG, Carlin BP, Van Der Linde A. Bayesian measures of model complexity and fit. J R Stat Soc Series B Stat Methodol. 2002;64(4):583‐639. [Google Scholar]
  • 26. Brooks SP, Gelman A. General methods for monitoring convergence of iterative simulations. J Comput Graph Stat. 1998;7(4):434‐455. [Google Scholar]
  • 27. Salanti G, Ades AE, Ioannidis JP. Graphical methods and numerical summaries for presenting results from multiple‐treatment meta‐analysis: an overview and tutorial. J Clin Epidemiol. 2011;64(2):163‐171. [DOI] [PubMed] [Google Scholar]
  • 28. Higgins J, Thompson SG. Quantifying heterogeneity in a meta‐analysis. Stat Med. 2002;21(11):1539‐1558. [DOI] [PubMed] [Google Scholar]
  • 29. Rucker G. Network meta‐analysis, electrical networks and graph theory. Res Synth Methods. 2012;3(4):312‐324. [DOI] [PubMed] [Google Scholar]
  • 30. netmeta: Network Meta‐Analysis using Frequentist Methods. R package version 0.8‐0. http://cran.at.r-project.org/web/packages/netmeta/. Accessed June 13, 2018.
  • 31. Rucker G, Schwarzer G. Ranking treatments in frequentist network meta‐analysis works without resampling methods. BMC Med Res Methodol. 2015;15:58. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32. Dieras V, Campone M, Yardley DA, et al. Randomized, phase II, placebo‐controlled trial of onartuzumab and/or bevacizumab in combination with weekly paclitaxel in patients with metastatic triple‐negative breast cancer. Ann Oncol. 2015;26(9):1904‐1910. [DOI] [PubMed] [Google Scholar]
  • 33. Baselga J, Gomez P, Greil R, et al. Randomized phase II study of the anti‐epidermal growth factor receptor monoclonal antibody cetuximab with cisplatin versus cisplatin alone in patients with metastatic triple‐negative breast cancer. J Clin Oncol. 2013;31(20):2586‐2592. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34. Baselga J, Segalla JG, Roche H, et al. Sorafenib in combination with capecitabine: an oral regimen for patients with HER2‐negative locally advanced or metastatic breast cancer. J Clin Oncol. 2012;30(13):1484‐1491. [DOI] [PubMed] [Google Scholar]
  • 35. Bell R, Brown J, Parmar M, et al. Final efficacy and updated safety results of the randomized phase III BEATRICE trial evaluating adjuvant bevacizumab‐containing therapy in triple‐negative early breast cancer. Ann Oncol. 2017;28(4):754‐760. [DOI] [PubMed] [Google Scholar]
  • 36. Bergh J, Bondarenko IM, Lichinitser MR, et al. First‐line treatment of advanced breast cancer with sunitinib in combination with docetaxel versus docetaxel alone: results of a prospective, randomized phase III study. J Clin Oncol. 2012;30(9):921‐929. [DOI] [PubMed] [Google Scholar]
  • 37. Brufsky A, Valero V, Tiangco B, et al. Second‐line bevacizumab‐containing therapy in patients with triple‐negative breast cancer: Subgroup analysis of the RIBBON‐2 trial. Breast Cancer Res Treat. 2012;133(3):1067‐1075. [DOI] [PubMed] [Google Scholar]
  • 38. Carey LA, Rugo HS, Marcom PK, et al. TBCRC 001: Randomized phase II study of cetuximab in combination with carboplatin in stage IV triple‐negative breast cancer. J Clin Oncol. 2012;30(21):2615‐2623. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39. Curigliano G, Pivot X, Cortes J, et al. Randomized phase II study of sunitinib versus standard of care for patients with previously treated advanced triple‐negative breast cancer. Breast. 2013;22(5):650‐656. [DOI] [PubMed] [Google Scholar]
  • 40. Finn RS, Press MF, Dering J, et al. Estrogen receptor, progesterone receptor, human epidermal growth factor receptor 2 (HER2), and epidermal growth factor receptor expression and benefit from lapatinib in a randomized trial of paclitaxel with lapatinib or placebo as first‐line treatment in HER2‐negative or unknown metastatic breast cancer. J Clin Oncol. 2009;27(24):3908‐3915. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41. Forero‐Torres A, Varley KE, Abramson VG, et al. TBCRC 019: a phase II trial of nanoparticle albumin‐bound paclitaxel with or without the anti‐death receptor 5 monoclonal antibody tigatuzumab in patients with triple‐negative breast cancer. Clin Cancer Res. 2015;21;2722‐2729. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42. Gonzalez‐Angulo AM, Akcakanat A, Liu S, et al. Open‐label randomized clinical trial of standard neoadjuvant chemotherapy with paclitaxel followed by FEC versus the combination of paclitaxel and everolimus followed by FEC in women with triple receptor‐negative breast cancer. Ann Oncol. 2014;25(6):1122‐1127. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43. Gray R, Bhattacharya S, Bowden C, Miller K, Comis RL. Independent review of E2100: a phase III trial of bevacizumab plus paclitaxel versus paclitaxel in women with metastatic breast cancer. J Clin Oncol. 2009;27(30):4966‐4972. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44. Jovanovic B, Mayer IA, Mayer EL, et al. A randomized phase II neoadjuvant study of cisplatin, paclitaxel with or without everolimus in patients with stage II/III triple‐negative breast cancer (TNBC). Clin Cancer Res. 2017;23:4035‐4045. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45. Kim SB, Dent R, Im SA, et al. Ipatasertib plus paclitaxel versus placebo plus paclitaxel as first‐line therapy for metastatic triple‐negative breast cancer (LOTUS): a multicentre, randomised, double‐blind, placebo‐controlled, phase 2 trial. Lancet Oncol. 2017;18(10):1360‐1372. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46. Kummar S, Wade JL, Oza AM, et al. Randomized phase II trial of cyclophosphamide and the oral poly (ADP‐ribose) polymerase inhibitor veliparib in patients with recurrent, advanced triple‐negative breast cancer. Invest New Drugs. 2016;34(3):355‐363. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47. Llombart‐Cussac A, Bermejo B, Villanueva C, et al. SOLTI NeoPARP: a phase II randomized study of two schedules of iniparib plus paclitaxel versus paclitaxel alone as neoadjuvant therapy in patients with triple‐negative breast cancer. Breast Cancer Res Treat. 2015;154:351‐357. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48. Nahleh ZA, Barlow WE, Hayes DF, et al. SWOG S0800 (NCI CDR0000636131): addition of bevacizumab to neoadjuvant nab‐paclitaxel with dose‐dense doxorubicin and cyclophosphamide improves pathologic complete response (pCR) rates in inflammatory or locally advanced breast cancer. Breast Cancer Res Treat. 2016;158(3):485‐495. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49. O'Shaughnessy J, Hellerstedt B, Schwartzberg L, et al. Phase III study of iniparib plus gemcitabine and carboplatin versus gemcitabine and carboplatin in patients with metastatic triple‐negative breast cancer. J Clin Oncol. 2014;32(34):3840‐3847. [DOI] [PubMed] [Google Scholar]
  • 50. O'Shaughnessy J, Osborne C, Pippen JE, et al. Iniparib plus chemotherapy in metastatic triple‐negative breast cancer. N Engl J Med. 2011;364(3):205‐214. [DOI] [PubMed] [Google Scholar]
  • 51. Pivot X, Schneeweiss A, Verma S, et al. Efficacy and safety of bevacizumab in combination with docetaxel for the first‐line treatment of elderly patients with locally recurrent or metastatic breast cancer: results from AVADO. Eur J Cancer. 2011;47(16):2387‐2395. [DOI] [PubMed] [Google Scholar]
  • 52. Robert NJ, Dieras V, Glaspy J, et al. RIBBON‐1: randomized, double‐blind, placebo‐controlled, phase III trial of chemotherapy with or without bevacizumab for first‐line treatment of human epidermal growth factor receptor 2‐negative, locally recurrent or metastatic breast cancer. J Clin Oncol. 2011;29(10):1252‐1260. [DOI] [PubMed] [Google Scholar]
  • 53. Tredan O, Campone M, Jassem J, et al. Ixabepilone alone or with cetuximab as first‐line treatment for advanced/metastatic triple‐negative breast cancer. Clin Breast Cancer. 2015;15(1):8‐15. [DOI] [PubMed] [Google Scholar]
  • 54. Yardley DA, Reeves J, Dees EC, et al. Ramucirumab with eribulin versus eribulin in locally recurrent or metastatic breast cancer previously treated with anthracycline and taxane therapy: a multicenter, randomized, phase II study. Clin Breast Cancer. 2016;16(6):471‐479.e471. [DOI] [PubMed] [Google Scholar]
  • 55. Robson M, Im SA, Senkus E, et al. Olaparib for metastatic breast cancer in patients with a germline BRCA mutation. N Engl J Med. 2017;377(6):523‐533. [DOI] [PubMed] [Google Scholar]
  • 56. Han HS, Dieras V, Robson M, et al. Veliparib with temozolomide or carboplatin/paclitaxel versus placebo with carboplatin/paclitaxel in patients with BRCA1/2 locally recurrent/metastatic breast cancer: randomized phase II study. Ann Oncol. 2018;29(1):154‐161. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57. Cameron D, Brown J, Dent R, et al. Adjuvant bevacizumab‐containing therapy in triple‐negative breast cancer (BEATRICE): primary results of a randomised, phase 3 trial. Lancet Oncol. 2013;14(10):933‐942. [DOI] [PubMed] [Google Scholar]
  • 58. Mayer IA, Abramson VG, Lehmann BD, Pietenpol JA. New strategies for triple‐negative breast cancer–deciphering the heterogeneity. Clin Cancer Res. 2014;20(4):782‐790. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59. Ge L, Tang Y, Zhang Q‐N, et al. A network meta‐analysis on the efficacy of targeted agents in combination with chemotherapy for treatment of advanced/metastatic triple‐negative breast cancer. Oncotarget. 2017;8:59539‐59551. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60. Lord CJ, Ashworth A. PARP inhibitors: synthetic lethality in the clinic. Science. 2017;355(6330):1152‐1158. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 61. Ma X, Wang X, Huang J, et al. Bevacizumab addition in neoadjuvant treatment increases the pathological complete response rates in patients with HER‐2 negative breast cancer especially triple negative breast cancer: a meta‐analysis. PLoS ONE. 2016;11(8):e0160148. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 62. Rose AA, Grosset AA, Dong Z, et al. Glycoprotein nonmetastatic B is an independent prognostic indicator of recurrence and a novel therapeutic target in breast cancer. Clin Cancer Res. 2010;16(7):2147‐2156. [DOI] [PubMed] [Google Scholar]
  • 63. Mittendorf EA, Philips AV, Meric‐Bernstam F, et al. PD‐L1 expression in triple‐negative breast cancer. Cancer Immunol Res. 2014;2(4):361‐370. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 64. Zhang M, Sun H, Zhao S, et al. Expression of PD‐L1 and prognosis in breast cancer: a meta‐analysis. Oncotarget. 2017;8(19):31347‐31354. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 65. Poggio F, Bruzzone M, Ceppi M, et al. Platinum‐based neoadjuvant chemotherapy in triple‐negative breast cancer: a systematic review and meta‐analysis. Ann Oncol. 2018;29(7):1497‐1508. [DOI] [PubMed] [Google Scholar]
  • 66. Lehmann BD, Bauer JA, Chen X, et al. Identification of human triple‐negative breast cancer subtypes and preclinical models for selection of targeted therapies. J Clin Investig. 2011;121(7):2750‐2767. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 67. Bareche Y, Venet D, Ignatiadis M, et al. Unravelling triple‐negative breast cancer molecular heterogeneity using an integrative multiomic analysis. Ann Oncol. 2018;29(4):895‐902. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 68. Rediti M, Messina C. Towards treatment personalisation in triple negative breast cancer: role of platinum‐based neoadjuvant chemotherapy. Breast Cancer Res Treat. 2018;172(1):239‐240. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

 

Click here for additional data file. (19.4MB, eps)

 

Click here for additional data file. (17.3MB, eps)

 

Click here for additional data file. (5.3MB, eps)

 

Click here for additional data file. (5.5MB, eps)

 

Click here for additional data file. (13.4MB, eps)

 

Click here for additional data file. (11.2MB, eps)

 

Click here for additional data file. (123.5KB, doc)

Articles from Cancer Medicine are provided here courtesy of Wiley

RESOURCES