Neoadjuvant Treatment with HER2-Targeted Therapies in HER2-Positive Breast Cancer: A Systematic Review and Network Meta-Analysis

Simple Summary Human epidermal growth factor receptor 2 (HER2)-positive breast cancer causes more aggressive progression of disease and poorer outcomes for patients. HER2-targeted medicines used as neoadjuvant systemic therapy could improve clinical outcomes in early-stage or locally advanced breast cancer patients. The purpose of this systematic review and network meta-analysis was to identify the neoadjuvant anti-HER2 therapy with the best balance between efficacy and safety. We found that trastuzumab emtansine + pertuzumab + chemotherapy had a high pathologic complete response with a low risk of adverse events compared to other neoadjuvant anti-HER2 regimens, while the pertuzumab + trastuzumab + chemotherapy regimen showed the highest disease-free survival. However, further trial data on neoadjuvant regimens with trastuzumab emtansine are needed to confirm these findings. Abstract This systematic review aimed to identify neoadjuvant anti-human epidermal growth factor receptor 2 (HER2) therapies with the best balance between efficacy and safety. Methods: A network meta-analysis was applied to estimate the risk ratios along with 95% confidence intervals (CIs) for pathological complete response (pCR) and serious adverse events (SAE). A mixed-effect parametric survival analysis was conducted to assess the disease-free survival (DFS) between treatments. Results: Twenty-one RCTs with eleven regimens of neoadjuvant anti-HER2 therapy (i.e., trastuzumab + chemotherapy (TC), lapatinib + chemotherapy (LC), pertuzumab + chemotherapy (PC), pertuzumab + trastuzumab (PT), trastuzumab emtansine + pertuzumab (T-DM1P), pertuzumab + trastuzumab + chemotherapy (PTC), lapatinib + trastuzumab + chemotherapy (LTC), trastuzumab emtansine + lapatinib + chemotherapy (T-DM1LC), trastuzumab emtansine + pertuzumab + chemotherapy(T-DM1PC), PTC followed by T-DM1P (PTC_T-DM1P), and trastuzumab emtansine (T-DM1)) and chemotherapy alone were included. When compared to TC, only PTC had a significantly higher DFS with a hazard ratio (95% CI) of 0.54 (0.32–0.91). The surface under the cumulative ranking curve (SUCRA) suggested that T-DM1LC (91.9%) was ranked first in achieving pCR, followed by the PTC_T-DM1P (90.5%), PTC (74.8%), and T-DM1PC (73.5%) regimens. For SAEs, LTC, LC, and T-DM1LC presented with the highest risks (SUCRA = 10.7%, 16.8%, and 20.8%), while PT (99.2%), T-DM1P (88%), and T-DM1 (83.9%) were the safest regimens. The T-DM1PC (73.5% vs. 71.6%), T-DM1 (70.5% vs. 83.9%), and PTC_T-DM1P (90.5% vs. 47.3%) regimens offered the optimal balance between pCR and SAE. Conclusions: The T-DM1PC, T-DM1, and PTC_T-DM1P regimens had the optimal balance between efficacy and safety, while DFS was highest for the PTC regimen. However, these results were based on a small number of studies, and additional RCTs assessing the efficacy of regimens with T-DM1 are still needed to confirm these findings.


Introduction
Breast cancer is the most common cancer among women worldwide, with 2.26 million new cases in 2020. It ranked fifth as the leading cause of mortality and accounted for 15% of cancer deaths in women [1]. Patients with human epidermal growth factor receptor 2 (HER2) positivity, accounting for approximately 20 to 24% of breast cancer patients globally, had more aggressive progression and poorer outcomes than those who were HER2negative. There have been many recent advances in HER2-targeted therapies, including trastuzumab (T), lapatinib (L), pertuzumab (P), and trastuzumab emtansine (T-DM1), which have significantly improved the progression and overall survival outcomes for HER2+ve breast cancer patients [2,3].
Network meta-analysis (NMA) is a method that combines data from direct comparisons to indirectly compare between different interventions by borrowing information from common comparators. In addition, NMA can estimate the probability associated with being the optimal treatment and enables the ranking of both positive and negative effects. Two NMAs published in 2018 and 2019 [8,9] identified the PTC regimen as the best treatment for pCR, whereas T-DM1P was the safest regimen with the lowest probability associated with adverse events. More recently, several RCTs have assessed newer neoadjuvant anti-HER2 regimens (i.e., T-DM1 alone, T-DM1 + lapatnib + chemotherapy (T-DM1LC), T-DM1 + pertuzumab + chemotherapy (T-DM1PC), and PTC +T-DM1P combination regimens) [10,11]. Furthermore, additional patient-relevant outcomes, such as disease-free survival (DFS), were not previously considered in the prior NMAs due to insufficient data. Therefore, this systematic review and NMA aimed to identify the current regimens of neoadjuvant anti-HER2 therapy with the highest probability of DFS and pCR coupled with the lowest risk of adverse events.

Materials and Methods
This systematic review was reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, and the review protocol was registered at the PROSPERO website (CRD42020211532).

Literature Search and Selection of Studies
Relevant studies were searched in Medline via PubMed, Scopus, and Cochrane Central Register of Controlled Trials from their inceptions to November 2021. Details of search terms and search strategies for each database are provided in Appendix A (see Tables A1-A3).

Selection of Studies
Studies were selected by two independent reviewers (A.D.M.G. and N.T.H.). Randomized controlled trials were eligible if they (1) included early-stage and/or locally advanced HER2+ve breast cancer patients, (2) compared any neoadjuvant regimens that included anti-HER2 therapies, with or without chemotherapy, and (3) reported any of the following outcomes: pCR, DFS, or overall survival. Trials that included HER2+ve breast cancer with distant metastasis or compared the different doses of the same treatment regimen were excluded.

Interventions of Interest
Interventions of interest included neoadjuvant therapies of HER2-targeted agents with or without chemotherapy, such as:
Dual HER2-targeted agents with chemotherapy (i.e., PTC, LTC, T-DM1PC, T-DM1LC, PTC_T-DM1P). T-DM1PC was a response guided regimen where T-DM1P was provided in cycles 1 through 4, followed by 4 cycles of chemotherapy in non-responders or a continuation of 2 cycles of T-DM1P in responders. PTC_T-DM1P was a regimen where PTC was provided in cycles 1 through 4, followed by T-DM1P in cycles 5 through 8; 3.

Outcomes of Interest
Primary outcomes of interest were DFS, overall survival, and pCR. DFS was defined as the time from randomization to first locoregional recurrence, contralateral breast cancer or distant metastasis or death from any causes [12]. Although overall survival was defined as the time from randomization to death from any cause [12,13], the limited number of studies reporting this outcome prevented its evaluation. Pathological complete response was defined as no histological evidence of residual invasive tumor cells in the breast and axillary lymph nodes (ypT0/Tis and ypN0) [5,14].

Data Extraction
Data were extracted by two independent reviewers (A.D.M.G., N.T.H.), including study characteristics (i.e., setting, follow-up time, participant numbers), patient characteristics (i.e., mean age, cancer stage, tumor size, nodal status, hormone receptor status, body mass index (BMI), family history of breast cancer, breast feeding, menopausal status), intervention characteristics (i.e., treatment regimens, dosages, course), and types of outcomes. For data pooling, contingency data between treatments and outcomes were extracted. For DFS, time (on x-axis), probability of outcome occurrence (on y-axis), number of events, and person-time at risk at each time point were extracted from Kaplan-Meier (KM) curves using webplot digitizer software [16].

Risk of Bias Assessment
Risk of bias of each study was assessed using the revised Cochrane risk-of-bias tool (RoB 2) by one reviewer (A.D.M.G.) and randomly checked by senior author (T.A.) [17,18]. Five domains (i.e., randomization, protocol deviation, missing outcome data, measurement of outcome, selection of results reported) were assessed and ranked as low risk, some concern, and high risk. An overall ROB was further classified as low, some concern, and high risk of bias; if all five domains were ranked as low risk, at least one domain was ranked as some concern without high risk, and at least one domain was ranked as high risk, respectively.

Statistical Analysis
Direct meta-analysis was performed if two or more studies with similar treatment comparisons and outcomes were available. Risk ratios (RR) of pCR, BCS, and SAEs were estimated and pooled across studies using the inverse variance method (fixed-effects) if there was no heterogeneity; otherwise, the DerSimonian-Laird method (random effects) was applied. Heterogeneity was assessed using Cochran's Q test and I 2 statistic. Heterogeneity was present if the p-value from Q-test < 0.10 and/or I 2 > 25%. Sources of heterogeneity were explored using a meta-regression by fitting co-variables (i.e., node positivity, hormone receptor negativity, chemotherapy regimens (i.e., taxane and taxane + anthracycline) and duration of anti HER2 treatment) one by one. Further subgroup analysis was performed if the Tau 2 value decreased by more than 50% after fitting the co-variable in the metaregression model. Publication bias was assessed using Egger's tests and funnel plots. If there was asymmetry of the funnel plot, a contour-enhanced funnel plot was generated to identify the cause of asymmetry (i.e., small study effect or heterogeneity) [19,20].
The NMAs for pCR, SAE, and BCS were performed using a two-stage NMA approach. First, a relative treatment effect (lnRR) was estimated along with the variance-covariance matrix using a binary regression model. Second, a multivariate meta-analysis with consistency model was applied to pool the lnRRs across studies. Treatment ranking was done using the rankogram and the surface under cumulative ranking curve (SUCRA). Cluster ranking considering both efficacy (pCR) and safety (SAE) was also performed to simultaneously weigh the risks and benefits of each intervention [21,22].
For DFS, the number of events, patients at risk, and probability of DFS at each time point were used to simulate individual patient data IPD for each study [23], and then pooled across all studies. A mixed-effect parametric survival analysis with Weibull distribution was applied to estimate relative treatment effects, i.e., hazard ratio (HR) between different neoadjuvant HER2-targeted regimens [24][25][26][27].
A comparison-adjusted funnel plot was constructed to assess publication bias across the network. The consistency assumption was assessed using design-by-treatment interaction inconsistency model [28]. If there was inconsistency (p-value < 0.05), treatment loops having high inconsistency factors (IF) were explored. Subgroup analysis was then performed by excluding studies with different characteristics to improve consistency of the model. All statistical analyses were conducted using STATA program version 16.0. A pvalue less than 0.05 was considered statistically significant for all tests except for Cochran's Q test, for which a p-value less than 0.10 was applied.

Risk of Bias Assessment
The results of the risk of bias assessment for the pCR, BCS, DFS, and SAE outcomes are presented in Table S3. The majority of the studies for pCR (13/21), BCS (9/12), and DFS (4/7) had a low risk of bias, and none had a high risk of bias. For the SAE outcome, only a single study had an overall high risk of bias due to missing outcome data, with 12 studies registering some concerns.
Given the high I 2 value observed in the LTC vs. LC comparison, sources of heterogeneity were explored (see Table S4A). Accounting for types of chemotherapy that decreased the I 2 value from 42% to 0%, a subgroup analysis showed that RR was higher with taxane only than taxane plus anthracycline, with pooled RRs (95% CI) of 2.13 (1.65, 2.75) and 1.33 (1.12, 1.58), respectively (see Figure S2). There was no significant variation in the comparisons from other subgroup analyses (Table S5). For LC vs. TC (see Table S4B), subgroup analyses showed reductions in heterogeneity when accounting for age, percentages of nodal positive, and T3 and T4, although none reached significance (Table S6).
Egger's tests did not provide any evidence of publication bias for any pCR pooled estimates (see Figure S3), in support of the funnel plots, with the exception for the comparison between TC vs. C, which was asymmetrical (see Figure S3A). A contour-enhanced funnel plot suggested the asymmetry may be a consequence of a small study effect ( Figure S3B).

Network Meta-Analysis
The NMA for pCR included 21 studies with 16 comparisons (Figure 2A) and showed consistency (global test chi-square = 2.01, p-value = 0.571). The data used for pooling are described in  Table 2). Both LC and PT regimens had significantly lower pCR when compared to TC. All dual anti-HER-2 agents plus chemotherapy (i.e., PTC, LTC, T-DM1LC, T-DM1PC, and PTC_T-DM1P) had significantly higher pCR than single anti-HER-2 agents plus chemotherapy regimens (i.e., LC, PC, TC) except LTC vs. PC, with pooled RRs (95% CI) of 1  Table 2). T-DM1 also had significantly higher pCR compared to LC, PC, and TC (see Table 2). In addition, all dual anti-HER-2 agents plus chemotherapy regimens T-DM1P and T-DM1 also significantly increased pCR when compared to PT (see Table 2). However, the chance of having pCR with dual anti-HER-2 agents plus chemotherapy regimens did not significantly differ from T-DM1P and T-DM1 regimens. In addition, both T-DM1LC and PTC_T-DM1P had significantly higher pCR compared to LTC, with RRs (95% CI) of 1.97 (1.06, 3.66) and 1.80 (1.06, 3.05), respectively. A SUCRA plot identified T-DM1LC with the highest probability of being the best regimen Cancers 2022, 14, 523 9 of 21 (91.9%), followed by PTC_T-DM1P (90.5%) and PTC (74.8%) (see Figure S4). Adjusted funnel plots were symmetrical, indicating no publication bias (see Figure S5).
The type of chemotherapy included could reduce the level of heterogeneity (Tau 2 ) for both LC vs. TC and LTC vs. TC, respectively (Table S8A,B). The subgroup analyses indicated that risks of SAEs in LC and LTC were higher than TC for the taxane regimen alone compared to taxane plus anthracycline regimens ( Figure S7A,B).
There was no evidence of publication bias as indicated by either Egger's tests or funnel plots for TC vs. C and LTC vs. LC ( Figure S8A,F). The funnel plots for LC vs. TC and LTC vs. TC were asymmetrical ( Figure S8B,D), which may be due to heterogeneity, as suggested from the contour-enhanced funnel plots ( Figure S8C,E).

Ranking of Regimens According to Efficacy and Safety
A cluster rank plot was constructed considering the SUCRA of pCR on the x-axis and SUCRA of lowering SAEs on the y-axis (Figure 3). T-DM1LC was located in the bottom right of the graph, representing the highest pCR with high SAEs, while T-DM1PC, PTC, T-DM1, and PTC_T-DM1P still ranked high for pCR but had better SAE profiles. Therefore, T-DM1PC, T-DM1, PTC, and PTC_T-DM1P provided an optimal balance between efficacy and risk of SAE.

Network Meta-Analysis
The NMA included 12 studies with 11 treatment regimens that passed the consistency assumption (global test chi-square = 0.16, p-value = 0.6937) (see Figure 2C). There were no significant differences in BCS for any of the treatment comparisons except for PTC vs. T-DM1P, with a pooled RR (95% CI) of 1.26 (1.02, 1.55) (see Table S11). PTC (68.5%) was ranked best in terms of having the highest BCS ( Figure S13). The comparisonadjusted funnel plots were symmetrical and indicated no evidence of publication bias ( Figure S14).

Network Meta-Analysis
The NMA included 12 studies with 11 treatment regimens that passed the consistency assumption (global test chi-square = 0.16, p-value = 0.6937) (see Figure 2C). There were no significant differences in BCS for any of the treatment comparisons except for PTC vs. T-DM1P, with a pooled RR (95% CI) of 1.26 (1.02, 1.55) (see Table S11). PTC (68.5%) was ranked best in terms of having the highest BCS ( Figure S13). The comparison-adjusted funnel plots were symmetrical and indicated no evidence of publication bias ( Figure S14).

Discussion
Our results suggest that T-DM1LC has the highest probability of achieving pCR, followed by PTC_T-DM1P and PTC, respectively. In addition, PTC also provided the longest DFS. PT was ranked first (i.e., lowest risk) for SAE, followed by T-DM1P and T-DM1. When considering both the optimal benefit (pCR or DFS) and risk (SAE) together, T-DM1PC, PTC_T-DM1P, T-DM1, or PTC were considered the best regimens for neoadjuvant anti-HER2 therapy.
Previous NMAs by Wu et al. [9] and Nakashoji et al. [8] identified PTC as the best neoadjuvant anti-HER2 regimen for achieving pCR in early-stage breast cancer. With our updated data, T-DM1LC now represents the top-ranked treatment, with PTC dropping to the third-ranked treatment in terms of pCR outcomes. In this fast-moving field of cancer therapeutics, the more recently available regimens of T-DM1 were not considered within the previous NMAs, including T-DM1, T-DM1LC, T-DM1PC, and PTC_T-DM1P. Trastuzumab emtansine consists of trastuzumab and the cytotoxic agent DM1 (derivative of maytansine), which can directly deliver cytotoxic molecules to tumors, potentially increasing the efficacy in comparison to trastuzumab [57,58]. Beyond pCR outcomes, our study also assessed the efficacy of neoadjuvant anti-HER2 therapies to prolong DFS, an outcome not considered in previous NMAs. PTC significantly increased DFS compared to TC, while none of the other anti-HER2 regimens significantly differed from TC. Nevertheless, there were some limitations associated with the more recent T-DM1 regimens, such as T-DM1 and T-DM1P, which had insufficient data to properly evaluate this outcome.
Although T-DM1LC was ranked the best for the pCR outcome, its SAE rate was very high. This may be due to lapatinib, which is associated with a higher risk of SAE grades 3-4, such as neutropenia, diarrhea, and hepatotoxicity [31,40,43,59]. Our study also indicated that all the regimens containing lapatinib (i.e., LTC, LC, and T-DM1LC) were ranked the worst for SAE. In addition to lapatinib, SAE grades 3-4 were commonly found in regimens with chemotherapy. Three regimens without chemotherapy (i.e., PT, T-DM1, and T-DM1P) were the best treatments for lowering SAEs. However, the efficacy of PT in terms of pCR and DFS was very low in comparison to the other regimens.
Moreover, the percentage of patients who discontinued the treatment was high in regimens with lapatinib. However, most studies applied intention to treat analysis for analyzing the data. Although this analysis might underestimate the true treatment efficacy in the ideal situation, it usually reflects the efficacy of the drugs when using them in the real-clinical setting where some patients may not be complying well with the treatment regimen.
When considering the benefit (pCR) and risk (SAE) together, T-DM1PC and T-DM1 were identified as the optimal neoadjuvant therapy for early-stage HER2+ve breast cancer given the greater chance of achieving pCR and the low risk of SAE. PTC_T-DM1P was identified as the next-best alternative regimen given the higher pCR to T-DM1PC coupled with the greater risk of SAE compared to T-DM1PC. These conclusions contrast to those from previous NMAs, which indicated PTC as the most effective treatment. However, our NMA has been updated to include the most recent treatment regimens and confirms that switching trastuzumab to T-DM1 in combination with PC could offer increased efficacy coupled with decreased risk of SAE. In addition, our findings failed to support chemotherapy alone or single anti-HER2 regimens in combination with chemotherapy (i.e., TC, PC, LC) as optimal neoadjuvant therapies for the treatment of early-stage HER2+ve breast cancer due to a very low chance of achieving pCR and an associated increased risk of SAE.
Although the recommended regimens of T-DM1 were top-ranked in our cluster ranking plot, its prohibitively high cost of $127,035 USD per patient per year should be considered [60]. Therefore, affordability and accessibility may be issues, especially in low-and middle-income countries, warranting economic evaluation before further consideration of these regimens.

Strengths and Limitations
To our knowledge, our NMA is the most up to date, and includes novel neoadjuvant anti-HER2 regimens with combinations of T-DM1 and chemotherapy, providing the most current evidence for treatment recommendations for HER2+ve breast cancer. Moreover, DFS is a clinically important outcome measure not considered in previous NMAs. However, not all novel regimens provided sufficient DFS outcome data for pooling, especially those regimens that included T-DM1. Therefore, we were unable to compare the regimen efficacy for DFS outcomes between new neoadjuvant anti-HER2 regimens, including T-DM1 and PTC, with other regimens, and, as such, recommendations on the optimal treatment regimens were mainly based on the outcomes of pCR and SAE. Nevertheless, several RCTs and meta-analyses suggest pCR following neoadjuvant therapy in HER2+ve breast cancer is a valid surrogate of long-term outcomes [61][62][63].
Our study also had some limitations. First, there were a small number of studies and participants that included newer treatment regimens, such as T-DM1 (i.e., T-DM1PC, T-DM1LC, PTC_T-DM1P, and T-DM1). We pooled the treatment effects based on a small number of included studies, each of which also had small sample sizes, which might result in the imprecision or uncertainty of treatment effects estimated from the network meta-analysis. Therefore, additional RCTs are needed to substantiate the findings from our study. Second, due to insufficient data, we were unable to analyze overall survival outcomes, which is the most important outcome of cancer treatment. Nevertheless, it has been shown that DFS is an adequate proxy for overall survival in HER2+ve breast cancer patients [64]. Third, all the included studies were funded by the pharmaceutical companies, which might have an influence on conducting the study, data analysis, or reporting the results. Therefore, further updated pooling treatment effects are required when there are more studies with new treatment regimens and/or with non-profit sponsors.

Conclusions
In conclusion, the T-DM1PC, T-DM1, and PTC_T-DM1P regimens had the optimal balance between efficacy (pCR) and safety (SAE) compared to other neoadjuvant anti-HER2 regimens for early-stage and locally advanced HER2+ve breast cancer, while DFS was the highest for the PTC regimen. Nonetheless, the results of regimens with T-DM1 are based on a small number of studies. Thus, additional RCTs to assess the efficacy of neoadjuvant regimens with T-DM1 are still needed to confirm these findings.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/ 10.3390/cancers14030523/s1, Figure S1: Forest plots of pairwise meta-analyses for pathological complete response (pCR), Figure S2: Subgroup analysis of pathological complete response (pCR) in LTC vs. LC according to types of chemotherapy, Figure S3: Funnel plots and contour-enhanced funnel plots of pairwise meta-analyses comparisons for pathological complete response (pCR), Figure S4: Surface under the cumulative ranking curves (SUCRA) for pathological complete response (pCR), Figure S5: Comparison-adjusted funnel plot of network meta-analysis of pathological complete response (pCR), Figure S6: Forest plots of pairwise meta-analyses for serious adverse events (SAE), Figure S7: Subgroup analyses according to types of chemotherapy for serious adverse events (SAE) (a) LC vs. TC; (b) LTC vs. TC, Figure S8: Funnel plots and contour-enhanced funnel plots of pairwise meta-analyses for serious adverse events (SAE), Figure S9: Inconsistency factor plot in network meta-analysis for serious adverse events (SAE), Figure S10: Surface under the cumulative ranking curves (SUCRA) for serious adverse events (SAE), Figure S11: Comparison-adjusted funnel plot of network meta-analysis for serious adverse events (SAE), Figure S12: Forest plots of pairwise meta-analyses for total breast conservation surgery, Figure S13: Surface under the cumulative ranking curves (SUCRA) of total breast conservation surgery (BCS), Figure S14: Comparison-adjusted funnel plot of network meta-analysis for total breast conservation surgery (BCS), Table S1: Summary of studies and total number of patients included for each treatment regimen,  Table S7: Treatment comparisons and data used for pooling pathological complete response (pCR) outcome in network meta-analysis, Table S8: Results of meta regression of LC vs. TC and LTC vs. TC comparisons for serious adverse events (SAE) outcome, Table S9: Treatment comparisons and data used for pooling serious adverse events (SAE) outcome in network meta-analysis, Table S10: Comparison of study characteristics in treatment loop with inconsistency for serious adverse events (SAE) outcome, Table S11: Risk ratios and 95% confidence intervals of network meta-analysis of total breast conservative surgery (BCS) outcome. References [11,29,30,34,38,49,51,54] are cited in the supplementary materials.