Dual- versus single-agent HER2 inhibition and incidence of intracranial metastatic disease: a systematic review and meta-analysis

Observational studies have suggested that HER2 inhibition with trastuzumab may be associated with an increased incidence of intracranial metastatic disease (IMD) due to its ability to prolong survival. We hypothesized that prolonged survival associated with dual-agent HER2 inhibition may be associated with an even higher incidence of IMD. This study pooled estimates of IMD incidence and survival among patients with HER2-positive breast cancer receiving dual- versus single-agent HER2 targeted therapy, as well as trastuzumab versus chemotherapy, observation, or another HER2-targeted agent. We searched PubMed, EMBASE, and CENTRAL from inception to 25 March 2020. We included randomized controlled trials that reported IMD incidence for patients with HER2-positive breast cancer receiving trastuzumab as the experimental or control arm irrespective of disease stage. Among 465 records identified, 19 randomized controlled trials (32,572 patients) were included. Meta-analysis of four studies showed that dual HER2-targeted therapy was associated with improved overall survival (HR 0.76; 95% CI, 0.66–0.87) and progression-free survival (HR 0.77; 95% CI, 0.68–0.87) compared to single HER2-targeted therapy, but the risk of IMD was similar (RR 1.03; 95% CI, 0.83–1.27). Our study challenges the hypothesis that prolonged survival afforded by improved extracranial disease control is associated with increased IMD incidence.


INTRODUCTION
Intracranial metastatic disease (IMD) is a common and serious complication of breast cancer 1 , with a median survival of 13.8 months 2 and reduced quality of life due to disease symptomatology and treatment toxicity 3 . Breast cancers expressing human epidermal growth factor receptor 2 (HER2) have a higher propensity to metastasize to the central nervous system (CNS) compared to hormone receptor (HR)-positive/HER2-negative disease subtypes 4,5 The anti-HER2-monoclonal antibody trastuzumab has been shown to improve overall survival (OS) for HER2-positive breast cancer patients and has become standard of care 6,7 . However, case series and cohort studies have reported a higher incidence of IMD among HER2-positive patients treated with trastuzumab for metastatic, unresectable, or recurrent breast cancer [8][9][10][11][12][13][14][15][16] . Metaanalyses of randomized controlled trials (RCTs) have corroborated such findings in non-metastatic disease, but have not included data accrued in recent years [17][18][19][20][21] . The increased IMD incidence following treatment with trastuzumab has been attributed to its improvement of OS: trastuzumab controls extracranial disease and prolongs survival until dormant micrometastases within the sanctuary of the CNS are able to proliferate and manifest clinically 22,23 .
HER2-targeted agents have also garnered interest in the treatment of IMD from HER2-positive breast cancer. Trastuzumab and other HER2-targeted antibodies and conjugates have been associated with a reduced number of intracranial tumors at IMD diagnosis, prolonged post-IMD OS, or intracranial responses, indicating possible intracranial efficacy despite minimal bloodbrain barrier penetrance [37][38][39] . Recent reviews address the landscape of HER2-targeted agents for the management of IMD, and suggest that the introduction of novel agents, greater inclusion of patients with IMD in clinical trials, and increased reporting of intracranial outcomes may all benefit future patients with IMD 40,41 .
The purpose of this systematic review and meta-analysis was to update existing estimates of the incidence of IMD among patients with HER2+ breast cancer, and to assess the impact of novel HER2-targeted regimens on the development of intracranial metastases. To address this question, we assessed IMD incidence and survival among patients with HER2-positive breast cancer who were treated with dual anti-HER2 therapy versus trastuzumab monotherapy; we also performed meta-analysis of IMD incidence in patients receiving trastuzumab versus chemotherapy, observation, or another HER2-targeted agent. 1

METHODS
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines 42 .

Search strategy
We searched MEDLINE (via PubMed), EMBASE (via Wiley), and CENTRAL (via Cochrane) on March 25, 2020. We also screened references of eligible articles and reviews, and queried Google Scholar, PubMed, and ClinicalTrials.gov for updated or IMDspecific publications of trials at full-text review. Full search queries are presented in the supplement (Supplementary Tables 1-3).

Study selection
Using a two-step process, we screened abstracts and then full texts of selected records to identify RCTs that reported the incidence of IMD that compared dual HER2-targeted regimens to trastuzumab, trastuzumab to another HER2-targeted therapy, or trastuzumab to standard chemotherapy or observation. Studies were screened in duplicate by two independent reviewers (AE, CH), and Cohen's κ statistic was calculated for inter-rater reliability at both steps. Disagreements were resolved through discussion. Studies could report IMD incidence overall or as the site of the first recurrence. Trials that did not report IMD incidence were excluded. Gray literature sources were not searched. Conference abstracts were eligible. No date range was applied, but studies were required to be in English. Full inclusion and exclusion criteria are presented in the supplement (Tables S4-5).
Data extraction and quality assessment The following data were extracted from included studies: trial name, treatment procedures, median follow-up, prior treatments, early (stage I-II) versus advanced (stage III-IV) HER2-positive breast cancer, number of intracranial events, number of recurrence events, OS (hazard ratio) and progression-free survival (PFS, hazard ratio), reported either as disease-free, progression-free, or eventfree survival. Outcomes specific to intention-to-treat analyses were preferentially extracted. Data were extracted by a single reviewer due to resource constraints. We performed a quality assessment using the Cochrane Risk of Bias 2 tool (RoB 2) to evaluate risk of bias across five domains (randomization, deviation from intended interventions, missing outcomes, measurement bias, and selection bias) and overall 43 . We assessed evidence quality using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework 44 .

Data synthesis and analysis
We performed a meta-analysis using the inverse variance method with random-effects models to produce summary risk ratios (RR) and hazard ratios (HR). We calculated RR using the number of cases of incident IMD divided by the number of patients in the intention-to-treat population per study arm. HR were extracted as adjusted HR when available, otherwise as reported. Missing HR values from one study 6 were imputed using the method by Guyot et al. 45 from digitized Kaplan-Meier plots and at-risk tables.
To assess heterogeneity, we calculated the Q-statistic for the ratio of observed to within-study variance, τ 2 for between-study variance, and I 2 for the percentage of observed variance attributable to between-study variance 46,47 .
We evaluated the incidence of IMD among patients receiving HER2-targeted therapy through two separate comparisons: dual HER2-targeted therapy versus trastuzumab; and trastuzumab versus chemotherapy, observation, or another anti-HER agent. Patients were pooled from treatment arms that received the same chemotherapy plus trastuzumab combination, but in concurrent or sequential order, or for different durations. As subgroup analyses, we estimated summary effects stratifying by disease stage, and in comparisons of trastuzumab monotherapy to control regimens, by the comparator. As sensitivity analyses, we compared summary estimates from fixedand random-effects models, and omitted studies one at a time as a "leave-out-one" assessment 47,48 .
We assessed publication bias by examining the funnel plot asymmetry visually and with Egger's test 49 . All statistical analysis was conducted using the R programming language (v3.6.1, R Foundation for Statistical Computing) 50 and the R package meta 51 .
Dual HER2-targeted therapy was associated with prolonged OS (four studies; HR 0.76; 95% CI, 0.66-0.87; Fig. 2; GRADE high) and PFS (four studies; HR 0.77; 95% CI, 0.68-0.87; Fig. 3; GRADE high) compared to single HER2-targeted therapy with trastuzumab. Heterogeneity in these comparisons was low (I 2 = 0% for OS and 11% for PFS). Stratification by disease stage showed dual HER2targeted therapy was associated with prolonged OS and PFS in both early stage (OS: three studies; HR 0.82; 95% CI, 0.68-0.99; Fig The risk of IMD incidence was not different between patients receiving dual versus single HER2-targeted therapy (four studies; RR 1.03; 95% CI, 0.83-1.27; Fig. 4). Heterogeneity in this comparison was low (I 2 = 0%). Subgroup analysis revealed no difference between early stage (three studies; RR 1.03; 95% CI, 0.78-1.37; Fig. 4) and advanced-stage disease (one study; RR 1.03; 95% CI, 0.74-1.42; Fig. 4 Patients receiving trastuzumab did not show an increased incidence of IMD compared to another HER2-targeted therapy (nine studies; RR 1.15; 95% CI 0.88-1.50), observation (two studies; RR 1.12; 95% CI 0.78-1.60), or chemotherapy (five studies; RR 1.32; 95% CI, 0.88-1.97) (Fig. 5). Heterogeneity in this comparison was moderate overall (I 2 = 37%), and higher but still moderate in the chemotherapy (I 2 = 57%) and anti-HER2 agent (I 2 = 38%) comparator subgroups. The summary estimate for IMD incidence from pooling the seven studies of trastuzumab monotherapy versus chemotherapy or observation comparators was RR 1. The overall risk of bias was low in 6/18 (33%) and moderate in 12/18 (66%) included RCTs (Fig. S5). Summary plots for risk of bias showed low to moderate risk for the meta-analyses in this study ( Supplementary Fig. 6). Assessment of funnel plots did not indicate publication bias, although these assessments were underpowered in all but one case . Sensitivity analysis showed that the findings were robust: comparison of random-and fixed-effects estimates did not render any significant summary estimates insignificant, and iterative omission of each study in the "leave-out-one" analyses did not significantly perturb summary estimates, including one study with follow-up <1 year 29 . GRADE certainty level was "high" for summary estimates ( Table 2).

DISCUSSION
Our study found prolonged OS and PFS without a significant difference in IMD incidence, with the addition of lapatinib or pertuzumab to trastuzumab for patients with HER2-positive breast cancer. Although there is mixed evidence for the ability of lapatinib to penetrate the blood-brain barrier [66][67][68] , our pooled analysis of two studies failed to show a difference in IMD incidence between lapatinib plus trastuzumab versus trastuzumab alone. While our study was not designed to assess this comparison, future reporting may clarify a role for HER2targeted TKIs in IMD prevention. Of note, several high-profile trials (HER2CLIMB 30 , DESTINY 34 , TH3RESA 69 , EMILIA 24 , MAR-IANNE 70 ) were captured in the literature search but did not meet inclusion criteria because they either have not yet reported IMD incidence, or featured absent or ineligible comparators.
Our study did not find a significant difference in IMD incidence between patients receiving dual anti-HER2 therapy versus trastuzumab alone. We also did not find a difference comparing trastuzumab with chemotherapy or observation, but this is in contradistinction to previous meta-analyses of RCTs 17-21 . One possible explanation for this difference is that our meta-analysis for this outcome involved more patients and longer follow-up for events to accrue. Further, previous meta-analyses included only studies of early stage disease, but our subgroup analysis of this population did not show a significant difference in IMD incidence between trastuzumab monotherapy and chemotherapy or observation (HR 1.01; 95% CI, 0.81-1.26; p = 0.92; Supplementary Fig. 3). Conversely, our subgroup analysis of advanced-stage disease showed an association between trastuzumab monotherapy and increased incidence of IMD, compared to chemotherapy or another HER2-targeted agent (HR 1.53; 95% CI, 1.19-1.97; p = 0.001; Supplementary Fig. 4).
These subgroup findings suggest that IMD may be more likely among patients with advanced-stage HER2+ breast cancer who receive trastuzumab monotherapy as compared to chemotherapy or another HER2-targeted agent. These findings also suggest that the impact of dual-versus single-agent HER2-therapy on IMD incidence could be different for patients with advanced-versus early stage disease. Our study did not detect such a difference (p = 0.97; Fig. 4), but this could be due to the small number of studies (advanced: n = 1, early: n = 3) in this comparison. Our subgroup analysis of early stage disease did not show a difference in IMD incidence between patients receiving trastuzumab monotherapy versus chemotherapy or observation. However, this  does not rule out the possibility that trastuzumab monotherapy influences IMD incidence for these patients, as the rarity of IMD in these studies renders this estimate susceptible to confounding. GRADE certainty level was high for summary estimates, although some indirectness may be present in the estimates due to pooling studies of early-and advanced-stage disease. As well, the magnitudes of summary effect sizes were modest.
Our study has several limitations. First, our study did not distinguish between the incidence of IMD overall versus the incidence of IMD as the first site of recurrence. This may have biased our results towards a lower estimate of overall IMD incidence, although one would expect this bias to impact both experimental and control arms similarly. The impact of HER2targeted agents on IMD incidence overall versus as the first site of recurrence remains underexplored 71 . Second, patients were pooled from treatment arms that received the same chemotherapy plus trastuzumab combination, but in concurrent or sequential order, or for different durations. This was necessary because there were few studies available for pooling. However, omitting these studies one at a time in the sensitivity analysis did not change our conclusions. Third, our study did not take into account the duration of HER2-targeted regimens, which is a potential confounder, although data from the HERA study suggest no difference in OS or IMD incidence between 1-year and 2-year trastuzumab treatment arms 55 . Fourth, IMD was not the main outcome of most RCTs, and RR was reported instead of HR. For patients with non-metastatic breast cancer, the occurrence of IMD may take years to develop, and differences in time-to-event comparisons may be obscured by comparing RR across studies with different follow-up durations. Future studies are needed to establish whether HER2-targeted therapy delays rather than prevent IMD. Fifth, blinding was infrequent in the included studies, which could have resulted in an overestimation of survival benefit for patients receiving the study regimen in included studies due to performance bias. Sixth, regular intracranial imaging was reported by only two studies, which could have    Fig. 3 Progression-free survival with the addition of pertuzumab or lapatinib to trastuzumab for patients with HER2-positive breast cancer. Hazard ratios for progression-free survival were extracted from eligible studies and pooled using a random-effects model. Studies here are stratified by disease stage: either early (stage I-II) or advanced (stage III-IV). The size of each box represents the weight of each study in the meta-analysis. The vertical solid line represents the point of equivalence between dual and single HER2 therapy. The vertical dashed and dotted lines represent the points of summary for fixed and random effects models, respectively, and the diamonds represent 95% CI for the summary hazard ratios. Analyses were performed with the R programming language 50 and the R package meta 51 .   Fig. 2 Overall survival with the addition of pertuzumab or lapatinib to trastuzumab for patients with HER2-positive breast cancer. Hazard ratios for overall survival were extracted from eligible studies and pooled using a random-effects model. Studies here are stratified by disease stage: either early (stage I-II) or advanced (stage III-IV). The size of each box represents the weight of each study in the meta-analysis. The vertical solid line represents the point of equivalence between dual and single HER2 therapy. The vertical dashed and dotted lines represent the points of summary for fixed and random effects models, respectively, and the diamonds represent 95% CI for the summary hazard ratios. Analyses were performed with the R programming language 50 and the R package meta 51 .
resulted in an underestimation of a difference in IMD incidence between groups. Seventh, this study was unable to address the impact of the therapy lines on outcomes due to too few studies. Finally, our results may have been impacted by the biases associated with RCTs. Attrition bias may have led to underestimates of IMD incidence following HER2-targeted therapy, although this is mitigated by the majority of included studies reporting analyses based on intention-to-treat populations 25,28,52,54,55,[57][58][59]62,[72][73][74][75][76] . Although most studies excluded at full-text review did not report IMD incidence despite evaluating the incidence of other safety outcomes in patients receiving HER2-targeted agents, we were unable to demonstrate publication bias with the small number of studies included.
While other meta-analyses have assessed survival and safety between dual-and single-agent HER2-targeted regimens 77-79 , our study compares IMD incidence in these groups and provides the   Fig. 4 Risk ratio of intracranial metastatic disease with the addition of pertuzumab or lapatinib to trastuzumab for patients with HER2positive breast cancer. Risk ratios were calculated from the proportion of patients in each study arm who developed the intracranial metastatic disease over the study course. Studies here are stratified by disease stage: either early (stage I-II) or advanced (stage III-IV). The size of each box represents the weight of each study in the meta-analysis. The vertical solid line represents the point of equivalence between dual and single HER2 therapy. The vertical dashed and dotted lines represent the points of summary for fixed and random effects models, respectively, and the diamonds represent 95% CI for the summary relative risks. Analyses were performed with the R programming language 50   Risk ratios were calculated from the proportion of patients in each study arm who developed the intracranial metastatic disease over the study course. Studies here are stratified by comparator regimen: either chemotherapy, observation, or another HER2-targeted agent. The size of each box represents the weight of each study in the meta-analysis. The vertical solid line represents the point of equivalence between trastuzumab and comparators. The vertical dashed and dotted lines represent the points of summary for fixed and random effects models, respectively, and the diamonds represent 95% CI for the summary relative risks. Analyses were performed with the R programming language 50 and the R package meta 51 . most recent update on the risk of IMD incidence following trastuzumab monotherapy. Our data suggest prolonged survival may not be associated with increased risk of IMD incidence when comparing dual to single HER2-targeted therapy, and when comparing trastuzumab to chemotherapy and observation in early stage disease. In an effort to maximize the knowledge gained from randomized treatment allocation, follow-up studies of these trials focusing on the incidence of IMD should be conducted. In addition, the incidence of IMD should be prospectively collected in future studies that involve patients who are at high risk of central nervous system metastases.

CONCLUSIONS
Dual-agent HER2-targeted therapy for eligible patients with breast cancer is associated with prolonged survival without increased risk of IMD compared to trastuzumab. Together, these findings suggest dual HER2-targeted therapy is associated with decreased IMD risk per unit time. The incidence of IMD among patients receiving trastuzumab was not greater than those who received other HER2targeted agents, chemotherapy, or observation in the setting of early stage disease. Future trials should monitor and report IMD incidence to assess whether the novel and existing systemic therapies may further impact IMD epidemiology across breast cancer subtypes.

DATA AVAILABILITY
The authors declare that the data supporting the findings of this study are available within the paper and its supplementary information files.