Clinicopathologic Characteristics and Prognosis of ERBB2-Low Breast Cancer Among Patients in the National Cancer Database

Key Points Question Do the demographics, clinicopathologic characteristics, and prognosis differ between breast cancers with no erb-b2 receptor tyrosine kinase 2 (ERBB2; formerly HER2 or HER2/neu) expression (ERBB2 negative) and those with low-level ERBB2 expression (ERBB2 low)? Findings In this cohort study of 1 136 016 patients from the National Cancer Database in the US, the proportions of ERBB2-low breast cancer were slightly lower among Hispanic and non-Hispanic Black patients compared with non-Hispanic White patients. ERBB2-low status was associated with slightly improved overall survival (≤2% difference at 5 years) compared with ERBB2-negative cancer. Meaning The findings of this study suggest that treatment response and long-term outcomes may be similar in ERBB2-low and ERBB2-negative cancers and do not support the classification of ERBB2-low breast cancer as a unique disease entity.

E rb-b2 receptor tyrosine kinase 2 (ERBB2; formerly HER2/ neu)-positive breast cancer was first defined as breast cancer with an amplification of ERBB2 leading to ERBB2 overexpression. 1 ERBB2 is clinically assessed with immunohistochemistry (IHC), with expression scored from 0 to 3+, and/or with in situ hybridization (ISH), which can detect ERBB2 amplification. ERBB2-positive cancers are currently defined as those with high expression (3+) via IHC or ERBB2 amplification via ISH. The precise classification of ERBB2-positive breast cancer has evolved to align with tumors predicted to respond to ERBB2-targeted therapies such as trastuzumab, constituting about 15% of all breast cancers. 2 Although approximately 50% to 60% of all breast cancers exhibit a low level of ERBB2 expression via IHC, cancers with low expression benefit from standard ERBB2-targeted therapies. [3][4][5][6][7] The paradigm of ERBB2 classification has recently shifted with the development of antibody-drug conjugates such as trastuzumab-deruxtecan (T-DXd), which has demonstrated efficacy in patients with ERBB2-low breast cancer, defined as 1+ ERBB2 expression via IHC or 2+ with negative ISH. 8,9 This has led to renewed interest in the subgroup of ERBB2-low breast cancer because greater understanding of the biological characteristics of this patient population could yield additional therapeutic approaches.
However, ERBB2-low tumors are heterogeneous, consisting of both hormone receptor-positive and triple-negative breast cancer (TNBC). 2,5,6 Multiple groups have undertaken studies to define ERBB2-low tumors in terms of epidemiological and clinicopathologic parameters; however, this has yielded conflicting results such as rates of ERBB2-low status among TNBC ranging from 40.1% 10 to 66.3%. 11 Moreover, there are conflicting reports of ERBB2-low status being associated with a negative, 5,10,11 neutral, 12,13 or positive prognosis. 14,15 Previous studies have had limited geographical distribution and/or relatively small sample sizes in the cohort analyzed, limiting the power to discern a prognostic difference between ERBB2low and ERBB2-negative (ERBB2 expression scored as 0 via IHC) breast cancer and the generalizability of the results. Furthermore, given the known racial and ethnic disparities in breast cancer prognosis, it is important to describe the epidemiological characteristics of ERBB2-low breast cancer to understand the potential association with novel antibody-drug conjugates such as T-DXd on breast cancer disparities. Our aim was to investigate whether ERBB2-low breast cancer is a clinically distinct subtype in terms of epidemiological characteristics, prognosis, and response to neoadjuvant chemotherapy.

Study Design and Data Source
This retrospective cohort study examined and compared the epidemiological differences and prognostic significance of ERBB2low vs ERBB2-negative breast cancer. We abstracted patient data from the National Cancer Database (NCDB), the largest cancer registry in the world, which includes data from approximately 70% of new invasive cancer diagnoses in the US. 16 The data were analyzed from November 1, 2021, through November 30, 2022. This study was deemed exempt from review by the institu-tional review board at the University of Chicago per 45 CFR part 46 of the Human Health Services regulations for human subjects research, which defines the criteria for secondary research for which consent is not required. No informed consent was obtained because NCDB data are deidentified. The data collected by the American Cancer Society/Commission on Cancer (ACS/CoC) for the NCDB are collected passively through cancer registries without informed consent collected from patients. The ACS/CoC and participating hospitals exempted the requirement of informed consent. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline for cohort studies. 17

Study Population and Covariates
Patients from the NCDB who were diagnosed with invasive breast cancer from January 1, 2010, to December 31, 2019, that was not classified as ERBB2 positive and who had ERBB2 IHC results available were included. ERBB2-low cases were defined as those with an IHC score of 1+, or 2+ with negative ISH results, and ERBB2-negative cases were defined as those with an IHC score of 0. We excluded patients with an ERBB2 IHC score of 2+ disease who did not have documented negative ISH results. We only included cases of invasive disease and excluded patients classified as stage 0 or with unknown stage.
The following epidemiological and clinicopathologic parameters were analyzed and compared between the ERBB2-low and ERBB2-negative groups: age at diagnosis; race and ethnicity; treatment facility type; tumor, nodal, and overall stage; tumor histology and grade; site of metastasis (including bone, brain, liver, and lung); quantitative estrogen receptor (ER) and progesterone receptor (PR) status; Ki-67 status determined via IHC, Oncotype DX (Exact Sciences Corp) multigene assay score; ERBB2 copy number and/or ERBB2/centromeric region of chromosome 17 (CEP17) ratio determined via ISH; receipt of chemotherapy and/or hormonal therapy; and survival status. Race and ethnicity were included in this analysis as part of a comprehensive assessment of the relationship of demographic factors and ERBB2-low status. The NCDB reported that the race and ethnicity information is collected from cancer program registries from patient self-report and medical/billing records. The race and ethnicity categories included Asian or Pacific Islander, Hispanic, Native American, Non-Hispanic Black, Non-Hispanic White, and other. Other is a subcategory listed in the NCDB and represents patients who were coded as Other by the local cancer registry. No explicitly defined racial or ethnic subgroup categorized by the NCDB was collapsed into Other. The TNM tumor staging was based on pathological stage if available, except in patients who received neoadjuvant chemotherapy (defined as receipt of chemotherapy at least 30 days prior to surgery), in which case clinical stage was used. Pathologic complete response (pCR) status was also recorded for patients who received neoadjuvant chemotherapy. As with all variables used in our analysis, race and ethnicity were abstracted by certified cancer registrars from the medical records for inclusion in the NCDB. Missing covariates were inferred with multiple imputation by chained equations. 18

Statistical Analysis
We first performed descriptive statistics to compare clinicopathologic and epidemiological features between ERBB2-low and ERBB2-negative subgroups, using a χ 2 test to compare categorical variables and an unpaired t test to compare continuous variables. A multivariable logistic regression analysis was used to quantify the independent association of these features with low-level ERBB2 expression, and adjusted odds ratios (aORs) were reported with significance and 95% CIs computed via Wald statistic. A similar logistic regression model was formulated to examine the association of ERBB2-low status with sites of metastatic disease. Overall survival (OS) was defined as the time from date of diagnosis to the last follow-up or death. Kaplan-Meier curves for OS were calculated for ERBB2-low vs ERBB2-negative breast cancer, stratified by receptor subtype and stage group. Multivariable Cox proportional hazards regression models were fit to examine the independent prognostic value of ERBB2-low breast cancer. For patients who received neoadjuvant therapy, a multivariable logistic regression analysis was used to examine the association of ERBB2-low status with pCR. All multivariable analyses controlled for age, sex, race and ethnicity, Charlson-Deyo Comorbidity Index score, treatment facility type, tumor grade and histology, and quantitative ER and PR status (except for analyses in the TNBC subgroup). Models for survival in the entire cohort and prediction of ERBB2-low status included overall stage group, and models for pCR included T and N stage groups. Statistical analysis was performed using Python, version 3.7.5 (Python Software Foundation), using the statsmodels, version 0.13.2, and lifelines, version 0.27.1. The code used for data extraction and statistical testing is publicly available. 19 All statistical tests were 2-sided with a significance threshold of P < .05.  Table 1 and eTable 1 in Supplement 1, 99.1% of patients were female, and the mean (SD) age at primary diagnosis was 62.1 (13.2) years for the ERBB2-negative group and 62.5 (13.0) years for the ERBB2-low group.

Association of ERBB2-Low Status With ISH Results
It is known that there is great variability in stratifying the ERBB2-low from ERBB2-negative subtype by IHC. 21 Thus, we assessed differences between ERBB2 copy number and ERBB2/CEP17 ratio to examine if the results of ISH testing results differ between ERBB2-negative and ERBB2-low patients to aid in selecting patients for ERBB2-low-directed therapies (eFigure 3 in Supplement 1). Optimal cutoffs for copy number and ratio were selected (maximizing Youden's index). The optimal ERBB2/CEP17 ratio of greater than 1.09 was associated with a sensitivity of 76.6% for ERBB2-low disease and a specificity of 36.8%, while an ERBB2 copy number of greater than 2.01 had a sensitivity of 61.1% and specificity of 58.2%. Performance characteristics at other cutoffs were also evaluated (eTable 5 in Supplement 1).

Discussion
This cohort study is, to our knowledge, the first to use the NCDB in over 1 million patients diagnosed with ERBB2-low or ERBB2negative breast cancer and reflects the geographic and ethnic distribution typically seen in the disease within the US. 22 Similar to prior studies, 10,11 we found that ERBB2-low status is more frequent in hormone receptor-positive than triple-negative disease. Previous studies 5,10 have also reported a higher rate of  ductal histology among ERBB2-low cases, but our study also found associations with other subtypes, such as less frequent ERBB2-low disease in metaplastic and medullary tumors. In contrast to prior work, 12 ERBB2-low disease was associated with lower rates of brain metastasis. Our analysis also found slightly lower rates of ERBB2-low disease in non-Hispanic Black and Hispanic women. For non-Hispanic Black patients, this difference was mediated by other known clinicopathologic features such as higher proportions of TNBC, 23,24 highlighting the complex interplay between low-level ERBB2 expression and patient ancestry that deserves further study. Whether ERBB2-low breast cancer represents a unique subtype is controversial 20,25 ; multiple studies have found associations of ERBB2-low status with response to therapy and prognosis. However, the size of these differences in our study and others 10,11 was small and confounded by hormone receptor status. We found a 2.6% reduction in pCR rates in hormone receptor-positive ERBB2-low disease and a 3.2% reduction in pCR rates among TNBC ERBB2-low patients-these reductions persisted on multivariable analysis, likely due, in part, to the large sample size. These findings are in line with the 4.6% lower pCR rates in ERBB2-low TNBC seen by Tarantino et al 20 or the approximately 6% lower rates of pCR in hormone receptor-positive ERBB2-low disease seen in multiple studies. 10,15 However, the clinical significance of these differences is questionable, and the size of the differences was similar to a 10% increase in ER expression on multivariable analysis seen in our study (ie, the aOR for ERBB2-low vs ERBB2-negative was 0.89, similar to a 10% increase in ER expression [aOR, 0.91]). Our data suggest that ERBB2-low status alone should not influence neoadjuvant treatment decisions with currently approved regimens in this setting but could perhaps provide an incremental improvement in multivariable or multiomic models for pCR. 26 We also found a slight improvement in OS in ERBB2-low breast cancer, particularly in advanced TNBC, although the size of the difference is of questionable clinical relevance, with overlapping survival curves when plotted by stage and receptor status. In a pooled analysis of 2310 patients receiving neoadjuvant chemotherapy, Denkert et al 15 found a 5.9% improvement in OS in ERBB2-low TNBC, but a survival advantage was not seen in ERBB2-low hormone receptor-positive disease. However, most other studies have not found survival differences based on ERBB2-low status, highlighting that any prognostic association of ERBB2-low status is likely subtle. 5,11,[27][28][29][30] Undoubtedly, the introduction of antibody-drug conjugates will define the prognostic implications of ERBB2-low breast cancer moving forward given the marginal nature of these survival differences.
Whether there is a molecular basis that could explain these subtle differences in survival and chemotherapy responsiveness in ERBB2-low breast cancer remains to be ascertained. Prior studies have found that ERBB2-low cancers have an overrepresentation of the luminal A molecular subtype, which is  known to have lower rates of pCR but maintains an excellent prognosis. 5,23 Furthermore, ERBB2-low cases may be associated with the luminal androgen receptor (LAR) subtype of TNBC, given higher rates of androgen receptor positivity among ERBB2-low cases 28 and high rates of ERBB2-enriched disease by 50-gene signature assay (PAM50) among patients with LAR TNBC. 31 Because androgen receptor-positive TNBC has a better prognosis but worse response to chemotherapy, further study is needed to examine how much of the prognostic association of ERBB2-low TNBC is attributable to enrichment for the LAR subtype. 32 The inaccuracies in IHC may further decrease the ability to accurately discern the length of survival or pCR differences associated with low-level ERBB2 expression and may have contributed to the inconsistent prognostic associations seen in the literature. More accurate quantification of levels of ERBB2 expression might aid in assessing associations with outcome. One criticism of using IHC for ERBB2 to define these tumors is that this assay was not designed to distinguish ERBB2-low from ERBB2-negative tumors but rather to distinguish the ERBB2-positive tumors that respond to traditional monoclonal antibodies such as trastuzumab. 21,33 This is further complicated by the variability in ERBB2 scoring from institution to institution, reflected in the variability in proportions of ERBB2-low tumors from study to study 11,28,34 and the lower rates of ERBB2-low diagnoses at academic centers seen in our study. Indeed, there is a pressing need to more precisely quantify low levels of ERBB2 expression to identify patients who might benefit from potent antibody-drug conjugates, as reflected in the DAISY trial, 35 in which even patients classified as ERBB2-negative responded to T-DXd. Our analy-sis of ERBB2 copy number and ERBB2/CEP17 ratio did not offer promising discriminative ability for ERBB2-low disease. However, other surrogate molecular markers such as quantitative polymerase chain reaction or flow cytometry may help to more accurately identify candidates for ERBB2-directed antibody-drug conjugates. 36

Strengths and Limitations
This study has strengths, including the large patient cohort available for analysis, which is reflective of most cancer diagnoses across the US. This study has several important limitations that should be considered when interpreting these results. The NCDB includes only OS data, which limits associations of ERBB2-low status with cancer-specific prognosis, especially in the hormone receptor-positive cohort, where survival may lag years behind recurrence. Additionally, the NCDB lacks centralized assessment of ERBB2 status via IHC, perhaps highlighted by the different proportions of ERBB2-low disease in different institution types; therefore, some of the results may be associated with regional variation in practice of classifying cases as ERBB2 0 vs ERBB2 1+ as scored via IHC. Our correlation of ERBB2-low status with ISH results is limited by the fact that most patients with ERBB2 expression scored as 0 or 1+ via IHC did not have ISH performed and therefore must be interpreted with caution.

Conclusions
The results of this cohort study suggest that there are associations between ERBB2 expression and hormone receptor expression and ductal tumor histological status but only slight differences in response to treatment and prognosis. These findings do not support classification of ERBB2-low breast cancer as a distinct clinical subtype. Further improvements are needed in molecular stratification of ERBB2 expression to understand the clinical significance of ERBB2-low breast cancer and identify patients who can benefit from novel therapies. Moving forward, the availability and use of ERBB2-directed antibody-drug conjugates will likely drive prognosis in ERBB2-low disease, rather than intrinsic differences in biology between ERBB2-low and ERBB2-negative breast cancer.