HER2-positive apocrine carcinoma of the breast: a population-based analysis of treatment and outcome

Purpose Apocrine carcinoma of the breast (APO) expresses HER2 in 30–50% of cases. This study explored the clinicopathological features and outcome of HER2+/APO and matched HER2+/NST cohort. Methods We used the SEER database to explore the cohorts. Univariate and multivariate analyses were used to assess the survival. Based on ER and PR [steroid receptors/SR/] and HER2 status, we divided the cohorts to match the intrinsic molecular subtypes for comparisons. Results We retrieved 259 cases of HER2+/APO. Most HER2+/APO were SR negative (65%). HER2+/APO were more prevalent in the 80+ age group (24.7% vs. 15.7%, p < 0.001). HER2+/SR−/APO had a significantly lower histological grade than the HER2+/SR−/NST (p < 0.001). Breast cancer-related deaths were more prevalent in HER2+/NST (7.8% vs. 3.9%, p = 0.019). This was particularly evident between SR− subgroups (10.4% in HER2+/SR−/NST vs. 4.2% in HER2+/SR−/APO, p = 0.008) and was reaffirmed in breast cancer-specific survival in univariate analysis (p = 0.03). Other than race and SR status, HER2+/APO subgroups did not differ in clinicopathological parameters. Conclusions Our study confirms the rarity of the APO and reveals that SR status in APO does not affect these patients' prognosis. HER2+/APO tumors tend to have a less aggressive phenotype and a more favorable outcome despite a markedly lower ER/PR positivity.

The available clinical data on APO are contradictory due to the rarity of the disease and the use of different diagnostic criteria for APO [14,19,21]. Several recent studies have reported a worse clinical outcome in patients with APO than invasive breast carcinoma of no special type (IBC NST) [4,24]. However, in an analysis of the Surveillance, Epidemiology, and End Results (SEER) population-based data, the clinical outcome for the APO patients was significantly better than NST patients following the adjustment for demographic and clinicopathological characteristics [24]. Some studies also revealed more favorable overall survival (OS) and breast cancer-specific survival (BCSS) for patients with AR-positive triple-negative APO compared with other triplenegative breast carcinomas (TNBC) [10][11][12]25]. In contrast, other studies found no significant differences [5,13]. Wu et al. recently reported better clinical outcomes of triplenegative APO patients than TNBC NST [23]. The authors also found chemotherapy associated with a more favorable outcome among the triple-negative APO patients [23].
Despite the common HER2 expression, most of the available clinical studies have been focused on the triplenegative APO, while the clinical and survival data on HER2+/APO are sparse [7,8]. The current study explored the association between APO subtypes and survival after adjusting for all other prominent clinical and demographic predictors of survival among SEER patients.

Patients' selection and cohorts
We explored the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) database to select our study cohort. The SEER database includes data on patients' demographics, tumor characteristics [Histotype, tumor grade, TNM stage (AJCC), tumor size, lymph node status, and distant metastases], the first course of treatment, treatment options (surgery, chemotherapy, radiotherapy), and follow-up for vital status. It encompasses data from eighteen population-based cancer registries covering approximately 1/3 of the US population.
For our study cohort, we selected patients diagnosed with histologically confirmed invasive HER2-positive APO (HER2+/APO) between 2010 and 2016. In addition, we selected patients diagnosed with IBC NST with HER2 positivity (HER2+/NST) within the same period. The two groups (HER2+/APO and HER2+/NST) were further divided based on the SEER variable that codes ER and Progesterone Receptor (PR) [Steroid receptors/SR/] and HER2 status to form groups based on the intrinsic molecular subtypes, i.e., Apocrine Luminal B (HER2+/SR+/APO) and Apocrine HER2-Enriched (HER2+/SR−/APO) for HER2+/ APO and HER2+/SR+/NST and HER2+/SR−/NST for HER2+/NST part of the cohort. SR status did not include androgen receptor. The study excluded in situ carcinomas (both apocrine and NST) and other special types of invasive breast carcinoma with HER2 positivity. We used the SEER*stat version 8.3.2 to generate a case-listing file.

Statistical analysis
Patients' demographical and clinicopathological characteristics, including age, hormone receptor status, stage, tumor grade, therapies (surgery, chemotherapy, radiotherapy), regional and distant metastases, as well as metastatic sites, were summarized by absolute and relative frequencies. Age was categorized as 18-49, 50-79, and ≥ 80 years. The difference in the distribution of clinicopathological characteristics was tested using Pearson Chi-square and/or Fisher's exact test, whichever was appropriate according to the expected cell counts. For continuous normally distributed variables, we give mean and standard deviation as measures of central tendency and variability; for those whose distribution is not normal, we give a median and interquartile range instead. We tested the equality of means/medians for normally/nonnormally distributed variables using the t test/Mann-Whitney U test.
BCSS and OS were defined as a time in years between the cancer diagnosis and death from breast cancer, death due to any cause, respectively. Patients alive at the end of the follow-up were censored for both types of survival. We ran univariate Cox proportional hazards analysis to identify candidate prognostic factors for the multivariate analysis. For each of these, we produced Kaplan-Meier plots and performed log-rank tests after checking the appropriateness of proportional hazard assumption graphically using martingale and weighted Schoenfeld residuals. After identifying and deciding which prognostic factors will enter the multivariate analyses, we fitted a multivariate Cox regression model to estimate the effect of subtypes on survival while adjusting for other prognostic factors. In the multivariate analysis of BCSS and OS, aside from the HER2+/APO status, we included age (categorical), stage, tumor grade, surgery, and systemic therapies. To provide an additional check of the robustness of our findings w.r.t. modeling assumptions assumed for multivariate Cox regression, we applied propensity score caliper matching in the ratio of 3:1 in favor of the "control" group (HER2+/NST). The propensity score was estimated via a multivariate logistic regression model that included age (in years, continuous), tumor grade, stage, size (in cm), lymph node status, the presence of distant metastasis, chemotherapy and radiation therapy. We applied the same model for propensity score and the same matching strategy for APO vs. NST comparisons (overall, SR+ and SR−) and the analysis for the effect of SR status among the APO subtype. All analyses were performed based on available cases using R Statistical Software (Foundation for Statistical Computing, Vienna, Austria) version 3.5.1. All tests were 2-sided, with p < 0.05 considered statistically significant.

HER2+/APO vs. HER2+/NST cohorts
The clinicopathological characteristics of the cohorts are summarized in Table 1.
The two groups did not differ significantly regarding the race, tumor stage, regional lymph node status or the presence of distant metastases (in general and in specific sites). The treatment options (surgery and chemotherapy) were similar between HER2+/APO and HER2+/NST (p > 0.05), while radiotherapy was more commonly used in the HER2+/APO group compared with HER2+/NST (p = 0.05) ( Table 1).
Although at 18 months and onwards the OS estimate in the HER2+/APO group was consistently higher, with unadjusted HR 1.26 (0.83, 1.92) in the univariate Cox proportional hazards model. This difference was not statistically significant (p = 0.264) (Fig. 1A). Nevertheless, the positive effect of apocrine morphology was more evident in BCSS analysis with unadjusted HR 1.83 (0.99, 3.42), hence 83% higher hazard for BCSS death in the HER2+/NST group (Fig. 1B).
Breast cancer as a cause of death was also more prevalent in HER2+/NST than in the HER2+/APO group (7.8% vs. 3.9%, p = 0.019). This difference was 1.5 times higher between ER-negative subgroups (4.2% in HER2+/SR−/APO vs. 10.4% in HER2+/SR−/NST, p = 0.008). It was reflected in the results of the univariate analysis of BCSS in those groups with unadjusted HR 2. 22
The survival curves for two groups defined by SR status (ER and PR) were not significantly different by the log-rank test (Fig. 3), although their shape was detectable by the log-rank test statistic. This was the case for both OS [HR 0.6 (0.54, 1.57)] and BCSS [HR 0.7 (0.18, 2.74)] in favor of SR− positive status. The patients aged ≥ 80 years had worse OS in univariate analysis (p < 0.001); however, this age-related difference in survival curves completely dissipated when it came to BCSS (p = 0.423).
Except for PR, which was less frequent in the HER2+/ SR+/APO cohort, the two groups did not differ in clinicopathological parameters, including the treatment modalities. Although in univariate analysis no significant differences were observed, multivariate Cox proportional hazards analysis revealed that the apocrine morphology in SR+/HER2+ carcinomas was associated with better OS (HR = 2.62, CI = 1.17-5.84, p = 0.018) and BCSS (HR = 3.14, CI = 1.01-9.76, p = 0.047) ( Table 4).
The cause of death was also significantly different between the groups, as only 4.2% of patients in the HER2+/ SR−/APO group died of breast cancer, while 10.4% died of breast cancer in the HER2+/SR−/NST group (p = 0.008). Consequently, the apocrine morphology was associated with a better BCSS in univariate analysis (p = 0.030) (Fig. 4). However, multivariate analysis did not reveal a significant difference in OS and BCSS between the HER2+/SR−/APO and NST groups.

Discussion
APO is a rare (frequency ~ 1%) subtype of breast carcinoma whose diagnosis and clinical data, including outcome, remain challenging and controversial [14,19]. Based on the SEER data (2010-2016), the current study represents the largest HER2-positive APO cohort reported to date.
Our findings indicate that HER2-positive APO do substantially better compared with HER2-positive NST carcinomas. This is mainly reflected in BCSS and was independent of ER and PR status. Our results are in line with a recently published SEER cohort of triple-negative APO, which also revealed that these cancers had a better prognosis than TNBC NST [23]. Several other independent studies confirmed a more favorable outcome for triple-negative APO patients [1,17]. In contrast, the studies of Dellapasqua NST no special type, APO apocrine, NA not available, SR steroid receptors (ER and PR), RLN regional lymph node, IQR interquartile range Only significant p-values are bolded  [16]. Some other findings from our study are consistent with the previous data, including a rarity of APO and a higher prevalence of APO among elderly patients [14,21,24]. APO typically lacks ER and PR receptors [6,14,21], as confirmed in our study, given that 2/3 of the cases were negative for ER and PR, which was significantly lower than in the HER2+/ NST cohort. In contrast, APO consistently overexpress AR [6], but the AR status was not provided in the SEER database. This is understandable given the timeline of the collected data (2010-2016) and the fact that AR testing was only recently incorporated into a recommended diagnostic work-up of the breast's APO [14,19]. Consequently, some of the tumors in the study, particularly ER/PR-positive, may not be true molecular APO. Nevertheless, our subgroup analysis (HER2-enriched vs. Luminal B APO) revealed no significant differences between the two groups regarding clinicopathological parameters and survival (both OS and BCSS).
We observed some clinically relevant differences between the APO and NST groups. Thus, HER2+/SR−/APO had significantly lower histological grade than the NST carcinomas, which is in line with several previous studies and is probably   (13.2) due to the lower mitotic activity (and lower Ki-67 labeling) of apocrine cells [9,17,19,20]. Lower tumor grade may also contribute to the less aggressive behavior of HER2+/ APO, as confirmed in a large cohort of IBS NST in which a high histological grade was a strong predictor of an adverse outcome [15]. In addition to this finding, the more advanced age of HER2+/APO patients probably contributed to substantially lower chemotherapy use in this group of patients. Nevertheless, the chemotherapy use did not affect the outcome, given a significantly better BCSS among HER2enriched APO than in the HER2-enriched NST cohort.
The current study has several limitations that reflect the nature of the SEER database. Firstly, the follow-up period was too short (~ 31 months), and long-term outcomes could not be assessed. This is particularly relevant for SR+ cases in both cohorts. Secondly, AR was not routinely provided, which could affect the classification of a proportion of HER2+/APO cases, particularly ER/PRpositive. In addition, the details (types and duration) on chemotherapy, endocrine, and anti-HER2 therapies were not provided in the SEER database, so their effects (sensitivity, resistance) and the overall impact on the outcomes could not be analyzed.
We conclude that our cohort represents the largest HER2+/APO study reported to date. It confirmed the rarity of the HER2+/APO and revealed only marginal differences within the HER2+/APO regardless of ER/PR status. Compared with HER2+/NST, HER2+/APO tended to have a less aggressive phenotype and were associated with a more favorable clinical outcome (BCSS) despite a markedly lower/absent ER/PR expression. Further prospective studies should confirm the provided observations.  Funding Open Access funding provided by the Qatar National Library.

Data availability
The datasets from the study can be obtained from the corresponding author on reasonable request.