Patterns of Failure in Triple Negative Breast Cancer Patients in an Urban, Predominately Black Population

Triple negative breast cancers (TNBC) behave more aggressively than hormone-receptor positive breast cancers. They are also known preferentially to affect young black women, often leading to poorer outcomes compared with those for white women. We sought to evaluate the comprehensive patterns of failure associated with treatment for TNBC at an urban institution with a predominantly black population and to assess the impact of social determinants of health on treatment failure. A retrospective review of TNBC patients treated from 2005 to 2015 was conducted. Detailed patient, tumor, and treatment characteristics and information on patterns of failure were included. With a median follow-up of 46 months, 32 (16%) documented failures occurred. Locoregional failures comprised 84% of failure patterns whether isolated or in combination with distant failure. Treatment failure was associated with insurance type and smoking status, as well as several tumor characteristics. On multivariate analysis, pathologic nodal staging was the most significant predictor of treatment failure. In contrast to previous studies, we found that black women had higher overall survival than white women, but race was not associated with differences in recurrence patterns or with likelihood of treatment failure. Regardless of race, of the patients who recurred, 53% failed in distant and locoregional sites simultaneously, with an additional 34% failing locally only. These results highlight the need for aggressive local therapies in high-risk patients and suggest a need for improved follow-up focusing on detecting locoregional failures. Multidisciplinary care is essential in the management of these patients at time of failure.

more aggressively than hormone-receptor positive breast cancers [1][2][3]. These tumors are negative for estrogen-receptor (ER), progesterone-receptor (PR), and human epidermal growth factor receptor 2 (HER-2) and tend to occur more commonly in black women and women < 40 years old [4,5]. Twenty percent of triple negative breast cancer patients will have a BRCA mutation present [6]. The vast majority of TNBC are in the basal-like subtype as defined by gene expression profiling [7]. These cancers tend to present aggressively with rapid growth and are diagnosed clinically more often than radiographically [5]. Patients have improved disease-free survival with the addition of neoadjuvant chemotherapy when pCR is achieved, especially in the locally advanced setting [8][9][10]. However, TNBC tumors generally relapse earlier after standard chemotherapy regimens and have a higher tendency for visceral and soft tissue involvement when metastasized [11]. Death within 2 years of diagnosis is much higher for women with triple negative breast cancer, though this recurrence rate starts to plateau when women with triple negative breast cancer survive longer than 2 years [5,12]. Moreover, patients who do not achieve pathologic complete response (pCR) have a poorer outcome relative to patients with receptor positive disease [11].
To date, most studies evaluating patterns of failure have treated failure patterns as discrete entities consisting of local, regional, or distant metastases [12][13][14]. However, locoregional recurrence is often underreported when distant metastases are found. Locoregional recurrences can cause significant morbidity and decreased quality of life for patients. As such, we sought to evaluate the comprehensive patterns of failure associated with treatment for triple negative breast cancer at a single urban institution. With this information, we hypothesized that further understanding of patterns of failure could inform future treatment directions and follow-up paradigms based upon the most likely site of failure.

Patient Selection
After institutional review board (IRB) approval, a health insurance portability and accountability act (HIPAA) compliant retrospective database was constructed to include all patients with triple negative breast cancer who completed their initial evaluation including pathologic diagnosis and all subsequent treatments at our institution (University of Maryland Greenebaum Comprehensive Cancer Center) between January 1, 2005, and December 31, 2015. Patients were included if they had a diagnosis of invasive ductal carcinoma (IDC) or invasive lobular carcinoma (ILC) with ER, PR, and Her-2 receptor proven negative by immunohistochemistry (IHC) or fluorescence in situ hybridization who received chemotherapy, surgery, and/or radiation therapy at our institution. ER, PR, and Her-2 negativity was considered < 1% IHC expression [15,16]. Other histologies included metaplastic (n = 1) and anaplastic carcinoma (n = 1). Exclusion criteria included patients any positivity (> 1%) in ER or PR, or Her-2 amplification, patients receiving treatment outside of our institution, final pathologic diagnosis of a non-breast primary origin, and any patient with DCIS histology without an invasive component.
Patient characteristics collected included age at diagnosis, sex, race, marital status, zip code (a surrogate for income), insurance status, BMI, smoking status, smoking pack-years, age at menarche, age at menopause, gravida, parity, oral contraceptive use, hormone replacement, HIV status, diabetes status, BRCA mutation status, and cardiovascular disease or risk factors. Tumor characteristics included histology, tumor location, radiographic or clinical detection, AJCC 7th edition clinical and pathologic tumor (T) stage, T size (cm), clinical and pathologic nodal (N) stage, metastatic (M) stage, overall clinical and pathologic stage grouping, histology, Ki-67 (%), overall grade, and presence or absence of lymphovascular space invasion (LVSI).

Patterns of Failure
For all patients who developed a disease failure on the basis of imaging (CT or PET) and/or biopsy, we collected information on patterns of treatment failure, including local, regional, distant, and combinations of these three (local and regional, local, regional, and distant, local and distant, and regional and distant). Documented sites of first failure and distant failure sites were included. Local failure was defined as a failure within the breast or chest wall. Regional failure was defined as any failure in the axilla, supraclavicular lymph node basins, or internal mammary lymph node chain.

Treatment
Medical and surgical oncology data included chemotherapy type, neoadjuvant versus adjuvant chemotherapy administration, dose reductions in chemotherapy, surgical procedure (lumpectomy vs mastectomy) and axillary assessment (sentinel lymph node vs axillary dissection), and pathology response (i.e., complete, partial, or no response).
Radiation oncology data included total radiation dose in Gy, dose per fraction, technique of treatment delivery, treatment volumes (breast/chest wall vs comprehensive nodal irradiation), dose and volume (dose volume histogram data) for the heart and lungs, and acute and late toxicities of treatment.

Statistical Analysis
Categorical variables are presented as numbers (percentages), and continuous variables are presented as medians with ranges. Univariate and multivariate analysis were used to assess for prognostic clinicopathologic factors, and chi-square testing (Pearson's) was used to assess for differences between failure and non-failure groups. Logistic regression with Kaplan-Meier estimate was used to calculate overall survival (OS) and freedom from recurrence (FFR). A p value < 0.05 was considered statistically significant. All univariate variables that were significant were entered into a forward conditional multivariable analysis to correct for any confounding variables. All analyses were performed using SPSS Statistical Software (Version 21, IBM Corporation).

Patient and Tumor Characteristics
A total of 197 women met inclusion criteria, with a median age of 54 years (range 22-86). Race was predominantly black (64%), 41% were married, and 64% of patients treated had private insurance. Comprehensive patient, tumor, and treatment characteristics stratified by failure status are outlined in Table 1.
In total, thirty-two patients (16%) in our cohort experienced treatment failure. Insurance type (p = 0.002) and active smoking (p = 0.002) differed between the two groups, but race and marital status were not found to be significantly different between the patients who failed treatment and those who did not (p = 0.36 and 0.94, respectively). The tumor histology was predominately (99%) IDC, the vast majority (85%) were grade 3, and LVSI was present in 28% of patients. Clinical (self-palpation) detection occurred in 73% of women. On chisquare testing, presence of LVSI (p = 0.036) and the method of tumor detection (p = 0.003) were both significantly different between failure and non-failure groups. Grade and histology were not associated with an increased risk of treatment failure in our patient cohort (p = 0.19 and 0.37, respectively) ( Table 1).
According to AJCC 7th edition, the most common clinical stage was II (47%), followed by stage I (32%). Prevalence of failure increased with increasing stage: 15% in stage I, 18% in stage II, and 24% in stage III patients. Of patients who failed treatment, 28% were stage I, 47% were stage II, and 25% were stage III.
There was a significant difference in pathologic staging between those who failed and those who did not fail treatment (p < 0.001). Largest tumor size was also documented and was stratified into ≤ 2 cm and > 2 cm. Forty-nine percent of tumors were > 2 cm, and increasing tumor size was significantly associated with treatment failure (p < 0.001). Pathologic tumor stage was significantly different between the two groups (p = 0.005); 35% of pT3/T4 patients failed compared with only 16% of pT1/T2 patients. Pathologic nodal stage was highly prognostic (p < 0.001) with 36% of treatment failure patients having either pN2 or pN3 pathologic nodal status compared with only 8% of the non-failure group.

Treatment
Almost all patients (95%) underwent surgical resection with lumpectomy performed in 53% of patients. Rates of sentinel lymph node biopsy and axillary dissection were not different between groups (p = 0.76). Thirty-one percent of women received neoadjuvant chemotherapy, while 63% of women received adjuvant chemotherapy. Neoadjuvant chemotherapy was not associated with decreased rates of treatment failure (p = 0.424). Adjuvant radiation was given in 62% of patients to a median dose of 60 Gy (range 10.8-70.4), and there was no difference in failure for those receiving < 60 Gy (p = 0.14). Radiation treatment modality (3D-CRT vs IMRT) did not differ between those who failed and did not fail treatment. Table 2 outlines comprehensive details for the 32 patients who failed treatment including site and pattern of failure, as well as subsequent treatments rendered. Patterns of failure in our TNBC cohort were varied and complex. The largest proportion of patients (53%) failed in distant and locoregional sites simultaneously, whereas 34% of patients developed local breast or chest wall recurrence only. Of those patients who had any distant failure, median OS was 23.5 months. Failure patterns did not differ by race (p = 0.41). The most common sites of failure were the lungs (28%), followed by the liver (25%) and bone and brain (22% each). Eight patients (25%) received further surgery for locoregional recurrences, while 17 (53%) received salvage chemotherapy at the time of relapse. The most common salvage chemotherapy regimens included carboplatin/gemcitabine, eribulin, and ixabepilone. Median OS from time of initiation of treatment for all patients who failed was 30 months, while median OS from time of failure for these patients was 8 months. Of those who recurred, black patients had a median OS of 32 months compared with 23 months for white patients, but this was not statistically significant (p = 0.23). Moreover, median overall survival for the entire cohort was also higher for black patients than white patients (50 vs 31.5 months).

Survival
Median follow-up for our entire cohort was 46 months (range 1-353 months). The median, 2-year and 5-year overall  .020), pathologic nodal stage (p = 0.001), and overall pathologic stage (p < 0.001) were also associated with treatment failure. In addition, patients with tumor size > 2 cm had significantly higher rates of failure (OR 6.996, 95% CI 2.293-21.342, p = 0.001). Treatment modalities including receipt of chemotherapy (both neoadjuvant and adjuvant), type of surgery, and radiation therapy dose were not found to be associated with an increased risk of treatment failure (Table 3).
On MVA, pathologic nodal staging was most strongly associated with treatment failure (p = 0.017) when adjusted for covariates of pathologic T stage, insurance and smoking status, and LVSI. Pathologic N2 stage, specifically, had a significant association with treatment failure (OR 25.0, 95% CI 3.454-180.972, p = 0.001) ( Table 4).

Discussion
In our retrospective analysis of TNBC patients, we found that 16% of patients failed initial treatment. We specifically further characterized failure patterns and found that locoregional All of our patients were women, predominantly black, and most underwent surgery and received anthracycline/taxanebased chemotherapy (neoadjuvant or adjuvant). The median time to failure was 16 months. Factors found to be predictive of failure in our univariate analyses included insurance, smoking status, presence of LVSI, clinical detection (palpating a lump) of tumor, increasing clinical tumor size (> 2 cm), and increasing pathologic tumor stage, nodal stage, and overall staging. When adjusted for covariates in MVA, increasing pathologic nodal staging was found to be the most significant predictor of treatment failure. Patients with increasing pathologic nodal involvement should be treated most aggressively, with neoadjuvant chemotherapy to prevent distant spread of disease, followed by surgical resection and adjuvant radiation. For patients with disease progression after initial local therapy, further systemic adjuvant chemotherapy is warranted.
This rate of failure (16%) is consistent with other reported literature with rates ranging from 4 to 33.9% [12,14,17]. In addition, we found an isolated locoregional recurrence pattern rate of 7.6%. Review of literature found wide variability in rates of locoregional only recurrences (3-22%) in TNBC patients. This heterogeneity stems from both differences in retrospective patient populations as well as from differences in their treatment regimens [18][19][20][21].
TNBC tends to metastasize more frequently to viscera compared with hormone-receptor positive cancers [11]. In our study, we found the distant recurrence only rate was 16% (5/32). Visceral organs accounted for the primary site of first relapse in our analysis. The most common sites of distant failure in our study included the lungs (28%), liver (25%), and bone and brain (22%). This is consistent with two prior reports which both described lung metastases as the most common sites of failure in TNBC patients [14,22]. Specifically Kennecke found that of the patients with TNBC who had distant disease, 42.8% had lung, 25.2% had brain, and 21.4% had liver metastases [22].
Other studies have reported distant recurrent rates of 16.9%-33.9% [12,17]. The cumulative distant recurrence rate in this cohort of failures was 8.6% (17/197), slightly lower than the earlier reported studies. Increased modalities to stage patients may account for the ability to detect both locoregional     and distant metastases, especially with increased usage of PET/CT which has higher sensitivity and specificity in detecting distant metastases which may have been missed with less sensitive imaging techniques in the past [23][24][25]. Previous analyses studied patients from 1999 to 2008 [17] and 1987 to 1997 [12]. Of note, our 5-year freedom from recurrence rate was nearly 84%. While many of the previous studies had longer follow-up times than our study (46 months), the time for recurrence in TNBC is often within 1-3 years of diagnosis [3,12], so it is unlikely that our study significantly underreported recurrence rates. Dent et al. found that in non-TNBC cancers, the recurrence risk can continue to increase even after 5 years [12]. Consistent with this unique pattern of early recurrence, 21 (66%) and 24 (75%) of the recurrences in our analysis occurred within 2 and 3 years or treatment initiation, respectively. Interestingly, nearly 20% of our recurrences happened after 5 years, even though documented analyses suggest a reduction in recurrence risk after a few years of initially high risk of recurrence [11,12]. In terms of survival outcomes, our study found that the 5year overall survival was 76.4%. Of those patients who lived longer than 5 years with TNBC, only 6/76 (7.9%) passed away from breast cancer, with an additional 2 patients dying of other causes. For all patients who had disease recurrence, the median survival time was 30 months, decreasing to a median of 23.5 months in those with any distant failure. Some previously published data suggests that the median survival time for metastatic TNBC patients is roughly 12 months [5,26], which is much lower than our cohort. This may reflect improvement in systemic therapy options over time. Ovcaricek and Steward reported an overall 5-year survival of 74.5% [27] and 76.3% [17], respectively, which is in line with our reported survival. In the latter study, median time to recurrence was 18.8 months, similar to the median time to recurrence in our cohort (16 months).
In our retrospective analysis, we also wanted to elucidate potential socioeconomic, tumor, and treatment factors that could potentially predict recurrence risk. Nodal status, tumor size, and LVSI are well-documented prognostic tumor characteristics [27][28][29]. We corroborated these findings on UVA showing that all three of these tumor characteristics were statistically significant predictors of recurrence. In addition, because of our unique population in an inner-city with a predominantly black population, we also sought to better understand socioeconomic factors that could potentially predict those at higher risk of recurrence. TNBC is known to preferentially affect young black women, and these women often have poorer outcomes than white women [30]. Possible explanations for a race-based disparity in survival have been linked to lack of screening, socioeconomic status, access to insurance, and differences in treatment [31,32]. Interestingly, we found that the median overall survival of black patients was 50 months as compared with 31.5 months for white patients. In those who failed treatment, black patients had a median OS of 32 versus 23 months for white patients who recurred, though this was not statistically significant (p = 0.23). These findings may reflect the higher proportion of black patients in our cohort which tend to be underrepresented in studies. Moreover, given the urban and predominantly black population where the majority of our patients are coming from, it is possible that black communities worked together to promote more regular cancer screening and treatment follow-up. For example, free health fairs and health promotion activities are regular occurrences with active engagement by members of our community.
Locoregional failure comprised 84% of failure patterns whether isolated or in combination with distant failure. This number is quite high and argues for the importance of locoregional therapy and possibly escalation of locoregional therapy for some patients. While death is typically caused by distant spread, locoregional disease can have a significant impact on morbidity and patient quality of life. For example, when women develop a dermal-lymphatic skin recurrence wound, care is often required. Several publications have suggested a critical role of radiation therapy in TNBC. Wang et al. performed a randomized trial of women with anatomic stage I and II breast cancer who underwent mastectomy followed by chemotherapy or chemotherapy and radiation therapy [33]. Adjuvant chemotherapy and radiation therapy were associated with improved overall survival and recurrence-free survival. Abdulkarim and colleagues performed a retrospective analysis of 768 patients treated at their center for TNBC [34]. At a median follow-up of 7.2 years, they found that women with anatomic T1-2N0 TNBC treated with mastectomy ± chemotherapy but no radiation therapy had increased risks of locoregional recurrence compared with the same women who elected breast conserving therapy with lumpectomy and radiation therapy suggesting an integral role of radiation therapy in the management of this disease. On the contrary, Dragun and colleagues found no difference in progression free-survival and locoregional-free survival in their cohort of patients treated with and without radiation therapy during the first 3 years of followup [35]. These recurrence patterns provide an opportunity for escalation of therapy in high-risk patients such as addition of radiation therapy for all post-mastectomy TNBC patients, doseescalation of radiation therapy, or addition of chemotherapy during radiation therapy for patients.
Our data support prior work showing increased rates of locoregional failures in the TNBC patient population. Despite this body of data, however, current follow-up guidelines support breast only imaging in the lumpectomy setting (no systemic imaging) and no imaging in mastectomy patients or patients who have undergone reconstruction [36]. This data show increasing rates of recurrence by anatomic tumor stage as well as increasing pathologic T and N stage. Thirty-five percent of patients with pT3/4 tumors experienced failure as did 47% of women with pN2/3 disease. We propose that these high recurrence rates warrant both breast/chest wall and systemic imaging. In most solid tumor sites, systemic or local imaging is performed when recurrence rates enter this range. Potential early detection of recurrence has the possibility to not only improve long term cancer outcomes but also allow for additional therapies and/or enrollment on clinical trials. Additionally, earlier detection of recurrence may also improve patient morbidity and quality of life outcomes.
Our study has some limitations, the first of which is the retrospective nature of our work. While we have meticulously maintained this database and incorporated as many potential clinical and demographic variables as available, it is possible that we may not have included all factors that could influence access to care and/or timely initiation of treatment and that could contribute to treatment failure. However, through a deep literature review, we have incorporated the well-known factors associated with treatment failure in TNBC. Additionally, we have included a wide range of TNBC patients, which may include lower risk older women. However, our higher proportion of black women in this study who are higher risk likely balances out this difference. The relatively small number of patients in our cohort and additionally our relatively low number of failures may decrease our power to detect factors associated with recurrence.

Conclusion
With modern BC therapies, we show good treatment outcomes for patients with TNBC, with 84% of patients free of disease at 5 years from initial diagnosis. Black women had higher overall survival than white women in our cohort, but race was not associated with differences in recurrence patterns or with likelihood of treatment failure. The patterns of failure in TNBC are complex, with 84% of patients experiencing locoregional failure as a portion of their recurrence pattern but 53% of patients experiencing it combined with distant failure. Improvements in local and systemic therapies as well as follow-up approaches including consideration of initial (2-3 years) post diagnosis periodic restaging are needed to allow for earlier detection of locoregional and systemic recurrence, thereby improving morbidity and quality of life of patients with TNBC.
Author's Contributions HRC and SRR compiled and maintained the database and contributed equally to the writing of this manuscript. EMN provided valuable feedback prior to submission. All authors read and approved the final manuscript.
Data Availability The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Compliance with Ethical Standards
Conflicts of Interest The authors declare that they have no conflicts of interest.
Ethics Approval and Consent to Participate All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. The study was approved by the ethics committee affiliated with the institutional review board of the University of Maryland Baltimore.
Consent for Publication Not applicable.