Neighborhood and racial influences on triple negative breast cancer: evidence from Northeast Ohio

Triple negative breast cancer (TNBC) is an aggressive subtype of breast cancer (BC) with higher recurrence rates and poorer prognoses and most prevalent among non-Hispanic Black women. Studies of multiple health conditions and care processes suggest that neighborhood socioeconomic position is a key driver of health disparities. We examined roles of patients’ neighborhood-level characteristics and race on prevalence, stage at diagnosis, and mortality among patients diagnosed with BC at a large safety-net healthcare system in Northeast Ohio. We used tumor registry to identify BC cases from 2007 to 2020 and electronic health records and American Community Survey for individual- and area-level factors. We performed multivariable regression analyses to estimate associations between neighborhood-level characteristics, measured by the Area Deprivation Index (ADI), race and comparative TNBC prevalence, stage at diagnosis, and total mortality. TNBC was more common among non-Hispanic Black (53.7%) vs. non-Hispanic white patients (46.4%). Race and ADI were individually significant predictors of TNBC prevalence, stage at diagnosis, and total mortality. Race remained significantly associated with TNBC subtype, adjusting for covariates. Accounting for TNBC status, a more disadvantaged neighborhood was significantly associated with a worse stage at diagnosis and higher death rates. Our findings suggest that both neighborhood socioeconomic position and race are strongly associated with TNBC vs. other BC subtypes. The burden of TNBC appears to be highest among Black women in the most socioeconomically disadvantaged neighborhoods. Our study suggests a complex interplay of social conditions and biological disease characteristics contributing to racial disparities in BC outcomes.

To better understand the growing evidence of higher burden of TNBC incidence among non-Hispanic Black women, it is critical to examine the contribution of both individual and environmental factors to disparities in TNBC incidence and mortality [24]. Patients' multifaceted individual-level characteristics, including molecular/epigenetic factors, health status and behaviors, interact with social circumstances, including family structures, neighborhood-level characteristics, and public policy environments (Fig. 1). Additionally, prominent medical professional organizations and publications have recently issued directives guiding scientists and authors (1) to more carefully denote the reason and basis of measuring variables that classify individuals (including race and ethnicity and socioeconomic position) and (2) to rigorously evaluate the contributions of these social indicators to health and disease incidence and outcomes [25,26].
Research has clearly established the notable and often substantial influence of neighborhood socioeconomic circumstances (such as the area deprivation index or ADI) on disease incidence [27], care patterns [28][29][30][31][32], and outcomes across many health conditions [33][34][35] for various health conditions, including multiple cancers [33,[36][37][38][39][40][41][42][43][44][45]. Even while accounting for individual characteristics and behaviors, various person, context, and population factors are understood to converge to produce different outcomes along the cancer care continuum (Fig. 1). This conceptualization of health disparities has been applied when examining subtypes of BC by hormone status [46][47][48], including TNBC [49][50][51][52][53]. Using data from state cancer registries and large patient cohorts, these studies showed that living in more disadvantaged neighborhoods was significantly associated with a higher comparative prevalence of TNBC and a more advanced stage at diagnosis. These studies, together, underscored the significant role of neighborhood characteristics in contextualizing each individual-level risk factor for TNBC and in informing current practices and interventions to deliver cancer care more effectively.
Our study expands this body of knowledge by investigating the role of patients' racial identity and neighborhoodlevel characteristics on the incidence and stage at diagnosis, and mortality among patients diagnosed with TNBC at a large safety-net hospital system in Northeast Ohio, a region with higher incidence and mortality rates of BC than those of the U.S. average [54]. We hypothesize a higher proportion Fig. 1 Conceptual framework of TNBC cases among Black patients than White patients and more late-stage diseases and higher mortality among patients living in more disadvantaged neighborhoods than those living in less disadvantaged neighborhoods.

Data sources
We drew primary data from the tumor registry of the Metro-Health System (MHS). The MHS tumor registry collects patients' demographic characteristics (age at diagnosis, race and ethnicity, marital status, and residential address at diagnosis), health insurance information at diagnosis, date of diagnosis, histopathology (i.e., tumor type, tumor stage, grade of differentiation, and receptor status), and date of death or last follow-up and reports this information to the state cancer registry, the Ohio Cancer Incidence Surveillance System (OCISS) [55]. The MHS tumor registry complies with the North American Association of Central Cancer Registries (NAACCR) to ensure collection and curation of high-quality data. The registry data were linked with MHS electronic health records (EHRs), using patients' medical record numbers, to compile data on patients' health conditions, including smoking history and diagnosis of conditions included in the Elixhauser Comorbidity Index (ECI) [56,57]. We additionally geocoded patients' residential addresses to the census tract to examine neighborhood-level data from the American Community Survey (ACS) [43,58]. Given the deidentified nature of the tumor registry data and the ACS data, the MetroHealth Institutional Review Board exempted the ethics review.
The MHS is Cuyahoga County's safety-net health system and serves more than 300,000 patients residing in the county, two-thirds of whom are uninsured or covered by Medicare or Medicaid [59,60]. The burden of BC has been increasing in Cuyahoga County with a steady increase in BC incidence rates since 2009. While the highest incidence rates were observed among White patients, Black patients have the highest age-adjusted death rates due to BC, which was about twice as high as the lowest rate among Hispanic women in 2018 [61].
Cuyahoga County is uniquely characterized by three distinct hospital systems: Cleveland Clinic, an internationally recognized health system with the largest number of beds in the county; University Hospitals (UH), a large regional integrated system with the second largest number of beds in the county; and the MHS, the oldest and largest safety-net health system of Cuyahoga County. The MetroHealth Cancer Center (MHCC) is located close to the neighborhoods of Cleveland with dense populations of Hispanic and Black residents. Along with the MHCC, Case Comprehensive Cancer Center, an NCI-designated cancer center based at Case Western Reserve University with its partnership organizations, UH Seidman Cancer Center and Cleveland Clinic Taussig Cancer Center, draws more diverse patient groups of over 4 million people in the northern region of Ohio [62].

Study population
We identified 2297 female patients ever diagnosed with BC at MHS from 2007 to 2020. Of those, 150 patients had 2 different BC diagnoses and 5 patients had 3 different BC diagnoses. To include a unique BC diagnosis per patient, we applied the following rules in sequence: we selected (1) the most recent case among patients with multiple BC diagnoses with different diagnosis dates (n = 107) and (2) a more advanced case (more advanced stage at diagnosis or larger clinical tumor size) among patients with multiple BC diagnoses on the same date (n = 34). We observed 1 patient with the same diagnosis date, stage at diagnosis, and tumor size and selected her second registry record. Finally, we excluded 13 patients who had the same date of diagnosis and stage at diagnosis but missing clinical/pathological tumor size. The final study includes 2284 female patients diagnosed with BC at MHS from 2007 to 2020.

Outcome measures
Our first outcome was comparative TNBC prevalence which estimates the proportion of TNBC cases among all BC patients. Our second outcome measure was stage at diagnosis, defined by the pathologic American Joint Committee on Cancer (AJCC) group staging and operationalized by 5 different levels: Stage 0 (reference), Stage I, Stage II, Stage III, and Stage IV. The third outcome was total mortality which indicates a patient's survival status at her last contact date.

Independent variables and covariates
Patients' racial identity was defined as a binary variable (white vs. Black) as recorded at clinical encounters and specified in the tumor registry; while patients from other racial and ethnic backgrounds are present in the registry data, sample size for other subgroups was insufficient for inclusion in our comparative analyses. Patients' neighborhood socioeconomic position was operationalized by the overall area deprivation index (ADI), previously demonstrated to be associated with lung cancer incidence [43,58]. We used the sociome R package [63] to match patients' geocoded addresses at the time of diagnosis to the ADI constructed based on the ACS data and, as in other studies using the ADI, operationalized the estimated ADI at diagnosis as deciles.

3
Patients' age (≤ 50 years, 51-64 years, and ≥ 65 years), marital status (not married vs. married), health insurance (private, Medicare, Medicaid, uninsured, and other including managed care provider, Tricare, Veterans Affairs, and not specific insurance information) were defined at diagnosis and collected from the tumor registry. Patients' smoking status was extracted the EHR based on the last entry prior to the diagnosis date. ECI score was ascertained from all historical encounter-based diagnostic codes and active problem list diagnostic codes documented at the time of the breast cancer diagnosis date. The total number of ECI conditions that a patient was diagnosed with was counted and categorized into five levels: none, 1, 2, 3, and ≥ 4 conditions.

Statistical analyses
We conducted chi-square tests to compare distributions of categorical variables by race and t-tests to compare distributions of continuous variables between non-Hispanic white and non-Hispanic Black patients. We estimated logistic regression models to examine the relationship between patients' neighborhood-level characteristics at diagnosis and patients' race on comparative TNBC prevalence and on total mortality and implemented ordinal logistic regression models to estimate the role of patients' neighborhood socioeconomic position at diagnosis and their racial identity on their odds of having a more advanced stage at diagnosis. To estimate the role of patients' racial identity and neighborhood characteristics more accurately on stage at diagnosis, we also excluded patients with unknown or missing information of stage and adjusted for TNBC status. We conducted bivariate analysis between each outcome and neighborhood socioeconomic position at diagnosis and race, and we expanded the bivariate model to adjust for patients' sociodemographic factors (age, marital status, and health insurance type) and health status (smoking history and comorbidities). To maximize the power for each regression model, we adjusted for TNBC status to estimate the effect of TNBC on the outcomes rather than conducting stratified analysis by TNBC status. Confidence intervals (95%, bias corrected, accelerated) were generated using the bootstrap technique with 5,000 samples. To determine the extent of clustering and shared variance in neighborhoods (i.e., census tracts), we estimated and compared random intercept models and intraclass correlation coefficients. All analyses were conducted in Stata 17.0. Table 1 presents baseline characteristics of the study population by race and TNBC status. About 8% of the study population (n = 176) was diagnosed with TNBC, and about 10% of the study population identified as Hispanic or other racial/ethnic groups. About half of the study population was 51-64 years old at diagnosis across racial groups and regardless of TNBC status. The proportion of Black patients among TNBC cases (58%) was about 1.5 times higher than that among non-TNBC cases (40%). About 1 in 3 patients in the study population were married,; however, we observed a significantly smaller proportion of married Black patients (around 20%) than married white patients (44%) regardless of TNBC status. Medicare was the most common source of health insurance, and we observed significant differences in the type of health insurance by racial identity. A smaller proportion of Black patients were uninsured, while a larger proportion of Black patients were covered by Medicare. Among non-TNBC cases, a significantly larger proportion of Black patients were covered by Medicaid (p value < 0.001). While almost 44% of the study population was diagnosed at stage 1, we observed more stage II cases at diagnosis among TNBC patients compared to non-TNBC patients (Table 1). Almost one third of the study population (30%) had a family history of BC, and about 86% of the study population had at least one ECI condition. A significantly larger proportion of Black patients had at least four ECI conditions among non-TNBC cases (p value < 0.001), and the same pattern is observed among TNBC cases. Supplementary Table 1 describes the prevalence of each ECI condition by race and TNBC status. About 17% of the study population died during the study period, and a significantly larger proportion of TNBC patients died of BC compared to non-TNBC patients. Figure 2 illustrates the distribution of BC cases by racial identity in each decile of neighborhood socioeconomic position at diagnosis (Panel A) and among TNBC cases (Panel B). While about 10% of all BC cases diagnosed among those living in the least disadvantaged neighborhoods were for non-Hispanic Black women, we observed about 80% of all BC cases were for non-Hispanic Black women in the most disadvantaged neighborhoods. We observed the same pattern among TNBC patients despite smaller sample size. Table 2 presents unadjusted and adjusted associations between neighborhood socioeconomic deprivation and race and comparative TNBC prevalence. Neighborhood socioeconomic position at diagnosis is a significant predictor of TNBC cases: patients living in the most disadvantaged neighborhood (i.e., decile 10 of ADI at diagnosis) had about 1.8 times higher odds of TNBC than those living in the least disadvantaged neighborhoods (i.e., decile 1 of ADI at diagnosis). Compared to their white counterparts, non-Hispanic Black patients diagnosed with BC had almost twice the odds of having TNBC diagnosis. When estimating the effects of neighborhood socioeconomic position at diagnosis together with race, we observed that association of ADI at diagnosis was reduced and no longer significant, but that having a non-Hispanic Black racial identity remained significant.  Table 3 presents the odds of having a more advanced stage by decile of neighborhood socioeconomic position at diagnosis and by race, adjusting for TNBC status. Here, due to sample size constraints, we adjusted for TNBC status rather than conducting subgroup analyses. The odds of being diagnosed at a more advanced stage among patients living in the most socioeconomically disadvantaged neighborhoods were 40% higher than those for patients living in the least disadvantaged neighborhood (p = 0.007), adjusting for TNBC status (Table 3, Model 1A). This magnitude of association between neighborhood socioeconomic position at diagnosis and stage at diagnosis remained unchanged after adjusting for race (Table 3, Model 2) and other sociodemographic and clinical characteristics ( Table 3, Model 3). We conducted a sensitivity analysis by (1) adjusting for ER, PR, and HER2 status (Supplementary Table 2); and (2) excluding patients whose stage at diagnosis was 0 as well as patients with unknown or missing stage at diagnosis (Supplementary Table 3). The effect size and statistical significance of neighborhood socioeconomic position remained unchanged in the sensitivity analysis. Table 4 presents results of the tests of association between neighborhood socioeconomic position at diagnosis and race with total mortality among the study population. Accounting for TNBC status, we observed that living in a more disadvantaged neighborhood at the time of diagnosis increased the odds of death (OR 1.06 [1.02, 1.11]). Odds of death among non-Hispanic Black patients were about 30% higher than among non-Hispanic white patients (Table 4, Model 1B). When estimating the joint effects of neighborhood socioeconomic position at diagnosis and race, adjusting for TNBC (Table 4, Model 2), patients living in the most disadvantaged neighborhood at the time of diagnosis had 50% higher odds of death compared to those living in the least disadvantaged neighborhoods. After adjusting for patients' sociodemographic and clinical factors, neither neighborhood socioeconomic position at diagnosis nor racial identity were significantly associated with total mortality (Table 4, Model 3), suggesting that the significant associations might have been accounted for by the covariates. We conducted a sensitivity analysis by excluding patients whose stage at diagnosis was 0 and patients with unknown/missing stage information (Supplementary Table 4) and results were similar.

Discussion
This study estimated associations between neighborhood socioeconomic position, race and TNBC prevalence, disease progression, and mortality by using cancer registry, EHR, and U.S. census data for a relatively large cohort of patients diagnosed with BC from 2007 to 2020 in Northeast Ohio. We found that patients' racial identity was the strongest predictor for comparative TNBC prevalence, and that neighborhood deprivation remained the significant predictor for stage at diagnosis and total mortality after accounting for other individual-level characteristics. In contrast to other studies reporting significant associations between neighborhood characteristics and TNBC incidence [48,49,53], we found that ADI became insignificantly associated with comparative TNBC prevalence when patients' racial Frequency and column percent is reported otherwise indicated White and black patients indicate non-Hispanic white and non-Hispanic black patients TNBC triple negative breast cancer, SD standard deviation a Frequency and row percent by race is indicated for TNBC and Non-TNBC population groups b Other health insurance includes managed care plan, Tricare, Veterans Affairs, or other not specified private plans c Stage at diagnosis is defined by pathologic AJCC group staging d Elixhauser comorbidity index includes the following conditions: congestive heart failure, cardiac arrhythmia, valvular disease, pulmonary circulation disorders, peripheral vascular disease, hypertension, paralysis, other neurological disorders, chronic pulmonary disease, diabetes uncomplicated, diabetes complicated, hypothyroidism, renal failure, liver disease, peptic ulcer disease including bleeding, HIV, lymphoma, metastatic cancer, solid tumor without metastasis, Rheumatoid arthritis, coagulopathy, obesity, weight loss, fluid and electrolyte disorders, blood loss, deficiency anemia, alcohol abuse, drug abuse, psychoses, and depression identity was included in the model. This may be due to limited variations in characteristics of our study population, which was constructed from the only safety-net hospital system and Medicaid deemed disproportionate share hospital in Cuyahoga County [60]. Given the persistent income gaps between non-Hispanic white vs. non-Hispanic Black [64], when patient's racial identity and neighborhood-level SES were estimated together with respect to comparative TNBC prevalence, the significant effects of ADI were masked by the effects of racial identity. However, our observations that more disadvantaged neighborhoods were significantly associated with more advanced diseases and higher total mortality agree with other studies, illustrating the general observation of increased social risks and barriers to access to care in more disadvantaged neighborhoods. All in all, our findings suggest complex mechanisms encompassing patients' individual-and neighborhood-level characteristics variations in comparative TNBC prevalence, stage at diagnosis, and total mortality along the lines of neighborhood socioeconomic position and racial identity. By including measures of neighborhood characteristics in combination with clinical variables from a tumor registry  1 3 and EHR, we extended insights provided by prior epidemiology or cancer biology studies of TNBC [7,22,65,66], which have been constrained by the lack of availability of geocoded addresses, comorbidities, and other indicators in tumor registries. A common data source of many studies investigating the contributions of patients' individual-and community-level characteristics and cancer incidence and mortality are cancer registries. Our study expands upon this approach by linking cancer registry data with patients' EHR to better capture patients' health status. Patients' smoking status was assessed based on their records in the EHR, and the number of ECI conditions was also extracted from the EHR over a longer time window prior to the cancer diagnosis. In addition to patients' sociodemographic factors and health insurance information, our study more accurately estimates the effects of patients' health conditions that are known to be associated with risks of developing cancer.
Incorporating the ADI to measure neighborhood socioeconomic position alongside patient-reported racial background enabled our study to evaluate the overlapping contributions of socioeconomic position and race more comprehensively. The racial compositions and socioeconomic conditions of neighborhoods are intricately connected based upon current and historical patterns of racial discrimination. Racial segregation of neighborhoods in the U.S. is closely tied to patterns of lending, employment, and housing discrimination that were part of government policy under rulings such as the U.S. Supreme Court decision of maintaining "separate but equal" policies of racial segregation in Plessy vs. Ferguson [67]. However, because our study was not able to complement the use of patients' self-reported racial identity with genetic ancestry data, our findings neither rule out nor confirm ancestral hypotheses. Still, given rich bodies of research on social research, we underscore evidence that the lives of Black women in America comprise unique constellations of social, economic, and political treatment that have deleterious effects on health. However, findings from this study must be considered in light of limitations. First, our study population may not be generalizable to the overall population of Northeast Ohio. As a safety-net hospital system, The MHS patient populations have larger proportions of non-Hispanic non-White residents and of low-income un-/under-insured residents than the general population of Northeast Ohio. Despite the potential over-representation of the low-income non-Hispanic non-White patient population, our study findings agree with other studies using more diverse populations.
Second, a small sample size of TNBC cases did not permit comparisons between neighborhood-level characteristics and race and cancer outcomes stratified by TNBC vs. non-TNBC cases. However, all our analyses report bootstrapped effect sizes and statistical significance to address this limitation. Third, our current data do not allow us to estimate the length of time for which patients had resided at their reported address at the time of diagnosis, and the scope of our study is limited to only cross-sectionally estimating the relationship between neighborhood characteristics and TNBC incidence, stage at diagnosis, and mortality. Fourth, EHR data used in this study do not provide patients' complete history of healthcare services use. However, the MHS Epic system is linked with Cleveland Clinic's EHR, and we verified the registry data with the information in the linked Epic system. Fifth, our study did not examine patients' reproductive history and differences in treatment patterns as well as treatment decision-making. These variables can have further influences on survival and be associated with socioeconomic and cultural preferences. Despite these limitations, our study effectively suggests the complex pathways of BC disparities likely operate differently across BC subtypes. Biological characteristics, health behaviors, and extra-individual social circumstances combine to influence BC outcomes, but the sequence and magnitude of these interrelated factors across the cancer care continuum and BC subtypes can be variable. Also, a clinical implication of this study includes more tailored efforts to increase BC prevention in deprived neighborhoods. More health education and screening for BC in these neighborhoods may increase the likelihood of early detection, reducing the burden of BC among non-Hispanic Black women living in disadvantaged neighborhoods. Future research can include other patient populations in Northeast Ohio to increase the generalizability of our study findings. In addition, our study recommends further investigations of the role of patients' and neighborhood socioeconomic position on survival by incorporating interactions among aging trajectory, disease progression, and neighborhood-level factors.