Effect of Neighborhood Deprivation Index on Breast Cancer Survival in the United States

Purpose To analyze the association between the Neighborhood Deprivation Index (NDI) and clinical outcomes of early-stage breast cancer (BC). Methods Surveillance, Epidemiology and End Results (SEER) database is queried to evaluate overall survival (OS) and disease-specific survival (DSS) of early- stage BC patients diagnosed between 2010–2016. Cox multivariate regression was performed to measure the association between NDI (Quintiles corresponding to most deprivation (Q1), above average deprivation (Q2), average deprivation (Q3), below average deprivation (Q4), least deprivation (Q5)) and OS/DSS. Results Of the 88,572 early-stage BC patients, 27.4% (n = 24,307) were in the Q1 quintile, 26.5% (n = 23,447) were in the Q3 quintile, 17% (n = 15,035) were in the Q2 quintile, 13.5% (n = 11,945) were in the Q4 quintile, and 15.6% (n = 13,838) were in the Q5 quintile. There was a predominance of racial minorities in the Q1 and Q2 quintiles with Black women being 13–15% and Hispanic women being 15% compared to only 8% Black women and 6% Hispanic women in the Q5 quintile (p < 0.001). In multivariate analysis, in the overall cohort, those who live in Q2 and Q1 quintile have inferior OS and DSS compared to those who live in Q5 quintile (OS:- Q2: Hazard Ratio (HR) 1.28, Q1: HR 1.2; DSS:- Q2: HR 1.33, Q1: HR 1.25, all p < 0.001). Conclusion Early-stage BC patients from areas with worse NDI have poor OS and DSS. Investments to improve the socioeconomic status of areas with high deprivation may help to reduce healthcare disparities and improve breast cancer outcomes.

Introduction percent of housing units that are owner occupied; percent of households without a telephone; percent of households without complete plumbing facilities) [16]. NDI values range from -3.6 to +2.8 and higher values indicate more neighborhood deprivation which implies lower socioeconomic status. We used the NDI quintiles weighted by the tract population for the analysis. The rst NDI quintile corresponds to most deprivation (Q1), second quintile (above average deprivation-Q2), third quintile (average deprivation-Q3), fourth quintile (below average deprivation (Q4)) and fth quintile corresponds to least deprivation (Q5) [17].

B. Patient selection
We queried the Surveillance, Epidemiology and End Results (SEER) registry November 2021 submission database which covers approximately 48% of the US population for our study. We included early-stage BC pts (clinical stage group I, II, III), aged >=18 years, who are diagnosed from 2010-2016, and studied the overall survival (OS) and disease-speci c survival (DSS) of BC in association with NDI. As data on HER2-neu status was accurately captured in SEER since 2010, this was chosen as the initial year of diagnosis for inclusion and selected patients diagnosed until 2016 to give adequate follow-up of 5 years. We excluded patients with unknown or missing data for each variable studied, or clinical/pathological evidence of distant metastases at the time of initial diagnosis. Institutional review board review was exempted as the data were deidenti ed and from publicly available databases upon request.
Statistical Analysis.
The demographical and clinical characteristics of patients by NDI were tabulated by summary statistics. The mean, median, standard deviation, and range were used for continuous variables and the Kruskal-Wallis test was used for comparisons. For the categorical variables, frequencies and relative frequencies were compared using the chi-square test. The median, 3-year, 5-year OS and DSS were summarized by NDI using standard Kaplan-Meier methods.
Cox multivariate regression modeling was performed to test the association between NDI and OS, DSS, with adjustment for age, race, stage, grade, insurance status, surgery, radiation, and chemotherapy (CT). Subset analysis was done based on the BC subtypes (Estrogen receptor and/or progesterone receptor positive HER2-neu negative (HR+), HER2-neu-positive (HER2 +) and triple-negative breast cancer (TNBC). All statistics were performed using SAS software version 9.4 (SAS Institute Inc.) and signi cance testing was 2-sided at p<0.05. Data were analyzed from June 1, 2022 through July 15, 2022.

Patient demographics
The baseline characteristics of the overall cohort are shown in Table 1. Of the 88,572 early-stage BC patients,27.4 % (n= 24,307) were in the most deprivation (Q1) quintile, 26.5% (n= 23,447) were in the average deprivation (Q3) quintile, 17% (n= 15,035) were in the above average deprivation (Q2) quintile, 13.5% (n= 11,945) were in the below average deprivation (Q4) quintile, and 15.6% (n= 13,838) were in the least deprivation (Q5) quintile. The median age of patients in the Q5 quintile was 59 yrs and Q1 quintile was 61 yrs (p<0.001). There was a predominance of racial minorities in the Q1 and Q2 quintile with Black women being 13-15% and Hispanic women being 15% compared to only 8% Black women and 6% Hispanic women in the Q5 quintile (p<0.001). There was a higher percentage of uninsured patients in the Q1 quintile compared to the Q5 quintile (2.2% vs 1.7%, p<0.001). There were more rural areas in Q1 quintile compared to Q5 quintile (25.9% vs only 0.7%, p<0.001). There were more patients with stage III and grade III disease in Q1 quintile compared to Q5 quintile (Stage III: 28.7% vs 14.2%, Grade III: 34% vs 31.9%, p<0.001), and therefore, a greater percentage of patients received CT in Q1 quintile compared to Q5 quintile (44.6% vs 42.1%, p<0.001). However, fewer patients underwent surgery and radiation in the Q1 compared to the Q5 quintile, with 96.1% and 49.7% of patients undergoing surgery and radiation in Q1 quintile compared to 97.1 % and 56.5% in the Q5 quintile (p<0.001 for both).
There was a higher percentage of aggressive cancers such as TNBC and HER2+ BC in Q1 quintile compared to Q5 quintile (14.5%, 17.7% vs 11.7%, 16.5% respectively, p<0.001). The baseline characteristics were strati ed by the subtype of breast cancer as shown in Table 2 -4. It was observed that the patterns are similar in all the subtypes as observed in the overall cohort except that the patients who received chemotherapy for the early-stage BC were higher in the Q5 when compared to the Q1 in both TNBC and HER2+ BCs.

Kaplan-Meier Survival Estimates
On univariate analysis, after a median follow-up of 44 months, the 5-year OS rate of the overall cohort was 87%. The 5-year OS of the early-stage BC patients who live in the Q1 and Q2 quintile was low when compared to those who live in the Q5 quintile (85%, 84% vs 89%, p<0.001). The DSS of the overall cohort also followed a similar pattern (DSS of Q1, Q2 vsQ5: 92%, 91% vs 94%, p<0.001) (Table 5, Figure 1).
In subset analysis strati ed by the BC subtypes, the 5-year OS and DSS were lower in the Q1 and Q2 quintiles compared to the Q5 quintile in all the subtypes of BC (HR+, HER2 + and TNBC). However, the 5-year DSS rate was not signi cantly different in the HR+ subtype (Q1, Q2 vs Q5: 95%, 95% vs 96%, p<0.001).

Multivariate Survival Analysis
In multivariate analysis after adjusting for socio-demographic, clinical, and treatment variables, in the overall cohort, those who live in Q2 quintile and Q1 quintile have inferior OS and DSS when compared to those who live in Q5 quintile (OS in Q2: Hazard Ratio (HR) 1.28; OS in Q1: HR 1.2; DSS in Q2: HR 1.33; DSS in Q1: HR 1.25, all p<0.001). In the subset analysis, similar results for OS and DSS by NDI were observed in hormone receptor positive HER2 negative (HR+) and HER2+ subtypes, but not in TNBC ( Figure 3).
Discussion study, we found that the deprivation index of the neighborhoods was in signi cant association with BC survival. Our analysis showed that the OS and DSS of patients with early-stage BC were lower for those who live in socioeconomically deprived neighborhoods compared to those who live in a uent neighborhoods. The survival differences were observed among all subtypes of BC. The differences in survival persisted even after adjusting for demographic, clinical, and treatment factors that could affect breast cancer survival. On multivariate analysis, the mortality difference among the patients living in different socioeconomic areas was statistically signi cant within the overall cohort, HR+ and HER2+ BC subtypes, but not within the TNBC subtype. This could be explained by the aggressive nature of TNBC. As TNBC is very aggressive and has high relapse rate [18], the survival of patients with TNBC could be poor regardless of their socioeconomic status.
Understanding the impact of neighborhood deprivation on BC survival will facilitate the development and implementation of policies and prioritize investments in communities with high deprivation scores. This could improve the socioeconomic conditions which could eventually improve clinical outcomes. [19] Several factors in the neighborhood in uence the health of an individual directly, as well as indirectly: poverty, access to the health care system, transportation system, housing quality, unemployment, environmental pollution including air and water pollution, neighborhood hygiene, waste management system, crime rates, racial composition, educational system, tobacco availability and marketing, access to healthy food [20][21][22][23]. These along with the factors that affect the individual such as marital status, family/social support, co-morbidities, mental health, nutritional status, healthy lifestyle, insurance status, and educational status play an inevitable role in the survival outcomes of malignancies. Studies have shown that prolonged and cumulative exposure to the above-mentioned deprivation-associated stressors can induce chronic in ammation which is one of the etiologies behind cancer development [24,25]. Therefore, a proper understanding of the deprivation factors of an individual and their neighborhood is essential to plan interventions to reduce the burden of cancer mortality.
Our study adds to the existing literature in multiple ways. This study is the rst to our knowledge to use a national database to examine the association between neighborhood deprivation with the clinical outcomes of early-stage BC; most prior studies used regional databases. Prior studies showed racial disparities in BC-related outcomes in the US and minoritized groups tend to have higher BC-related mortality [22]. In a study by Luningham et. al. which examined the association between racial disparities and SES in BC survival between Black and White women across Georgia, it was found that Black women with BC had higher mortality than White women, but this disparity was not explained by the socioeconomic deprivation of their residential area. White patients living in socioeconomically a uent areas had lower rates of BC mortality compared to those who reside in deprived neighborhoods [26]. This study nding critically shows the important role of area of residence on clinical outcomes, and thus emphasizing that socioeconomic factors of the neighborhood play a vital role in determining clinical outcomes. There were several studies conducted to understand the reason behind the observed racial disparities. One of them was attributed to poor neighborhoods. Black and Hispanic women generally live in poor neighborhoods and Black patients were found to live in neighborhoods with high poverty rates and this difference persists even after adjusting for their income status [27]. In our study, we found that Black and Hispanic women with BC were more commonly residing in the deprived neighborhoods compared to socioeconomically a uent neighborhoods; however, the disparities in BC-related mortality of the patients from these socioeconomically different neighborhoods persisted even after accounting for the racial disparities.
Similarly, uninsured patients, rural areas were found more commonly in the deprived neighborhoods. Advanced BCs (higher stage and grade) and aggressive BC such as HER2 + and TNBC were predominantly found in the deprived neighborhoods compared to the a uent neighborhoods. In our study, BC patients from a uent neighborhoods received more surgical and radiation treatments which can be explained by the higher percentages of urban areas in these regions with better facilities for treatments, and better referral systems. Our study showed that in the overall BC cohort, patients who received chemotherapy were slightly higher in the deprived neighborhoods than in the a uent neighborhood. One possible explanation for this is that the advanced diseases that require chemotherapy were more prevalent in the deprived neighborhood regions. Nevertheless, the disparities in BC-related mortality remain unaffected when adjusted for the demographic, clinical, pathological, and treatment-related factors such as age, race, stage, grade, insurance, surgery, radiation, and chemotherapy. This suggests that additional factors related to neighborhood SES that are not captured by the NDI play an important role in BC-related outcomes. The access to genetic and somatic testing which are important for deciding the appropriate treatments in BC might be limited to patients from poor neighborhood which could have impacted their survival.
Poverty, unhealthy food habits, decreased access to healthy food, environmental pollution, increased advertising of tobacco in poor neighborhood leads to increased incidence of various cancers in patients from these neighborhoods [20,21,23,28,29]. Furthermore, the transportation barriers, decreased access to better comprehensive cancer centers with standard of care treatments or novel clinical trials, poor nutritional and educational status of patients, nancial toxicity associated with cancer treatment leads to increased cancer-related mortality in socioeconomically poor neighborhoods [10,14,30,31]. In addition to this, poor environmental hygiene and pollution can add to the increased mortality by causing infections in cancer patients who are already immunocompromised due to cancer and associated treatments.
[32] In a patient-reported outcomes study in advanced cancers, it was found that patients from socioeconomically deprived areas have a higher level of anxiety.
[33] Factors such as anxiety, depression, and poor social support which are subjective measures of poor mental health are not accounted for in any of the tools to measure the neighborhood/individual SES and have been shown to be associated with cancer-related mortality. [13,33,34] Disparities in BC survival related to neighborhood SES re ect the systematic barriers in policies related to health care, education, employment, insurance, environment, and judicial system. Our study ndings may support restructuring the policies, to implement new policies and investments in socioeconomically deprived neighborhoods which would help to decrease inequalities in opportunities, improve healthcare facilities, and increase access to timely cancer treatments.
Our study has many strengths and certain limitations. We used large real-world data to assess the impact of neighborhood deprivation on clinical outcomes of BC patients. These data capture more than 50% of the US population, and therefore, the results are generalizable. We adjusted for multiple factors that are known to in uence survival, including racial distribution, demographic factors, clinicopathological factors of the disease [35]. Although NDI is a comprehensive tool to assess socioeconomic disadvantage, it may not capture all the factors associated with neighborhood SES. We could not assess the in uence of several neighborhood factors that may contribute to the mortality of BC such as access to transportation, environmental: air, water pollution, poverty level, accessibility to healthy food, and crime rate of the neighborhood. As we do not have one comprehensive tool to assess the socioeconomic status of neighborhoods and individuals together, future studies warranting the combination of multiple indices such as area deprivation index, Yost index might be bene cial. As it is a retrospective national database study, several individual factors that can affect the mortality rates of BC such as comorbidities of patients, social support, details of factors such as anxiety, and depression that can affect the mental and physical condition of the patients were not collected.
Incompleteness of individual-level data collected on cancer risk and treatment, and incomplete values for several variables collected from multiple registries were other limitations. Further tumor recurrence data, speci c details on the type, dose, and duration of chemotherapy, radiation, oral chemotherapy, targeted agents were not available in the SEER database and these factors could have been associated with the mortality of BC.

Conclusion
The ndings from our study suggest that neighborhood deprivation is signi cantly and independently associated with worse clinical outcomes among patients with BC in the US. Early-stage BC patients from areas with worse NDI have poor OS and DSS, after accounting for demographic, clinicopathological, and treatment-related factors. Identi cation of these poor-resource neighborhoods is critical to guide investments in these neighborhoods and implement policies focusing on improving the SES of these areas with high deprivation to reduce healthcare disparities and improve breast cancer outcomes. Future studies are warranted to understand the factors affecting the neighborhood socioeconomic status other than what is mentioned in our study and to assess their relationship with BC-related survival. The data from these studies might be extrapolated to other cancers which would help us to improve the quality of life of patients and cancer-related mortalities.

Declarations Funding
Research reported in this publication was supported by the National Center for Advancing Translational Sciences of the National Institutes of Health under award numbers KL2TR001413 and UL1TR001412. The content is solely the responsibility of the authors and does not necessarily represent the o cial views of the NIH.

Con ict of Interest
The authors declare that they have no con ict of interest.

Author Contributions
Arya Mariam Roy and Shipra Gandhi contributed to the study conception and design. Data acquisition, interpretation of data and statistical analysis was performed by Arya Mariam Roy, Anthony George, Kristopher Attwood and Shipra Gandhi. Initial draft of the manuscript was written by Arya Mariam Roy. Initial draft was reviewed and edited by Archit Patel and Sabah Alaklabi. All authors commented on revision of manuscript for important intellectual content. The project was conducted under the supervision of Shipra Gandhi. All authors have read and approved nal manuscript.

Data Availability
All data utilized in this article is available in public datasets such as SEER and NCI Neighborhood deprivation index. Data analyzed during this study are included in this published article and its Appendix.

Ethics approval:
The study does not require Institutional Review Board (IRB) approval as all data used for analysis are deidenti ed and from publicly available databases.