A comparison of neighborhood socioeconomic deprivation measures and the association with survival among Black and White women with endometrial cancer

Abstract

Background:

Black women with endometrial cancer (EC) have twice the mortality compared to White. Survival disparities remain after accounting for individual-level socioeconomic and cancer-related factors. We investigated associations between area-based deprivation and survival and explored whether area-based deprivation attenuates the association between race and survival, among a cohort of Black and White women.

Methods:

Data from ECs diagnosed between 2013–2022 were collected from a comprehensive cancer registry covering Metropolitan Detroit. Addresses at diagnosis were linked to Area Deprivation (ADI) and Social Vulnerability (SVI) indices. Adjusted Fine & Gray and Cox proportional hazard models were run investigating associations between area-based deprivation measures and survival; analyses were conducted estimating the proportion of the association between race and survival that was attenuated by area-based measures.

Results:

Higher deprivation was associated with poorer survival, adjusted for race, insurance status, and tumor characteristics. Compared to the least disadvantaged quartile, the quartile with the highest disadvantage using ADI and SVI had 1.18 (95% CI: 0.99, 1.43) and 1.40 (1.14, 1.71) times the hazard of EC-specific mortality, respectively. ADI and SVI attenuated 18% (3–38%) and 27% (10–48%) of associations between race and mortality overall, and 24% (95% CI: 3–61%) and 40% (95% CI: 16–78%) among those with high-grade histology.

Conclusions:

This study demonstrates a clear association between neighborhood-level disadvantage and survival among women with EC living in Metropolitan Detroit. Neighborhood disadvantage attenuates the relationship between race and survival, particularly among those with high-grade histology.

Impact:

These findings serve as motivation to understand how neighborhood impacts cancer outcomes.

Introduction

Endometrial cancer (EC) is the most common gynecologic cancer in the United States (US), with nearly 68,000 incident cases and 13,250 deaths each year(1). In addition, EC is one of the few cancers in the US with increasing mortality and decreasing survival over the past 40 years(1). Over 80% of EC cases are non-aggressive histologic subtypes with a good prognosis, but recently the incidence of aggressive histologic subtypes (e.g., serous carcinomas, clear-cell carcinomas, carcinosarcomas), which have poor prognosis and account for over 40% of EC deaths(2), has been rising(3). EC also has one of the largest racial disparities for survival among all cancers: EC mortality rates are twice as high in non-Hispanic Black women (NHBW) compared to non-Hispanic White women (NHWW)(1). This inequity is partially related to elevated rates of aggressive EC in the NHB population(4), but mortality rates are higher among NHBW across all histologic subtypes and stages at diagnosis. Individual-level social characteristics, such as income and insurance status, are also known to contribute to racial disparities in EC survival(5). However, even after adjustment for known biological factors and social factors, one study found that 15–20% of the excess relative risk of death from EC in NHBW remained unexplained(6). This warrants a need for additional studies focusing on the impact of social determinants of health on EC disparities.

Along with individual-level biological and social factors, neighborhood-level factors are known to play a part in cancer-related health(7–9) (e.g., screening behaviors(10), incidence rates(11,12), survival(13,14)). For example, multiple investigations have found that living in a persistent poverty neighborhood, defined by the US Census Bureau as at least 20% of the population living in poverty for at least 30 years(15), is associated with decreased cancer survival across a variety of cancers including head and neck, breast, lung, pancreatic, hepatocellular, colon, bladder, prostate, ovarian, and endometrial(16–25).

There are various singular measures that can be used to quantify aspects of area deprivation in the context of cancer health disparities, such as persistent poverty(26,27), air quality(27), and racial/ethnic residential segregation(28). Composite measures of deprivation, which capture multiple dimensions of the Social Determinants of Health(29), are often preferred as they provide a more comprehensive picture of neighborhood socioeconomic disadvantage and thus are often more strongly related to outcomes(30). Recently, several studies investigating the association between neighborhood-level factors and the cancer continuum have used one of two composite area-based measures, the Area Deprivation Index (ADI)(31–34) and the Social Vulnerability Index (SVI)(35–37), to measure area-level socioeconomic disadvantage. Both use a principal component analysis (PCA) to combine several sociodemographic data elements from the American Community Survey (ACS) into an index that, in turn, ranks the social disadvantage of a geographic area. Critically, although studies on the association between neighborhood-level indices and the cancer continuum have treated the ADI and SVI as transposable measures, these national indices were developed for different purposes and are not interchangeable(38). The ADI, designed to assess neighborhood resource deprivation, is evaluated at the census block group-level. In contrast, the SVI, created to spatially identify vulnerable communities most in need of health-related resources, is often estimated at the county- or census tract-level (a census tract is composed of several census block groups, which are composed of several census blocks). Census tracts have population sizes ranging from 1,200 to 8,000 people and are made up of at least one census block group (usually 2–5 groups), each of which have a population ranging from 600 to 3,000 people(City of Detroit Census Data Map: https://detroitmi.gov/webapp/census-data-map)(39,40). Because of their different purposes, these measures include different sociodemographic data elements from the ACS (e.g., the SVI contains minority status and language factors, while the ADI contains housing price factors). As such, even when the ADI is aggregated up to the same census tract-level as the SVI, the agreement on which areas are “vulnerable” or “deprived” is modest (R = 0.51)(41). This is significant because the use of varying composite variables can yield different results; for example, one study on organ transplant patients observed markedly different distributions of patients across deprivation quartiles when using different composite measures(42).

Despite these essential differences in composition and use, as well as the relative novelty of using any composite area-based socioeconomic disadvantage measure in a cancer survival study, to our knowledge only one prior study has compared these indices in their association with cancer outcomes(43). It is important to understand if the way in which deprivation is operationalized affects the association between area deprivation and cancer outcomes so that appropriate measures can be selected to better understand and address health inequities. In this study, we investigated the association of ADI and SVI with all-cause and EC-specific survival, using EC cases among NHBW and NHWW identified through the Metropolitan Detroit Cancer Surveillance System (MDCSS). Further, we investigated whether ADI and SVI attenuated the association between race/ethnicity (NHB vs NHW) and survival, which could in part explain known racial survival disparities.

Materials and Methods

We conducted a retrospective analysis investigating the association between two area-based deprivation indices (ADI and SVI) with survival among NHWW and NHBW with EC. We examined the hazard ratios for EC survival according to quartiles of the deprivation indices for all cases, as well as stratified by high- or low-grade histology group. Additionally, we calculated the proportion of the association between race/ethnicity and survival that was attenuated by the area-based measures. Cases were derived from 10 years of data from the MDCSS (2013–2022).

Study Population and MDCSS-derived variables

The MDCSS is a population-based cancer registry that covers the tri-county area of metropolitan Detroit, including Wayne, Oakland, and Macomb counties. All cases were ascertained from the MDCSS with the inclusion criteria of: (1) diagnoses between 2013–2022; (2) tri-county resident at diagnosis; (3) invasive disease; (4) EC case as defined by the cancer site recodes of “corpus uteri” and “uterus, NOS”; and (5) NHB or NHW. No exclusions were made for cases that were identified by autopsy or death certificate only. Sarcomas of the corpus uterus and gestational trophoblastic tumors were excluded. Races and ethnicities other than NHB and NHW were excluded due to small counts, which corresponds to the overall population of the Metro Detroit area.

The MDCSS, a gold star North American Association of Central Cancer Registries (NAACCR) registry and a research support Surveillance, Epidemiology, and End Results (SEER) registry, complies with the requirement to report vital status and date of last contact within 22 months of the date of annual data submission for a minimum of 90% of all registered cancer patients. Vital status information is collected through multiple methods, including linkage to the State of Michigan’s death certificate records, updates from reporting medical facilities, and indirect verification through linkages with administrative databases such the Centers for Medicare and Medicaid Services (CMS), the Social Security Administration (SSA), the National Death Index (NDI), and State of Michigan’s voter registration and department of motor vehicle’s office. The dataset for this project was created in June of 2024 and follow-up was current for a minimum of 90% of the patients within 22 months of that date.

The following variables were obtained from MDCSS records. Histology was grouped according to International Classification of Diseases for Oncology, Third Edition (ICD-O-3) codes, as described previously in the literature(44). Due to the high proportion of unknown grade in the MDCSS database, electronic pathology reports and abstractor notes were reviewed to record grade information. Grade was categorized as low grade or high grade. Low grade was based on either standard grade 1 “well-differentiated” or 2 “moderately differentiated,” International Federation of Gynecology and Obstetrics (FIGO) grade 1 or 2, or a clinical note of “low grade.” High grade was categorized based on either standard grade 3 “poorly differentiated” or 4 “undifferentiated,” FIGO grade 3, clinical note of “high grade,” or histologic subtype of serous, clear cell, or carcinosarcoma. Due to the heterogeneity of tumors in the historical Type 1/Type 2 grouping of EC, we instead grouped tumors by histology and grade, consistent with recent literature(44,45). Tumors were defined as either low-grade histology (low-grade endometrioid, low-grade mixed, mucinous, other) or high-grade histology (clear cell, high-grade endometrioid, high-grade mixed, carcinosarcoma serous)(44). FIGO staging was collapsed into four categories (Stage I, II, III, and IV). Surgery was categorized as yes or no. Insurance status was categorized as private, Medicare, Medicaid, military, no insurance, or unknown.

In the MDCSS, the primary cause of death is coded from death certificates in valid ICD-10 codes. For EC-specific mortality analyses, an event was defined as death due to EC by the following ICD-10 (RRID:SCR_010349) codes: C541, C549, C55, and C559. Survival time was calculated from the date of disease diagnosis. Three cases had missing months of disease diagnosis that were coded as June.

Area-based deprivation measures

This project utilized and compared the ADI and SVI, two measures that have been previously used to study the association between neighborhood-level deprivation and outcomes along the cancer continuum. ADI is calculated using a PCA of 17 ACS 2016–2020 five-year estimates (Supplementary Table S1) covering four social domains (income/poverty, education, employment, and housing quality) at the census block group-level, and was downloaded and linked to each 2020 census block group within the MDCSS catchment area that had at least one EC case (n=2,631 census block groups)(Area Deprivation Index v4.0.1: https://www.neighborhoodatlas.medicine.wisc.edu). SVI is calculated using a PCA of 16 ACS 2016–2020 five-year estimates (Supplementary Table S1) covering four social domains (socioeconomic status, household composition, racial and ethnic minority status, housing type and transportation) at the census tract-level, and was similarly downloaded and linked to each 2020 census tract within the catchment area with at least one EC case (n=1,131 census tracts)(Social Vulnerability Index 2020 Database: https://www.atsdr.cdc.gov/placeandhealth/svi/data_documentation_download.html). The 2020 census tracts likely best represent the underlying populations for the ACS 2016–2020 five-year estimates used in both the ADI and SVI. The ADI values range from 0–10 with 1-unit increments, while the SVI values range continuously from 0–1. Using the state-specific version of each measure, census block group-level ADI and census tract-level SVI were linked to each EC case based on their geocoded address at cancer diagnosis, available through MDCSS. For statistical analyses, we defined both socioeconomic disadvantage measures as quartile variables, where Q1 represented the least disadvantage and Q4 the most, based on the distribution of cases in our analytic sample.

Statistical analyses

We reported frequencies and proportions of categorical population characteristics and mean and standard deviation for continuous elements overall and by race/ethnicity (i.e., NHB or NHW), including sociodemographic (age, insurance, ADI, SVI) and disease-related measures (vital status, follow-up time, histology, histology-grade category, grade, stage, and surgery). Survival curves based on Kaplan-Meier technique were used to plot all-cause and EC-specific survival over the follow-up period stratified by ADI or SVI quartile. Follow-up time was calculated from date of diagnosis to date of death, and time to censor was calculated from date of diagnosis to date of last contact. Multivariable Cox proportional hazard models were constructed to compare hazard of all-cause mortality across increasing levels (i.e., quartiles) of neighborhood socioeconomic disadvantage; Fine and Gray cumulative incidence models(46) were constructed to compare hazard of EC-specific mortality, accounting for competing risks. We verified proportional hazard assumptions were met using graphical diagnostics and statistical tests based on Kaplan-Meier curves and Schoenfeld residuals. Covariates that did not meet proportional hazard assumptions were included as a stratification variable in the models. Models accounted for race/ethnicity, insurance status (only compared among those with private or public [Medicare/Medicaid] due to small counts in other categories), and histology-grade category (for the EC-specific models). For the all-cause mortality models, histology-grade category did not meet proportional hazards assumptions and was included as a stratification variable. A cluster term was included to account for interdependence between people living in the same census block group or census tract. While stage at diagnosis is a strong predictor for survival, we did not adjust due to its collinearity with the histology-grade variable. Observations with missing information in exposure or outcome variables were removed from analyses. A P value of < 0.05 was considered statistically significant.

To evaluate estimated attenuation due to area-based measures, we first investigated the independent association between race/ethnicity and survival, using Kaplan-Meier and adjusted Cox proportional hazard and Fine and Gray models. We then looked at the association between race/ethnicity and area-based measures, using chi-squared tests and logistic regression. Next, we confirmed that, after adjusting for each area-based measure, the association between race/ethnicity and survival decreased compared to the model without the area-based measure. Finally, we added the area-based measure into an adjusted Cox proportion hazard models and used the mediate package in R to calculate the proportion of the association between race/ethnicity and survival that was attenuated by the area-based measure. This method has been described in the literature(47) and used in other similar studies(13). For these analyses, we removed those with zero follow-up time (i.e., diagnosis date and last date of contact were the same) as well those with missing cause of death, histology-grade category, and/or area-based measure. All analyses were conducted in R Version 4.4.1 (https://cran.r-project.org/, RRID:SCR_001905).

Ethical approval

This study was deemed exempt from Institutional Review Board approval as all data are deidentified and coded for public use. The researchers did not have access to any identifying information. Geocoding and linkage to area-based measures was performed by the Epidemiology Research Core at Karmanos Cancer Institute to maintain confidentiality. The research team received a dataset where census tract and block group IDs were replaced with random IDs.

Data Availability

The data generated in this study are available upon request from the corresponding author.

Results

From 2013–2022, there were 7,788 diagnoses of EC in the Wayne, Macomb, and Oakland counties of Metropolitan Detroit. Of these, 7,430 were either NHWW or NHBW. We excluded 358 women who were other races or ethnicities due to small counts (147 Hispanic, 161 non-Hispanic Asian, and 50 other race). Among the 7,430 cases included in the analytic dataset, 39 (0.05%) were identified through autopsy or death certificate only.

Characteristics of the study population are described in Table 1. Of the 7,430 cases included in this analysis, 5,673 (76%) were NHW and 1,757 (24%) were NHB. The mean age at diagnosis (64 years) did not differ by race/ethnicity. All other variables were statistically significantly different by race/ethnicity: compared to NHWW, a higher proportion of NHBW were deceased (45% vs 26%), had non-endometrioid histology (49% vs 22%), and had later stage cancer at diagnosis (64% vs 45% stage 2+ at diagnosis among those with known stage). Additionally, more NHBW had public insurance (41% vs 26%) and fewer received surgery (78% vs 89%) than NHWW.

Table 1.

Characteristics of endometrial cancer cases in metropolitan Detroit, 2013–2022

	Overall		NHW		NHB		P
	N	%	N	%	N	%
Total	7430		5673		1757
Age at diagnosis, mean, SD	63.9	11.3	63.9	11.3	63.9	11.2	0.79A
Vital Status							<0.001B
Alive	5165	69.5	4190	73.9	975	55.5
Deceased	2265	30.5	1483	26.1	782	44.5
Follow-up time (years), mean, SD							<0.001A
Alive	5.81	2.9	5.88	2.9	5.51	3.0
Deceased	2.41	2.4	2.6	2.5	2.06	2.0
Histology							<0.001B
Clear cell	163	2.2	105	1.9	58	3.3
Endometrioid	5298	71.3	4405	77.6	893	50.8
Mixed	382	5.1	275	4.8	107	6.1
Carcinosarcoma	501	6.7	279	4.9	222	12.6
Mucinous	31	0.4	25	0.4	<11D	<1.0
Serous	807	10.9	420	7.4	387	22.0
Other	248	3.3	164	2.9	~80D	~5.0
Histology-grade category							<0.001B
Low	5042	67.9	4217	74.3	825	47.0
High	2259	30.4	1365	24.1	894	50.9
Missing	129	1.7	91	1.6	38	2.2
Grade							<0.001B
Low	4797	64.6	4049	71.4	748	42.6
High	2340	31.5	1416	25.0	924	52.6
Missing	293	3.9	208	3.7	85	4.8
FIGO stage							<0.001B
1	3733	50.2	3099	54.6	634	36.1
2	386	5.2	292	5.1	94	5.4
3	776	10.4	567	10.0	209	11.9
4	356	4.8	211	3.7	145	8.3
Missing	2179	29.3	1504	26.5	675	38.4
Insurance							<0.001C
Private	2977	40.1	2449	43.2	528	30.1
Medicare	3484	46.9	2579	45.5	905	51.5
Medicaid	489	6.6	325	5.7	164	9.3
Military	21	0.3	13	0.2	<11	<1.0
Uninsured	55	0.7	31	0.5	24	1.4
Unknown	404	5.4	276	4.9	~125	<10.0
Surgery							<0.001B
Yes	6419	86.4	5042	88.9	1377	78.4
No	969	13.0	596	10.5	373	21.2

^AT-test used to compare continuous variables

^BChi-squared test used to compare categorical variables

^CFisher’s exact test used to compare Insurance, due to small cell count

^DCells with “<” or “~” are included to comply with confidentiality guidelines for registry data, which require suppression of numbers less than eleven to protect patient confidentiality.

Irrespective of composite measure, those living in neighborhoods with more disadvantage had lower all-cause and EC-specific survival (Figure 1). The adjusted Fine and Gray hazard models (Table 2) demonstrated a statistically significant trend with increasing disadvantage quartiles of SVI corresponding to increasing hazard ratios. This pattern was less apparent with the ADI measure, but Quartiles 2, 3, and 4 all had higher mortality hazard compared to Quartile 1. After adjusting for race/ethnicity, insurance, and histology-grade category (i.e., low-grade or high-grade histology), women living in neighborhoods in ADI Quartiles 2, 3, and 4 had 1.30 (95% CI: 1.08, 1.56), 1.27 (95% CI: 1.06, 1.53), and 1.18 (95% CI: 0.99, 1.43) the hazard of EC-specific death compared to those living in neighborhoods in ADI Quartile 1, respectively. Similarly, those living in neighborhoods in SVI Quartiles 2, 3, and 4 had 1.32 (95% CI: 1.10, 1.60), 1.36 (95% CI: 1.12, 1.66), and 1.40 (95% CI: 1.14, 1.71) the hazard of EC-specific death compared to those living in neighborhoods in SVI Quartile 1. Results were similar for models where the outcome was all-cause mortality (Supplementary Table S2).

Figure 1. All-cause and endometrial-cancer specific survival by quartile of area-based deprivation measure.

The figures in the left column show the all-cause (top) and endometrial-cancer specific (bottom) survival for women in the tri-county area by quartile of state-specific ADI measure. The figures in the right column show the all-cause (top) and endometrial-cancer specific (bottom) survival for women in the tri-county area by quartile of state-specific SVI measure. For both measures, Q1 (yellow) encompasses women living in neighborhoods with the least amount of deprivation; Q4 encompasses women living in neighborhoods with the most amount of deprivation.

Table 2.

Sub-distribution hazard ratios^A for endometrial-cancer specific mortality by neighborhood deprivation quartile for women with endometrial cancer, adjusted for race, insurance status, and histology-grade category

	ADIB	N = 7,105		SVIB	N = 7,124
	SHR	95% CI	P	SHR	95% CI	P
Deprivation quartile			<0.001C			<0.001C
Q1 (Least disadvantage)	Reference			Reference
Q2	1.30	(1.08, 1.56)	0.005	1.32	(1.10, 1.60)	<0.001
Q3	1.27	(1.06, 1.53)	0.010	1.36	(1.12, 1.66)	<0.001
Q4 (Most disadvantage)	1.18	(0.99, 1.43)	0.082	1.40	(1.14, 1.71)	<0.001
Race
White	Reference			Reference
Black	1.38	(1.19, 1.61)	<0.001	1.30	(1.11, 1.51)	<0.001
Insurance
Private	Reference			Reference
Public	1.67	(1.45, 1.91)	<0.001	1.63	(1.42, 1.87)	<0.001
Histology-grade category
Low	Reference			Reference
High	4.62	(4.02, 5.31)	<0.001	4.65	(4.04, 5.34)	<0.001

^AFine and Gray model of cumulative incidence used to account for competing risks

^BModels account for clustering at the census tract (SVI) and census block group (ADI)

^CTest for trend

Before evaluating attenuation, we confirmed that race/ethnicity was significantly associated with survival in our dataset (Figure 2, Table 3). Results from a Fine and Gray hazard model adjusted for insurance and histology-grade category demonstrated that NHBW had a hazard of EC-specific mortality that was 1.43 times that of NHWW (95% CI: 1.25, 1.63). Fine and Gray hazard models were also run stratified by histology-grade category and adjusted for insurance. Among women with low-grade histology, the hazard ratio was 1.65 (95% CI: 1.29, 2.12) comparing NHBW to NHWW; among women with high-grade histology, the hazard ratio was 1.36 (95% CI: 1.16, 1.59). Similar results were seen for Cox proportional hazards models where the outcome was all-cause mortality (Supplementary Table S3).

Figure 2. All-cause and endometrial-cancer specific survival by race and histology-grade group.

The figures in the left column show the all-cause (top) and endometrial-cancer specific (bottom) survival for all women in the tri-county area by race (non-Hispanic White or non-Hispanic Black). The figures in the middle column are subset to only those with high-grade histology (HGH); the figures in the right column are subset to only those with low-grade histology (LGH).

Table 3.

Sub-distribution hazard ratios^A for endometrial cancer-specific mortality comparing Black versus White women with endometrial cancer, adjusted for insurance status and histology-grade category

	OverallB; N = 7,125			Low-grade histologyB; N = 4,938			High-grade histologyB; N = 2,187
	SHR	95% CI	P	SHR	95% CI	P	SHR	95% CI	P
Race
White	Reference			Reference			Reference
Black	1.43	(1.25, 1.63)	<0.001	1.65	(1.29, 2.12)	<0.001	1.36	(1.16, 1.59)	<0.001
Insurance
Private	Reference			Reference			Reference
Public	1.66	(1.45, 1.91)	<0.001	2.70	(2.11, 3.45)	<0.001	1.28	(1.09, 1.50)	0.003
Histology-grade category
Low	Reference
High	4.60	(4.00, 5.28)	<0.001

^AFine and Gray model of cumulative incidence used to account for competing risks

^BClustering between census tract accounted for in model

Next, we assessed the association between race/ethnicity and each area-based deprivation measure using Chi-squared tests (Supplementary Table S4). For ADI, Q1 included scores less than 3, Q2 from 3–4, Q3 from 5–7, and Q4 from 8–10. For SVI, Q1 included scores less than 0.2134, Q2 from 0.2134 to less than 0.4775, Q3 from 0.4775 to less than 0.7453, and Q4 from 0.7453 to 1. The distribution was statistically significantly different between NHBW and NHWW for both ADI and SVI, where more NHBW were in the higher quartiles (i.e., more disadvantage) for ADI (e.g., 60% of NHBW lived in neighborhoods in ADI Q4 vs 14% of NHWW) and SVI (e.g., 56% of NHBW lived in neighborhoods in SVI Q4 vs 15% of NHWW). We then compared the results from Tables 2 and 3 and confirmed that the association between race/ethnicity and survival decreased with the addition of the neighborhood-based deprivation index (HR = 1.43 without ADI/SVI; HR = 1.38 with ADI; HR = 1.30 with SVI).

Upon evaluation of attenuation (Table 4), we found that ADI and SVI attenuated 18% (95% CI: 3–38%) and 27% (95% CI: 10–48%) of the association between race/ethnicity and EC-specific survival, respectively. Among only those with low-grade histology, ADI and SVI attenuated 12% (95% CI: −4–31%) and 14% (95% CI: −5–38%), respectively; among those with high-grade histology, ADI and SVI attenuated 24% (95% CI: 3–61%) and 40% (95% CI: 16–78%), respectively. Results were similar for all-cause mortality (Supplementary Table S5).

Table 4.

Proportion of association between race and endometrial cancer-specific mortality attenuated by neighborhood deprivation measures

	HR	95% CI	P	Prop Attenuated	95% CI	P
Overall (N=7052)
Adjusted	1.49	(1.31, 1.70)	<0.001
Adjusted + ADI	1.38	(1.18, 1.59)	<0.001	0.18	(0.03, 0.38)	0.02
Adjusted + SVI	1.32	(1.13, 1.54)	<0.001	0.27	(0.10, 0.48)	<0.001
Low-grade hist (N=4886)
Adjusted	1.72	(1.36, 2.19)	<0.001
Adjusted + ADI	1.61	(1.25, 2.07)	<0.001	0.12	(−0.04, 0.31)	0.15
Adjusted + SVI	1.59	(1.22, 2.06)	<0.001	0.14	(−0.05, 0.38)	0.13
High-grade hist (N=2166)
Adjusted	1.42	(1.21, 1.66)	<0.001
Adjusted + ADI	1.28	(1.07, 1.54)	0.007	0.24	(0.03, 0.61)	0.03
Adjusted + SVI	1.22	(1.02, 1.46)	0.03	0.40	(0.16, 0.78)	0.002

^*All models adjusted for insurance; overall adjusted for histology group category

^*Accounting for clustering done in all models

Discussion

Here, we investigated the association between two neighborhood-level deprivation indices and survival among women with EC in Metropolitan Detroit. We found that neighborhood deprivation was associated with survival in this population: those who lived in neighborhoods with more socioeconomic disadvantage had higher mortality. Additionally, our results indicate that neighborhood deprivation partially attenuates the widely established association between race/ethnicity and survival among women with EC(1,5). Interestingly, the proportion of this association attenuated by neighborhood deprivation is higher among women with high-grade histology compared to low-grade histology. Finally, the findings of these analyses suggest that SVI serves as a stronger predictor of survival and accounts for a greater proportion of the relationship between race/ethnicity and survival compared to ADI.

Our findings align with prior work that has demonstrated the association between increased neighborhood deprivation and poorer cancer outcomes(16–25). The majority of these studies, however, have focused on singular measures of area deprivation, such as persistent poverty(48). This study enhances our understanding of the relationship between neighborhood deprivation and EC survival by incorporating deprivation indices that account for factors across multiple social determinants, rather than focusing on a singular social context of health. Additionally, much research on socioeconomic deprivation and EC outcomes has thus far focused on individual- rather than area-level factors(49). However, one recent study from Gamble et al(50) found similarly increased rates of mortality among women living in more disadvantaged neighborhoods using the ADI. Our findings add to this work by exploring this association stratified by histology-grade category and the amount of the association between race/ethnicity and survival that can be attributed to neighborhood-level deprivation using composite indices.

The specific mechanisms underlying the association between neighborhood disadvantage and cancer outcomes remain poorly understood, however we propose several potential hypotheses. Access to diagnostic and treatment-related health care has been shown to differ by area deprivation. For example, those living in areas with greater deprivation have lower rates of cancer screening(10) and present with more advanced stage at diagnosis(51). Additionally, one study found treatment delays were increased among women with breast cancer living in areas of higher deprivation(52). In our cohort, a higher proportion of NHBW lived in neighborhoods with higher levels of deprivation and were diagnosed at later stages compared to NHWW. Less access to diagnostic care likely plays a role in stage at diagnosis, and this, along with treatment delays, may contribute to the survival disparities seen between NHBW and NHWW with EC. Differential utilization of healthcare services across varying levels of neighborhood deprivation may be linked to concepts of medical mistrust in health care settings and trustworthiness of health care institutions, particularly within populations that include racially and ethnically marginalized groups. For instance, a framework for evaluating medical mistrust suggests that unequal distribution of societal resources could serve as a pathway contributing to such mistrust(53).

We saw that neighborhood disadvantage attenuated a larger proportion of the association between race/ethnicity and survival among those in the high-grade histology subgroup. These subtypes are more aggressive than low-grade histologic tumors and are often diagnosed at later stages(54). Thus, for those with less access to health care, there may be delays in detection that have more significant impacts on survival for those with high-grade histology. Additionally, while treatment options are still limited for high-grade histologic subtypes of EC, recent drug trials have demonstrated improved treatment response among certain molecular subtypes(55). Grant et al. demonstrated that areas with more disadvantage have fewer clinical trial sites(56), likely limiting access to the most advanced treatments. Lack of both diagnostic and treatment-related health care in areas with more disadvantage may explain why area deprivation appeared to contribute more to the association between race/ethnicity and survival among those with high-grade histology. Additionally, due to myriad complex and interconnected factors, those living in areas with higher disadvantage often have higher rates of multimorbidity(57,58), and management of these morbidities is often challenging for those experiencing socioeconomic deprivation(59). Thus, those with high-grade histology living in areas with higher disadvantage may be more likely to have comorbidities that would exclude them from more aggressive drug trials due to either the assumption or the reality that they may not be able to tolerate more toxic treatments(60).

While living in a neighborhood with higher levels of disadvantage as categorized by either ADI or SVI was associated with higher hazard of mortality in our analytic cohort, SVI better differentiated hazard by group. Our Cox models demonstrated a clearer linear pattern of increasing hazard by increasing deprivation, and our attenuation analyses suggested that SVI attenuated a larger proportion of the association between race/ethnicity and survival compared to ADI. ADI is constructed using 17 variables from the ACS focused on socioeconomic status (N=6), household characteristics (N=3), and housing type and price (N=8). SVI is constructed using 16 variables related to socioeconomic status (N=5), household characteristics (N=5), race and ethnicity (N=1), and housing type and transportation (N=5). Variables common to both indices include percent living in poverty, percent unemployment, percent with less than a high school diploma, percent single parent households, percent of houses with no vehicles, and percent of houses with crowding. ADI includes additional economic variables (e.g., household poverty and income, average rent, mortgage and home value), while SVI includes additional population characteristics (e.g., age 65 and older, age 17 and younger, disabilities, English language proficiency, race and ethnic minority). A potential hypothesis for the stronger trend between SVI quartiles and survival compared to ADI is that SVI, with the inclusion of those under 18, over 65, and with disabilities, better captures populations with high rates of caregivers, a population that may experience barriers in accessing healthcare(61,62). Additionally, a study comparing ADI and SVI found that the inclusion of housing price variables in the ADI may lead to incorrect classifications of less deprivation in urban gentrified areas that still have high levels of poverty, crowding, and limited education(41).

It was notable that the SVI, which is linked by census tract, exhibited a clearer trend compared to the ADI, which uses census block group and represents a more granular geolocation variable. We hypothesize that meaningful exposures related to area socioeconomic deprivation may be more relevant at larger geographic levels. For instance, access to health care may be better represented at the census tract level, as resource allocation is often determined at broader geographic units. For example, the Centers for Medicare and Medicaid Services frequently reference county and census tract areas when discussing resource allocation(63). The modifiable areal unit problem is a known issue when conducting area-based analyses, leading to different estimates based on the selected area unit(64). The results from this study demonstrate variations in estimates at differing areal units, providing important evidence for the importance of considering the areal unit most appropriate for a given analysis.

While access to healthcare likely contributes to the association between area-based socioeconomic deprivation and survival among women with EC, there are likely other important factors contributing to this association. One such factor may be chronic stress, which has been associated with a range of diseases, including cancer(65). Biological mechanisms related to chronic stress are associated with outcomes across the cancer continuum, including tumorigenesis, progression, metastasis, drug resistance, and survival(66). Given that neighborhood deprivation is associated with chronic stress(67–69), chronic stress may be contributing to the association seen between area deprivation and survival. Furthermore, there is some evidence that Black cancer patients experience higher levels of stress compared to their White counterparts(70). This could in part explain the attenuation between the association between race/ethnicity and survival among women with EC when area-based deprivation measures are added to the model.

The strengths of this study include the use of 10 years of comprehensive and complete data from a NAACCR certified cancer registry. Additionally, the data come from a population with a high proportion of NHBW with EC, a population of high interest due to large survival disparities that have persisted for decades. EC cases recorded in the MDCSS from 2013–2022 included a cohort of women who were 72% NHW, 23% NHB, 2% Hispanic, 2% Non-Hispanic Asian, and 1% other race/ethnicity. This is similar to the racial and ethnic distribution of the tri-county area of Metropolitan Detroit (Macomb, Oakland, and Wayne counties)(71). Conducting this study in the metropolitan Detroit area, which has a large NHB population, allowed for the inclusion of a large proportion of NHBW (23% compared to 14% in the US population(United States Census Bureau QuickFacts for Detroit Michigan: https://www.census.gov/quickfacts/fact/table/detroitcitymichigan,MI/PST045221)), who experience the worst EC outcomes(1,5), in our study population. There are also potential limitations which must be considered when interpreting our findings. Our data were limited to the variables available in the MDCSS, and thus we were unable to consider individual variables such as education, income, and health history. Additionally, due to low counts, we excluded races and ethnicities other than NHW and NHB and thus our results may not be generalizable to other diverse populations. Finally, we were only able to consider the associations at the census tract and block group level; these associations may differ when considering other areal units(64).

This study demonstrated a significant association between neighborhood-level disadvantage and both EC-specific and all-cause mortality among women with EC living in metropolitan Detroit, even after adjustment for known individual social and biological risk factors, such as race/ethnicity, insurance status, and tumor characteristics. Additionally, we saw that neighborhood disadvantage attenuates the relationship between race/ethnicity and survival, particularly among those with high-grade histology. These findings serve as motivation to understand how neighborhood impacts cancer outcomes. While composite measures are in many ways better at capturing area disadvantage compared to single factors, it is challenging to determine which factors, or set of factors, are driving the relationship between composite measures and outcomes. Additional studies are needed to identify factors driving the association between composite area-based socioeconomic disadvantage measures and survival in this population, as well as develop location-specific disadvantage measures facilitating a deeper understanding of the mechanisms underlying the association.