Disparities in geospatial patterns of cancer care within urban counties and structural inequities in access to oncology care

Abstract

Objective

To examine geospatial patterns of cancer care utilization across diverse populations in New Jersey—a state where most residents live in urban areas.

Data Sources/Study Setting

We used data from the New Jersey State Cancer Registry from 2012 to 2014.

Study Design

We examined the location of cancer treatment among patients 20–65 years of age diagnosed with breast, colorectal, or invasive cervical cancer and investigated differences in geospatial patterns of care by individual and area‐level (e.g., census tract‐level) characteristics.

Data Collection/Extraction Methods

Multivariate generalized estimating equation models were used to determine factors associated with receiving cancer treatment within residential counties, residential hospital service areas, and in‐state (versus out‐of‐state) care.

Principal Findings

We observed significant differences in geospatial patterns of cancer treatment by race/ethnicity, insurance type, and area‐level factors. Even after adjusting for tumor characteristics, insurance type, and other demographic factors, non‐Hispanic Black patients had a 5.6% higher likelihood of receiving care within their own residential county compared to non‐Hispanic White patients (95% CI: 2.80–8.41). Patients insured with Medicaid and those without insurance had higher likelihoods of receiving care within their residential county compared to privately insured individuals. Patients living in census tracts with the highest quintile of social vulnerability were 4.6% more likely to receive treatment within their residential county (95% CI: 0.00–9.30) and were 2.7% less likely to seek out‐of‐state care (95% CI: −4.85 to −0.61).

Conclusions

Urban populations are not homogenous in their geospatial patterns of cancer care utilization, and individuals living in areas with greater social vulnerability may have limited opportunities to access care outside of their immediate residential county. Geographically tailored efforts, along with socioculturally tailored efforts, are needed to help improve equity in cancer care access.

1. INTRODUCTION

Inequities exist along the cancer care continuum, creating disparate cancer outcomes among diverse populations. For example, minoritized racial and ethnic groups, populations without health insurance, and underinsured individuals are less likely to receive screenings and treatment for various cancers, which can lead to increased morbidity and mortality. ¹ , ² , ³ , ⁴ , ⁵ Prior work has demonstrated factors contributing to cancer disparities at all levels, from the patient to the health care system. ⁶ An emerging area of research has focused on the role of geography and/or geospatial patterns of cancer care, such as rural/urban differences in cancer care delivery. ⁷ , ⁸ , ⁹ , ¹⁰ Results from this research suggest that disparities between rural and urban populations exist along the cancer continuum from screening, follow‐up, and treatment to mortality, with rural areas faring worse. ⁷ , ¹¹ , ¹² For example, research suggests the availability of providers treating colorectal cancer (e.g., radiation oncologists, gastroenterologists, and surgeons) is more limited in rural settings when compared to urban settings. ¹³

Prior research has largely focused on this urban/rural dichotomy; most studies have examined the availability of cancer care services in urban versus rural areas ⁸ , ¹² , ¹⁴ , ¹⁵ because of the assumption that the increased availability of providers in urban areas ensures equal access and utilization to all residents within a given urban area. However, there is an incomplete understanding of factors that may influence where cancer care is received in urban areas. ¹⁶ , ¹⁷ , ¹⁸ Importantly, patients residing in urban areas with limited resources (e.g., higher percentages of residents at or below the federal poverty level and without insurance coverage) and who are affected by structural racism (e.g., through residential racial segregation) may be unable to access cancer care and experience poor cancer outcomes despite the increased geographic availability of oncology providers in urban areas. In fact, Pruitt and colleagues found that residential racial segregation was associated with increased mortality among patients with breast cancer living in urban areas. ¹⁹ Historical forces related to structural racism—such as redlining (i.e., the racially discriminatory federal government practice of grading communities' mortgage credit‐worthiness, resulting in forced racial segregation and intentional underdevelopment of those communities), ²⁰ mortgage discrimination, and racially restrictive covenants (e.g., the prohibition of Black Americans and/or other racial and ethnic groups from the legal purchase or leasing of housing)—have caused low‐income and minoritized racial and ethnic populations to disproportionately reside in communities that have been refused resources needed to promote health and well being. ²¹ , ²² , ²³ , ²⁴ , ²⁵ Redlining in particular impacts urban populations in both the demographic composition of these environments (e.g., minoritized racial and ethnic populations are more likely to reside in urban areas) and overall health inequities, including those related to cancer. ²⁴ , ²⁶ , ²⁷

Populations with increased social risks and limited resources living in urban areas navigate multiple barriers to health care. For example, they often experience transportation barriers, discrimination, and limited access to specialty providers. ⁶ , ²⁸ , ²⁹ , ³⁰ Factors related to urbanicity, or urban area characteristics and their interaction with urban populations to influence health, can affect direct geographic/physical access to care and humanistic elements, such as perceived lack of high‐quality care or perceptions on proximity to care within one's immediate neighborhood. ³¹ , ³² , ³³ Overall, more research is needed to understand inequities in cancer care utilization within urban areas.

Inequities in cancer care utilization are particularly relevant in New Jersey, a state consisting of over 9 million residents with considerable racial, ethnic, and socioeconomic diversity. Approximately 90–95% of New Jersey residents reside in urban areas, as defined by the U.S. Census Bureau. ³⁴ New Jersey is situated between two large metropolitan areas with high densities of medical facilities (i.e., New York City and Philadelphia) and—for reasons that are not completely understood—has higher overall age‐adjusted cancer incidence rates compared to the national average. ³⁵ , ³⁶ Structural determinants for cancer care delivery (e.g., access to income, transportation, childcare, and health insurance) become especially nuanced in densely populated areas like New Jersey where the geographic availability of oncology specialty providers is high, ³⁷ but important disparities in care utilization may exist among populations who cannot access these specialty providers. Understanding geospatial patterns of cancer care is critical to unmasking potential inequities in care access and utilization within urban populations. Therefore, in this study, we aimed to describe geospatial patterns of cancer care utilization across diverse population subgroups in New Jersey—a state where most residents live in urban areas.

2. METHODS

We conducted a cross‐sectional geospatial analysis to describe the relationship between residential location at time of cancer diagnosis and location of the first course of cancer treatment among persons aged 20–65 with breast (BC), colorectal (CRC), and invasive cervical cancer (ICC) identified by the New Jersey State Cancer Registry (NJSCR). The NJSCR is the state‐designated repository for population‐wide cancer surveillance data, collecting individual demographic, tumor, treatment, and vital status information for every New Jersey resident diagnosed and/or treated for cancer. We selected these cancer types because they all have clear, national screening guidelines. ³⁸ , ³⁹ , ⁴⁰ We explored the relationships between individual and area‐level factors and the location of cancer care. We focused on differences in location of cancer care by insurance type and race/ethnicity to inform whether differences emerge in access to and receipt of care across population subgroups. We employed the socio‐ecological framework because it considers multiple levels of influence on health outcomes. We examined the individual, neighborhood or community, and local/state health system levels to guide our analysis and interpretation of findings across the cancer continuum. ⁴¹ , ⁴² NJSCR has interstate data exchange agreements with other states to collect information on residents if they were diagnosed with or treated for cancer out‐of‐state. We linked NJSCR geocoded information with data from the 2014 Centers for Disease Control and Prevention Social Vulnerability Index (CDC SVI) using 5‐year estimates (2010–2014) by census tract (CT). ⁴³ We used CTs in this study to define residential neighborhood characteristics for each person with cancer within our study sample.

2.1. Study Population

We identified persons with a first primary breast (n = 9561), colorectal (n = 3311), or invasive cervical (n = 457) cancer diagnosed between January 1, 2012, and December 31, 2014. Primary site and histology were coded according to the International Classification of Diseases for Oncology, third edition (ICD‐O‐3). ⁴⁴ Male and female CRC patients and female BC and ICC patients were eligible for inclusion in this study. Persons with multiple primary cancers occurring within the same time frame (2012–2014), persons who were not residents of the state of New Jersey at the time of diagnosis (i.e., residential ZIP code at the time of diagnosis outside of New Jersey), younger than 20 or older than 65 years of age at time of diagnosis, and those reported to the registry via death certificate or autopsy only were excluded. We included adults ages 20–65 to focus on the pre‐Medicare population and capture younger individuals diagnosed with cancer and enrolled in Medicaid. A small number of individuals age 65 (n = 710) were included to conservatively capture individuals who may have experienced cancer diagnosis and workup prior to Medicare eligibility. Our analyses only focused on the first incident cancer diagnosis.

2.2. Measures

2.2.1. Individual characteristics

Individual demographic and tumor‐related characteristics were obtained from the NJSCR. We examined the surveillance, epidemiology, and end results (SEER) summary stage (early versus late) and age at diagnosis for each person in our 3‐year cohort. SEER summary stage was grouped as either early‐stage (in‐situ or localized) or late‐stage (regional or distant). Individual‐level demographic information included race/ethnicity, gender, year of diagnosis, county of residence at the time of diagnosis, and primary payer at time of diagnosis or treatment. Race/ethnicity was measured using SEER categories, which apply the following mutually exclusive categories: Hispanic, Non‐Hispanic (NH) White, NH Black, and Other NH Racial Groups combined (NH Asian/Pacific Islander, NH Other/Unknown/Missing). ⁴⁵ Primary payer at the time of diagnosis or treatment, from SEER, included the following mutually exclusive categories: private insurance (managed care, HMO, PPO, or fee‐for‐service), Medicaid (administered through a managed care plan), Medicare (administered through a managed care plan, Medicare supplemental insurance, with private supplement, and with Medicaid eligibility), and uninsured. We used ZIP codes from person's residence and treatment facility to determine if they stayed within or went outside their area to seek care. We excluded persons with unknown, missing, or other public primary payer status, unknown disease stage, residential ZIP codes outside the state of New Jersey, or with missing residential ZIP codes, as well as those missing treatment facility information and those who did not receive treatment. In total, we excluded 6008 persons from the final analytic dataset (Figure S1).

2.2.2. Area‐level measures

We used geocoded data from the NJSCR, which routinely collects addresses at the time of diagnosis for each person with cancer reported to the registry. Geocoding to county and CT is conducted through a combination of automated and manual processes through the NJSCR database management system, SEER*DMS. Nearly all (99.5%) of the 2010 CTs in New Jersey consisted of at least one person with BC, CRC, or ICC diagnosed between 2012 and 2014.

Residential census tract for each NJSCR person with cancer was linked to the 2014 Social Vulnerability Index, a 5‐year population estimate (2010–2014) collected by the CDC. ⁴³ , ⁴⁶ The SVI provides measures of relative vulnerability (socially and spatially) for every U.S. CT. We used the composite measure of overall social vulnerability that included four major themes: socioeconomic status (persons below poverty, persons unemployed, per capita income, and persons with no high school diploma); household composition and disability (persons aged 65 or older, persons aged 17 or younger, civilians with a disability, and single‐parent households); minority status and language (persons who identify as a racial/ethnic minority and persons who speak English “less than well”); and housing type and transportation (persons living in multi‐unit structures, mobile homes, group quarters, or crowded conditions and have no vehicle). For each CT, we calculated the percentile of overall social vulnerability based on the CT population estimates and created quintiles (1 = lowest 20% vulnerability and 5 = highest 20% vulnerability) based on the distribution across all CTs in our sample.

2.2.3. Geographic measures

All health care facilities, such as physician's offices and ambulatory care facilities, that provide treatment for patients with cancer (i.e., treatment locations) must report information to the NJSCR. We therefore constructed three measures for treatment location using a combination of residential location at the time of diagnosis and treatment facility location for each person: received treatment within same county of residence, received treatment within the same hospital service area (HSA) of residence, and received treatment outside the state of New Jersey. Because health care delivery and organization are often determined at larger geographic units than CTs or ZIP codes, we used residential county and HSA‐catchment area to examine geospatial patterns of cancer care. ⁴⁷ HSAs are defined by the Centers for Medicare & Medicaid Services and represent a collection of ZIP codes whose Medicare residents receive most of their hospitalizations from the hospitals in that area. There are 65 HSAs in New Jersey. We conducted a sensitivity analysis of receipt of treatment within residential ZIP code tabulation areas (ZCTA) but found that nearly 95% of persons with cancer were treated outside of residential ZCTA; therefore, the limited variation did not warrant further exploration. We examined the completeness of geographic information for diagnosis and treatment locations among persons with New Jersey residential ZIP codes. While 99.9% had a viable residential ZIP code, a substantial proportion of the cohort were missing diagnosing facility ZIP codes (28.2%). A smaller proportion had a missing treatment facility location (9.5%).

2.3. Statistical analysis

Descriptive statistics were used to summarize the characteristics of the cohort for all cancer sites combined, for each cancer site, and between early and late‐stage diagnoses. For ease of interpretation, we used logistic regression models to examine the relationship between each treatment outcome measure and individual‐ and area‐level measures. For our measure of social vulnerability, we compared persons living in the quintile of highest vulnerability (Quintile 5) to persons living in the lowest quintile of vulnerability (Quintile 1). We conducted generalized estimating equation (GEE) models with an exchangeable correlation structure and robust standard errors, accounting for clustering of persons within residential CTs, for the three treatment location measures described above. Adjusted models included the following variables: cancer site, insurance, race/ethnicity, age at diagnosis, year of diagnosis, late versus early‐stage diagnosis, and Social Vulnerability Index score. Model specifications were based on significant relationships in unadjusted GEE models, prior literature, and assessments for multicollinearity. Site‐specific GEE models were used to assess if relationships between predictors and the three treatment location‐related outcomes of interest varied across cancer sites, and results were reported using average marginal effects. We estimated 95% confidence intervals (CI) and determined statistical significance at the p < 0.05 level.

While the majority of NJ residents live in urban areas, approximately 5%–10% of the population resides in seven counties that meet the U.S. Census density definition of rural (e.g., <500 people/square mile). ⁴⁸ To assess if our results remained consistent when only urban counties were considered, we conducted a sensitivity analysis excluding the seven rural counties. Data were analyzed using Stata v15.1 (College Station, TX).

2.3.1. Geospatial data visualization

In addition, we created maps to better visualize cancer incidence rates and distance traveled for treatment. Breast, cervical, and colorectal cancer incidence rates by ZCTA were age‐standardized to the 2000 U.S. Census population (10 age groups) by means of the direct method. Rates are presented as the number of persons with cancer per 100,000 population. The ZCTA rates were smoothed using a weighted head‐banging algorithm. Head‐banging is a weighted two‐dimensional median‐based smoothing algorithm developed to reveal underlying geographic patterns in data where the values to be smoothed do not have equal variances. ⁴⁹ Rates were smoothed using Headbang software developed by the National Cancer Institute. ⁵⁰ ZCTA Census populations used to estimate the incidence rates were from the American Community Survey, 5‐year estimates (2010–2014). ⁵¹ We estimated the straight‐line distance from ZCTA origin of persons with cancer for each cancer type to the treatment ZCTA. The maps show persons traveling ≤20 km and >20 km to treatment locations. Maps were created using QGIS 3.2.

3. RESULTS

3.1. Descriptive characteristics

Demographic and tumor‐related characteristics are shown in Table 1. There were a total of 13,329 persons with cancer, 9561 with breast cancer (BC), 3311 colorectal cancer (CRC), and 457 invasive cervical cancer (ICC), diagnosed from 2012 to 2014 in New Jersey that met our inclusion criteria for the analytic cohort. Over half (59.6%) of all persons with CRC and nearly half (47.3%) of all persons with ICC were diagnosed at a late stage, while roughly a quarter (27.2%) of persons with BC were diagnosed at a late stage. Approximately two‐thirds of persons with BC (69.1%) and CRC (66.4%) were NH White persons, which is slightly higher than the percent of NH‐White persons (59%) in the state of New Jersey overall. ⁵² Larger proportions of persons with BC (81.7%) were privately insured compared to CRC (68.2%) and ICC (63.5%).

TABLE 1.

Individual and area‐level characteristics of persons diagnosed with breast, colorectal, and cervical cancer, New Jersey State Cancer Registry, 2012–2014.

	Total (n = 13,329)		Breast (n = 9561)		Colorectal (n = 3311)		Cervical (n = 457)
Characteristics	n	%	n	%	n	%	n	%
Individual‐level
Race/Ethnicity
NH White	9072	68.1	6608	69.1	2198	66.4	266	58.2
NH Black	1601	12.0	1048	11.0	494	14.9	59	12.9
NH Other	1107	8.3	853	8.9	210	6.3	44	9.6
Hispanic	1549	11.6	1052	11.0	409	12.4	88	19.3
Insurance type
Private	10,363	77.7	7815	81.7	2258	68.2	290	63.5
Medicaid	548	4.1	334	3.5	173	5.2	41	9.0
Medicare	1282	9.6	796	8.3	452	13.7	34	7.4
Uninsured	1136	8.5	616	6.4	428	12.9	92	20.1
Sex
Female	11,516	86.4	9561	100.0	1813	54.8	457	100.0
Male	1813	13.6	—	—	1498	45.2	—	—
Age at diagnosis
20–39 Years	982	7.4	663	6.9	192	5.8	127	27.8
40–49 Years	3620	27.2	2880	30.1	610	18.4	130	28.4
50–65 Years	8727	65.5	6018	62.9	2509	75.8	200	43.8
Stage at diagnosis
Early	8535	64.1	6957	72.8	1337	40.4	241	57.2
Late	4794	35.9	2604	27.2	1974	59.6	216	47.3
Diagnosis year
2012	4673	35.1	3335	34.9	1172	35.4	166	36.3
2013	4819	36.2	3532	36.9	1130	34.1	157	34.4
2014	3837	28.8	2694	28.2	1009	30.5	134	29.3
Area level*
Social Vulnerability Index Score
Quintile 1 (lowest vulnerability)	2670	20.0	2074	21.7	536	16.2	60	13.1
Quintile 2	2665	20.0	2020	20.5	578	18.9	67	17.9
Quintile 3	2665	20.0	1956	20.5	627	18.9	82	17.9
Quintile 4	2670	20.0	1807	18.9	743	22.4	120	26.3
Quintile 5 (highest vulnerability)	2659	19.9	1704	17.8	827	25.0	128	28.0

3.2. Area‐level characteristics

Figure 1 visually presents the locations of where persons with cancer received treatment relative to overall cancer incidence by ZCTA. Figure 2 visually presents travel distances to BC, ICC, and CRC treatment locations. Overall, 87.1% of persons with CRC, 68.2% of persons with BC, and 66.0% of persons with ICC traveled less than 20 km to receive treatment. Conversely, 12.9% of persons with CRC, 31.8% of persons with BC, and 34.0% of persons with ICC traveled greater than 20 km to receive treatment. A higher proportion of persons with ICC resided in the CT with the highest quintile of social vulnerability.

FIGURE 1.

Cancer incidence and cancer treatment receipt location. [Color figure can be viewed at wileyonlinelibrary.com]

FIGURE 2.

Distance to treatment locations. [Color figure can be viewed at wileyonlinelibrary.com]

Location of cancer treatment by cancer site across racial/ethnic and insurance subgroups is shown in Table 2. Half (53.4%) of all persons with cancer received cancer treatment within the same county as their residence, and a third of these same individuals (32%) received treatment within their HSA‐catchment area. Eleven percent of persons with cancer were treated out of state. The proportion of persons with cancer staying within county or HSA‐catchment area of residence for treatment differed by insurance type, race/ethnicity, and all area‐level socio‐demographic measures. For example, 66% of Medicaid insured and 68% of uninsured patients remained within their residential county for cancer treatment, compared to only 50% of privately insured patients. Similarly, rates of receiving treatment within residential county were significantly higher among NH Black patients (63.9%) and Hispanic patients (56.4%) compared to NH White patients (50.8%). Rates of receiving treatment out‐of‐state were significantly higher among privately insured (13%), and for NH White patients (12.5%) and NH Other race (12.9%) patients compared to other groups. Similar patterns were observed when examining treatment location by quintiles for area‐level social vulnerability, where higher proportions of persons with cancer residing in CTs with the highest vulnerability (Quintile 5) remained within their residential county (59.2%) or HSA‐catchment area (37.9%) for cancer treatment. Conversely, a higher proportion of patients living in CTs with the lowest vulnerability (Quintile 1) traveled out of state for cancer treatment (13.2%) compared to those living in CTs with the highest vulnerability (Quintile 5; 6.1%).

TABLE 2.

Geospatial treatment patterns by individual and area‐level characteristics among persons diagnosed with cancer, New Jersey State Cancer Registry, 2012–2014.

	% Treated within county	% Treated within HSA‐catchment area	% Treated out of state
Overall
All persons diagnosed with cancer	53.4	32.0	11.0
Cancer site
Breast	52.3	31.1	11.4
Colorectal	57.9	35.9	10.0
Cervical	45.1	23.0	9.4
Stage at diagnosis
Early	51.9	31.2	11.4
Late	56.1	33.5	10.1
Age at diagnosis
20–39 Years	46.9	27.5	16.0
40–49 Years	51.1	29.7	12.8
50–65 Years	55.1	33.5	9.6
Race/Ethnicity
NH White	50.8	30.2	12.5
NH Black	63.9	37.3	5.9
NH Other	55.6	35.0	12.9
Hispanic	56.4	34.9	5.5
Insurance Type
Private	50.1	29.7	13.0
Medicaid	66.1	40.7	1.3
Medicare	61.3	37.8	7.4
Uninsured	68.3	42.1	0.9
Social Vulnerability Index Score
Quintile 1 (lowest vulnerability)	50.7	31.3	13.2
Quintile 2	51.2	33.0	12.7
Quintile 3	52.7	28.4	11.8
Quintile 4	53.3	29.3	11.0
Quintile 5 (highest vulnerability)	59.2	37.9	6.1
Note. NH = Non Hispanic; HSA = Hospital Service Area

3.3. Adjusted model results

Adjusted average marginal effects are presented in Table 3 and show significant individual‐ and area‐level factors associated with all three geospatial treatment location measures. Persons with cancer with Medicaid, Medicare, or no insurance had a significantly higher likelihood of receiving treatment within their HSA‐catchment area compared to those who were privately insured (Table 3). Similar relationships were observed for treatment within county. NH Black persons had a 5.6% higher likelihood of receiving treatment within their residential county (95% CI 2.80–8.41) compared to NH White patients. Persons living in the CTs with the highest quintile of vulnerability (Quintile 5) had a 4.6% higher likelihood of receiving treatment within their residential county (95% CI 0.00–9.30), a 4.3% higher likelihood of remaining within their HSA‐catchment area (95% CI 0.30–8.41), and a 2.7% lower likelihood of seeking care out of state (95% CI −4.85 to −0.61).

TABLE 3.

Average marginal effects of factors associated with treatment location among persons diagnosed with cancer, New Jersey State Cancer Registry, 2012–2014, N = 13,329.

	Treated in the same county				Treated in the same HSA‐catchment area				Treated out of state
	Unadjusted ME		Adjusted ME		Unadjusted ME		Adjusted ME		Unadjusted ME		Adjusted ME
	PP	95% CI	PP	95% CI	PP	95% CI	PP	95% CI	PP	95% CI	PP	95% CI
Cancer site
Breast	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐
Colorectal	5.29	(3.44, 7.13)	2.63	(0.72, 4.54)	4.62	(2.76, 6.48)	2.61	(0.72, 4.51)	−1.18	(−2.39, 0.03)	0.78	(−0.60, 2.15)
Cervical	−6.20	(−10.4, −2.03)	−7.55	(−11.8, −3.29)	−7.85	(−11.7, −3.96)	−8.91	(−12.7, −5.12)	−1.48	(−4.17, 1.20)	−0.46	(−3.38, 2.46)
Age at diagnosis
20–39 years	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐
40–49 years	3.01	(−0.17, 6.18)	3.14	(−0.02, 6.30)	1.01	(−1.96, 3.97)	0.82	(−2.16, 3.81)	−3.36	(−5.93, −0.78)	−4.23	(−6.84, −1.62)
50–65 years	7.75	(4.78, 10.7)	6.62	(3.65, 9.60)	5.17	(2.44, 7.90)	4.01	(1.25, 6.76)	−6.51	(−8.94, −4.08)	−7.01	(−9.51, −4.50)
Diagnosis year
2012	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐
2013	0.53	(−1.24, 2.3)	0.50	(−1.26, 2.26)	0.75	(−0.91, 2.42)	0.71	(−0.95, 2.37)	−2.48	(−3.81, −1.14)	−2.31	(−3.63, −0.99)
2014	3.01	(1.10, 4.92)	3.03	(1.12, 4.94)	4.14	(2.22, 6.06)	4.06	(2.13, 5.98)	−6.29	(−7.64, −4.95)	−6.22	(−7.56, −4.89)
Race/Ethnicity
NH White	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐
NH Black	8.02	(5.31, 10.7)	5.61	(2.80, 8.41)	4.91	(2.28, 7.55)	2.58	(−0.04, 5.19)	−6.17	(−7.6, −4.74)	−4.33	(−6.06, −2.60)
NH Other	0.72	(−2.13, 3.58)	0.42	(−2.43, 3.27)	1.20	(−1.56, 3.96)	0.92	(−1.88, 3.72)	0.83	(−1.24, 2.91)	1.77	(−0.29, 3.83)
Hispanic	3.72	(1.02, 6.41)	1.37	(−1.40, 4.15)	3.59	(0.98, 6.20)	1.61	(−1.01, 4.24)	−6.53	(−7.93, −5.13)	−4.41	(−6.09, −2.74)
Insurance type
Private	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐
Medicaid	12.7	(8.91, 16.6)	11.7	(7.80, 15.6)	10.5	(6.25, 14.8)	9.09	(4.87, 13.3)	−11.6	(−12.8, −10.4)	−10.9	(−12.3, −9.44)
Medicare	10.6	(7.95, 13.2)	8.65	(5.97, 11.3)	7.73	(5.01, 10.5)	6.16	(3.43, 8.89)	−5.46	(−7.02, −3.91)	−4.20	(−5.90, −2.49)
Uninsured	15.1	(12.5, 17.8)	14.1	(11.4, 16.9)	10.7	(7.87, 13.6)	9.79	(6.88, 12.70)	−11.9	(−12.8, −10.9)	−11.4	(−12.4, −10.46)
Stage at diagnosis
Early	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐
Late	3.63	(2.02, 5.23)	2.13	(0.49, 3.78)	2.42	(0.86, 3.99)	1.16	(−0.44, 2.75)	−1.06	(−2.13, 0.00)	−0.38	(−1.49, 0.73)
Social Vulnerability Index Score
Quintile 1 (lowest vulnerability)	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐	Ref.	‐
Quintile 2	0.36	(−4.27, 4.99)	−0.44	(−5.06, 4.17)	2.11	(−2.01, 6.24)	1.58	(−2.62, 5.79)	−0.74	(−2.97, 1.50)	−0.12	(−2.07, 1.83)
Quintile 3	3.46	(−1.10, 8.02)	2.20	(−2.31, 6.72)	−1.48	(−5.43, 2.47)	−2.42	(−6.41, 1.57)	−1.62	(−3.70, 0.46)	−0.50	(−2.36, 1.36)
Quintile 4	4.61	(0.11, 9.11)	1.79	(−2.70, 6.28)	−0.13	(−4.06, 3.79)	−2.06	(−6.03, 1.92)	−2.42	(−4.45, −0.39)	−0.10	(−2.00, 1.80)
Quintile 5 (highest vulnerability)	11.2	(6.73, 15.6)	4.65	(0.00, 9.30)	8.87	(4.96, 12.8)	4.36	(0.30, 8.41)	−7.44	(−9.27, −5.60)	−2.73	(−4.85, −0.61)

3.4. Sensitivity analysis results

Results of the sensitivity analysis are presented in Table S1. The pattern of results remained largely consistent when the analyses were restricted to only urban counties, with one exception. In our main analyses, people experiencing the highest quintile of social vulnerability had a higher likelihood of staying within their residential county for treatment, but this association was no longer statistically significant in our sensitivity analysis.

4. DISCUSSION

Our study is one of only a few that uses population‐based cancer registry data to examine individual‐ and area‐level factors associated with location of cancer care in an urban, diverse population of persons aged 20–65. ⁵³ , ⁵⁴ , ⁵⁵ Our findings show that NH Black individuals are more likely to stay within their own residential county for cancer care compared to NH White individuals, even after controlling for tumor size and other individual and neighborhood demographic characteristics. This finding is consistent with prior studies that have examined the role of census tract poverty, historical racism, residential segregation, and other facets of the social context that contribute to health care utilization patterns among NH Black individuals. ⁵⁶ , ⁵⁷ , ⁵⁸ In addition, we observed that patients without insurance, patients with public insurance, and patients living in areas with the highest levels of social vulnerability are more likely to utilize care within their residential area than those with private insurance and living in areas with the lowest levels of social vulnerability. Similarly, a recent study also found that other forms of vulnerability (e.g., lower socioeconomic status) are associated with lower access to cancer care providers. ⁵⁹ Collectively, this result suggests that NH Black individuals and individuals who do not have private insurance may have more limited choices when seeking cancer care, especially across state lines, which may manifest as utilization patterns that encompass a smaller geographic area than other populations with more resources and, therefore, better access.

Our findings also have implications for conceptualizing HSA‐catchment areas in ways that are more centered on patients than individual health care facilities. The notion of medical neighborhoods, which evolved out of the patient‐centered medical home model, somewhat encompasses this idea by focusing on centralized health care coordination through primary care providers “surrounded by specialty clinics, ancillary service providers, and hospitals.” ⁶⁰ Medical neighborhoods for cancer care delivery become especially nuanced in densely populated areas where the geographic availability of oncology specialty providers is high, ³⁷ but factors related to a lack of high‐quality care, insurance type, or proximity to care within an individual's immediate neighborhood can influence where patients seek or eventually receive care. ³¹ , ³² , ³³ Our study suggests that medical neighborhoods for cancer care delivery may vary for different populations due to individual, provider, insurance, and community‐level factors. Indeed, medical neighborhoods for privileged populations may differ from those of populations exposed to structural racism and other disadvantages. Even though medical neighborhoods and HSAs have been used for health planning purposes, they may not encompass the areas where large population subgroups actually obtain their health care. Future research is needed to better conceptualize medical neighborhoods that are equitable, capturing geographic patterns of cancer care utilization across diverse urban populations. For example, health system planning and redesign efforts should thoughtfully consider how the composition and scope of provider networks within medical neighborhoods may differentially affect cancer care options for population subgroups.

Indeed, understanding geospatial patterns of health care access and utilization for urban populations is critical to inform local health system redesign efforts. Increasingly, these redesign efforts aim to reduce health care fragmentation and strengthen the integration of health services and systems—an action area of the Robert Wood Johnson Foundation's Culture of Health Action Framework. The Culture of Health Action Framework creates a holistic idea of health and wellness for individuals and their communities, incorporating social determinants of health (e.g., housing) and the re‐imagining and/or better integration of health care systems. ⁶¹ , ⁶² , ⁶³ This framework includes four action areas: (1) making health a shared value; (2) fostering cross‐sector collaboration to improve well‐being; (3) creating healthier, more equitable communities; and (4) strengthening integration of health services and systems. Our study suggests that health system redesign efforts to implement Action Area 4 are needed in urban areas with the greatest proportions of NH Black residents, individuals with Medicaid and/or who are uninsured, and in census tracts with the highest amounts of social vulnerability. Increasing the availability and continuity of coordinated medical care and social support in these areas may lead to better outcomes for all patients diagnosed with cancer.

Our study also found that at least 1 in 10 persons with cancer from the NJSCR received their primary treatment out of state. Geospatial approaches for cancer control research are often limited by predetermined, arbitrary state boundaries due to data limitations. For populations like New Jersey residents, who live near large metropolitan areas in other states with dense medical care, these predetermined boundaries may limit researchers' ability to accurately examine cancer care utilization patterns when patients cross state lines or other boundaries to receive care. Therefore, future studies may benefit from obtaining multi‐state data to accurately assess patterns of care in areas where patients may cross state boundaries. Additionally, future research is needed to understand the reasons patients may seek care from out‐of‐state providers (e.g., due to potentially shorter travel times, challenges finding providers who accept their health insurance, or other reasons).

There are important strengths to this study. First, we used population‐based cancer registry data to examine patterns of care in a diverse urban population. We also focused on location of actual care utilization rather than availability of services, which provides insight into how urban residents utilize and engage with cancer care in the urban environment. ⁶⁴ Overall, our study provides evidence for the need to recognize the limitations of what arbitrary geographic boundaries across socioeconomic and demographic subgroups can teach us about understanding and addressing inequities in cancer care delivery.

Some limitations should be noted for our analyses. First, we could not address missing geographic information for diagnosis/treatment facilities in the cancer registry. The proportion of persons missing ZIP code information for either diagnosis or treatment facility was higher among persons with ICC compared to persons with BC and CRC, as well as those with facilities in other states. Missing data and quality of facility information in cancer registries should be further explored. Second, we used primary payer at the time of diagnosis or treatment to distinguish between insurance groups, with the understanding that primary payer could be either at diagnosis or treatment. Prior studies on the primary payer variable in registries from other states have suggested there is mixed accuracy. ⁶⁵ , ⁶⁶ It is also unclear whether facilities in New Jersey consistently report insurance information from patient records. Patients may also switch insurance types between diagnosis and treatment, which may limit or expand their access to care. This limitation is particularly relevant to breast and cervical cancer, as federal legislation facilitates temporary Medicaid access when uninsured patients are diagnosed with these cancer types. ⁶⁷ , ⁶⁸ Third, we examined travel distance using straight‐line distances as opposed to driving distances on a network. However, a study by Boscoe et al. shows that straight‐line distances are highly correlated with distances based on street network. ⁶⁹ Fourth, we used established geographic boundaries for counties and HSAs in our analysis and did not have information to examine travel time by car or public transit. Future research should examine travel time, by car or public transit, to understand how different populations access cancer care. Fifth, we did not have access to information related to quality of care received at facilities within and outside of patients' geographic county and whether this quality of care affected patient outcomes. Future research should examine quality of care received when staying within a residential county versus going beyond this geography and potential associations with health outcomes using a resource such as American Hospital Association quality and patient safety data. ⁷⁰ Sixth, our analyses were conducted using data from 2012 to 2014. Future research is needed to understand geospatial patterns of cancer care within urban populations using more recent data. A final limitation is that our study did not have information on patient preferences or other structural barriers impacting decisions on where to obtain cancer care, such as structural racism and residential racial segregation. However, our observed differences for treatment location do suggest insurance‐based limitations, and similar factors should be examined in future work.

5. CONCLUSIONS

Our study suggests that urban populations are not homogenous in their geospatial patterns of cancer care utilization, as individuals living in areas with higher SVI scores may have more limited opportunities to access care outside of their immediate residential county than their counterparts living in areas with lower SVI scores. Efforts to build a culture of health and strengthen the integration of health services and systems (e.g., by reducing care fragmentation) may have a significant impact in urban communities with high proportions of NH Black residents, individuals with Medicaid and/or who are uninsured, and in areas with the highest levels of social vulnerabilities. Indeed, geographically tailored health systems (e.g., systems where the availability of and access to coordinated health care systems and providers are tailored to the needs of people in a geographic region) are needed to help improve equity in cancer care outcomes. Furthermore, cancer prevention and control research should continue to leverage existing data sources, such as cancer registries, to understand geospatial aspects of cancer care delivery and identify multilevel strategies to improve care quality and outcomes.

FUNDING INFORMATION

This study was funded by the Rutgers Cancer Institute of New Jersey (CINJ) Cancer Prevention and Control Pilot Award Program, including support from the Rutgers CINJ Cancer Center Support Grant (P30 CA072720). New Jersey State Cancer Registry is funded by the National Cancer Institute's Surveillance, Epidemiology and End Results (SEER) Program (#75N91021D00009) and Centers for Disease Control and Prevention's National Program of Cancer Registries (#5NU58DP006279) with additional support from the State of New Jersey and the Rutgers Cancer Institute of New Jersey. Dr. Richmond received support from the Agency for Healthcare Research and Quality (T32HS026122) and the National Cancer Institute (K99CA277366 and L60CA264691).

CONFLICT OF INTEREST STATEMENT

The authors declare no potential conflicts of interest.

Supporting information

Figure S1. Description of cohort inclusion criteria

Table S1. Average marginal effects of factors associated with treatment location among persons diagnosed with cancer in urban counties only, New Jersey State Cancer Registry, 2012–2014, N = 11,763

McGee‐Avila JK, Richmond J, Henry KA, Stroup AM, Tsui J. Disparities in geospatial patterns of cancer care within urban counties and structural inequities in access to oncology care. Health Serv Res. 2023;58(Suppl. 2):152‐164. doi: 10.1111/1475-6773.14182