The Role of Rurality, Travel Time, and Neighborhood Socioeconomics on Patterns of Adjuvant Therapy Receipt among Patients with Endometrial Cancer

Abstract

Background:

Rural patients with endometrial cancer are more likely to receive lower-quality treatment compared with their urban peers. We evaluated the role of contextual factors [rurality, distance to care, and community socioeconomics (SES)] on the receipt of adjuvant therapy (AT): vaginal brachytherapy (VBT), external beam radiation, and chemotherapy.

Methods:

We analyzed Surveillance Epidemiology and End Results–Medicare and included stages IB grade 3 and stages II to IV. We used county-level rural–urban continuum codes to define rurality, the Yost index to measure community SES, and measure of average driving time to gynecologic oncology care. Multivariable logistic regression was used to estimate adjusted ORs (aOR) and 95% confidence intervals (CI) evaluating AT receipt, adjusting for patient-level clinical and demographic characteristics.

Results:

A total of 7,572 individuals met inclusion criteria; 15% were rural residing. Rurality was only associated with lower odds of any AT receipt among patients with stage IB endometrial cancer (aOR = 0.62; 95% CI, 0.46–0.83). Increasing travel time was associated with lower odds of VBT (aOR = 0.89; 95% CI, 0.84–0.95). Residence in a low-SES neighborhood was associated with lower odds of chemotherapy (aOR = 0.79; 95% CI, 0.67–0.92) and VBT (aOR = 0.81; 95% CI, 0.69–0.95); however, associations were no longer significant after adjusting for individual SES.

Conclusions:

Travel time to gynecologic oncology care negatively affects the receipt of treatment regardless of rural or urban residence. Travel time may be a proxy for access to brachytherapy services and may explain the associations between travel and receipt of VBT.

Impact:

Factors characterizing the place of residence beyond rural/urban residence are important for predicting inequitable access to AT.

Introduction

Endometrial cancer is the most common gynecologic malignancy in the United States and the second most common cancer diagnosed among women who have not had a hysterectomy (1). It is estimated that in 2025, roughly 13,860 people will die as a result of this disease, and it is one of the few cancers for which both incidence and mortality continue to increase (2). Access to high-quality gynecologic oncology care is critical for optimizing outcomes for patients with endometrial cancer.

Adjuvant therapy (AT) is a central component of high-quality care for individuals with endometrial cancer and includes vaginal brachytherapy (VBT), external beam radiation (EBRT), and chemotherapy. AT is utilized across stages and grades with a high risk of recurrence (3). For patients with early-stage endometrial cancer, AT is beneficial for locoregional control of disease with a high risk of recurrence (4, 5). For patients with advanced disease, an AT regimen of chemotherapy and/or EBRT (with optional VBT) is recommended per National Comprehensive Cancer Network (NCCN) guidelines to reduce the risk of recurrence and improve survival (6, 7). Selection of AT is guided by prognostic factors such as age, grade, histology, and primary surgical treatment (3).

Because of the centralization of gynecologic oncology care, patients living in rural and remote areas face greater difficulties accessing care, particularly patients with disease stages and grade that need AT, because of the frequent trips required to receive care. EBRT, for example, requires daily treatments for 5 weeks, which can place tremendous burdens on patients and caregivers. Patients with endometrial cancer living in rural areas are less likely to receive timely, high-quality treatment compared with their urban peers (8–11). Rural residence is associated with lower odds of receiving lymph node assessment and minimally invasive hysterectomies and, among stages I and II, reduced likelihood of receiving adjuvant radiotherapy (8, 10, 12).

Rural location, in and of itself, is only one factor that may increase an individual’s risk of receiving delayed, poor-quality endometrial cancer care. Rural locations differ in their distance to urban areas, availability of care, and in community-level socioeconomics (SES), all of which have been shown to be associated with the receipt of AT and treatment outcomes in other cancer populations (13, 14). Rural areas in the United States are diverse, and using rural/urban status as the only measure to characterize a patient’s place of residence can result in overlooking variation in outcomes within as well as between rural and urban areas (15–17).

AT treatment patterns across rural/urban status and other contextual factors are not well characterized among the endometrial cancer population. Understanding the role of contextual factors in patterns of AT receipt is key to more precisely identifying populations at risk for low-quality care. The purpose of this study was to describe treatment patterns among rural patients with endometrial cancer and evaluate the associations between AT receipt and rurality, distance to gynecologic oncology care, and community SES.

Materials and Methods

Design and data

We used a retrospective cohort study design to conduct a secondary analysis of the Surveillance, Epidemiology and End Results linkage with Medicare claims (SEER–Medicare). SEER is comprised of cancer data from 18 cancer registries throughout the United States, including Kentucky, Iowa, New Mexico, Utah, and Georgia, and covers roughly 35% of the US population (18). Cancer registry information included in SEER is comprised of patient demographics, diagnosis information, cancer characteristics, and vital status. The rural population within the SEER registry, although proportionally smaller, is comparable to the US rural population overall (19). More than 97% of adults 65 years and older are enrolled in Medicare (18). When linked with Medicare claims data, SEER–Medicare connects incident primary endometrial cancer cases with data on diagnosis and treatment procedures in Medicare administrative data on enrollment and claims data (18). The Medicare administrative data we used in this study included beneficiary enrollment information and fee-for-service medical claims from parts A (inpatient) and B (outpatient; SEER–Medicare does not include individuals enrolled in Medicare Advantage). For these analyses, we used SEER data from 2006 to 2017 linked to Medicare claims fields from 2005 to 2018. This study was deemed exempt from full review by the Institutional Review Board at the University of North Carolina at Chapel Hill.

Sample selection

Our sample selection diagram is presented in Fig. 1. We included individuals in our analyses with stage IB grade 3 and stages II to IV (any grade) endometrial cancer diagnosed between 2006 and 2017 with complete staging information available in SEER. We derived the International Federation of Gynecology and Obstetrics stage from the tumor, lymph node, and metastasis staging information within the SEER database (3). According to current NCCN guidelines, these are all stages and grades in which AT is indicated; however, prior iterations of NCCN guidelines provided observation as an option for stages IB grade 3 and stage II grades 1 and 2. Individuals diagnosed at 66 years of age and older were included. We excluded patients who were diagnosed at autopsy or death, died within the same month as diagnosis, received a second cancer diagnosis within one year of their initial endometrial cancer diagnosis, or lacked continuous enrollment within the Medicare claims data. We also excluded patients with coverage under a health maintenance organization. Because we were interested in treatment patterns for patients receiving platinum-based therapies, we excluded those histologies for which non–platinum-based chemotherapy regimens were primarily indicated using the histology information in SEER.

Figure 1.

Patient selection diagram showing inclusion and exclusion of cases for analytic sample from SEER–Medicare.

Contextual factors

The contextual factors included to characterize patterns of AT included rurality, census tract socioeconomic index (SES), and travel time to the nearest gynecologic oncologist. These variables were selected by reviewing studies that examined associations between contextual factors and endometrial cancer treatment (15, 16, 20). Rurality was measured at the county level using the 2013 Rural–Urban Continuum Codes (RUCC) developed by the US Department of Agriculture (https://www.ers.usda.gov/data-products/rural-urban-continuum-codes/). RUCCs use the degree of urbanization and proximity to metro areas to divide counties into nine classifications. Urban counties were defined as counties with RUCC values 1 through 3, and rural counties were defined as those with RUCC values of 4 through 9.

The Yost index is provided in SEER–Medicare as a measure of community SES. This index was constructed using a factor analysis of variables from the American Community Survey including median household income, median house value, median rent, percent below 150% of the poverty line, education index, percent working class, and percent unemployed (21). We used the state-based SES quintile that reflects the quintile of an individual’s census tract relative to the census tracts within their state. The first quintile reflects census tracts with the lowest SES, whereas the fifth quintile reflects the highest SES tracts relative to the rest of the tracts within the state. We excluded individuals missing a value for the Yost index for any of the years of data available (Fig. 1). For analysis, the quintiles were collapsed into three categories: quintile 1 was considered “low SES”, quintiles 2 and 3 “mid SES”, and quintiles 4 and 5 “high SES.”

To obtain a measure of availability of gynecologic cancer care, we generated a measure of the average countywide driving time to the nearest gynecologic oncologist for the continental United States. Using Google Maps, we geocoded the addresses of gynecologic oncologists as provided by the Society for Gynecologic Oncology database. Within each county, we geocoded 100 points, randomly scattered in the county, and calculated the driving time in minutes from each point to the nearest gynecologic oncologist using the Open Source Routing Machine (22). We used the travel time points to interpolate a surface using R software, generating a raster file of driving times within the continental United States. We calculated county means from the raster file and used the mean travel time in hours to the nearest gynecologic oncologist as a proxy of availability of gynecologic care.

Patient-level factors

Patient-level factors identified from SEER information included age at diagnosis, year of diagnosis, race, ethnicity, Medicaid eligibility, comorbidities, and tumor stage. We used Medicaid eligibility as a proxy for individual SES as individual income information is not available in SEER–Medicare. Using all years of beneficiary enrollment available, we included an indicator of Medicaid eligibility in analyses if, at any point during the enrollment years available, an individual was eligible for Medicaid. We used eligibility rather than dual enrollment at diagnosis to capture individuals who (i) may experience financial instability and (ii) may have been eligible but for administrative reasons (e.g., due to Medicaid redetermination) were not enrolled at the time of diagnosis (23, 24).

Surgical modality (minimally invasive hysterectomy/total abdominal hysterectomy/other/no hysterectomy) and lymph node assessment were measured using diagnosis and procedure codes in Medicare claims after diagnosis. The Charlson Comorbidity Index was generated using diagnosis and procedure codes from the first year of enrollment in Medicare (25). The final weighted Charlson Comorbidity Index score excluded any malignancy or solid tumor and was collapsed into three categories: 0, 1, and ≥2.

Treatment outcomes

We identified the receipt of AT by evaluating the receipt of VBT, EBRT, and chemotherapy using the Current Procedural Terminology (CPT) and International Classification of Diseases (ICD) diagnosis and procedure codes from Medicare claims within 9 months after diagnosis as has been documented in prior studies (Supplementary Table S1; refs. 12, 26). An individual was considered to have received VBT or EBRT if any claims were detected within 9 months after diagnosis. For chemotherapy receipt, an individual was considered to have received chemotherapy if any claims codes for any of the chemotherapies used in the platinum-based regimens as recommended by the NCCN guidelines were detected 9 months after diagnosis. The Healthcare Common Procedure Coding System (HCPCS) codes for chemotherapies were verified using the NCI Observational Research in Oncology Toolbox (Supplementary Table S1; https://seer.cancer.gov/oncologytoolbox/canmed/).

Statistical analyses

Descriptive statistics were calculated to summarize the distributions of contextual and individual-level variables for the sample overall and by the receipt of AT. We used logistic regressions to estimate ORs and their 95% confidence intervals (CI) for the associations of the contextual factors (rurality, travel time, and census tract SES) with the receipt of any AT, VBT, EBRT, and chemotherapy (treatment outcomes). First, we estimated the association of each contextual factor with each treatment modality while controlling for clinical characteristics–adjusted OR1 (aOR1): age, stage, grade, histology, surgical treatment (hysterectomy with or without lymph node assessment), and year of diagnosis. Second, we examined the associations between each contextual factor on AT receipt while controlling for clinical characteristics and the remaining individual-level, demographic factors (aOR2). Interactions among rurality, travel time, and community SES were tested using likelihood ratio tests to examine within rural/urban associations of travel time and community SES with treatment outcomes. We conducted these analyses for the overall sample and in analyses stratified by stage.

All tests were two-sided at α = 0.05, with P < 0.05 considered statistically significant. Data management was conducted using SAS (SAS Inc.) and R software (version 3.6.3). Logistic regressions and figures were generated using R software.

Data availability

SEER–Medicare data are not available from the investigator. Original data used to derive the analytic cohort can be accessed through application to the US NCI at https://healthcaredelivery.cancer.gov/seermedicare/.

Results

A total of 7,572 individuals diagnosed with endometrial cancer met inclusion criteria. Urban patients comprised 85% of the total sample (Table 1). Within the overall sample, 64% lived in counties where the average travel time to the nearest gynecologic oncologist was within 1 hour. The majority of rural patients (47%) lived in counties where the average travel time was between 1 and 2 hours and a quarter lived in counties with an average travel time between 2 and 3 hours. The majority of urban patients (74%) lived in counties with an average travel time under 1 hour.

Table 1.

Contextual and sociodemographic characteristics of patients diagnosed with stage IB grade IV endometrial cancer from 2006 to 2017 in the SEER–Medicare database by rural/urban residence (N = 7,572).

Characteristic	Overall N (%)	Urban N (%)	Rural N (%)
Rurality
Urban	6,433 (85.0)	–	–
Rural	1,139 (15.0)	–	–
Travel time (hours)a
<1	4,875 (64.4)	4,731 (73.5)	144 (12.6)
1–2	1,763 (23.3)	1,233 (19.2)	530 (46.5)
2–3	695 (9.2)	404 (6.3)	291 (25.6)
3+	239 (3.1)	65 (1.0)	174 (15.3)
Census tract SES quintile
High SES	3,269 (43.2)	3,114 (48.4)	155 (13.6)
Mid SES	3,124 (41.3)	2,445 (38.0)	679 (69.6)
Low SES	1,179 (15.5)	874 (13.6)	305 (26.8)
Age, years [median, (IQR)]	74 (69–79)	74 (69–80)	74 (69–80)
Race
White	6,409 (84.6)	5,384 (83.7)	1,025 (90.0)
Black	826 (10.9)	732 (11.4)	94 (8.3)
Other	337 (4.5)	317 (4.9)	20 (1.7)
Non-Hispanic	7,109 (93.9)	6,001 (93.3)	1,108 (97.3)
Medicaid eligibleb	1,648 (21.0)	1,353 (21.5)	295 (25.9)
Year of diagnosisc
2006–2009	2,479 (32.7)	2,094 (32.6)	385 (33.8)
2010–2013	2,630 (34.7)	2,257 (35.1)	373 (32.7)
2014–2017	2,463 (32.5)	2,082 (32.4)	381 (33.5)
Stage
IB	1,546 (21.0)	1,297 (20.2)	249 (21.9)
II	1,592 (21.4)	1,357 (21.1)	235 (20.6)
III	2,918 (37.6)	2,476 (38.5)	442 (38.8)
IV	1,516 (19.9)	1,303 (20.2)	213 (18.7)
Histology and grade
Endometrioid
Low grade	1,824 (32.6)	1,564 (32.8)	260 (31.2)
High grade	2,840 (50.7)	2,400 (50.3)	440 (52.9)
Unknown grade	936 (16.7)	804 (16.9)	132 (15.9)
Nonendometrioid
Low grade	126 (6.4)	100 (6.0)	26 (8.5)
High grade	1,471 (74.6)	1,249 (75.0)	222 (72.3)
Unknown grade	375 (19.0)	316 (19.0)	59 (19.2)
Comorbidities
0	3,456 (46.0)	2,914 (45.3)	542 (47.6)
1	1,988 (26.3)	1,670 (26.0)	318 (27.9)
2+	2,128 (27.8)	1,849 (28.7)	279 (24.5)
Hysterectomy
Total	5,759 (76)	4,882 (76)	1,018 (77)
Minimally invasived	3,017 (52.5)	>2,553 (>52.3)	>442 (>43.4)
Total abdominald	2,731 (47.3)	2,308 (47.3)	423 (41.6)
Unspecifieda	11 (0.2)	<11 (0.4)	<11 (<1.0)
Lymph node assessmentd	2,775 (48.2)	2,398 (49.1)	377 (43.0)

^aDetermined from Medicare enrollment, any eligibility for Medicaid in all years of enrollment data available.

^bCountywide average travel time to the nearest gynecologic oncologist.

^cYears of diagnosis groupings were determined by the years of release revisions for NCCN guidelines for endometrial cancer treatment during the study period (2009–2014).

^dProportions reflective only among those who received hysterectomy.

In our sample, 16% of patients lived in the lowest, most socioeconomically disadvantaged census tracts. More rural than urban patients lived in the most socioeconomically disadvantaged census tracts in the first SES quintile (27% vs. 14%). Among the overall sample, 43% patients resided in a high-SES community; nearly half of patients in urban areas resided in a high-SES community, whereas only 14% of rural patients did.

A higher proportion of patients living in rural areas compared with urban areas were White (90% vs. 84%), non-Hispanic (97% vs. 93%), and Medicaid eligible (26% vs. 22%; Table 1). The proportions of the type of hysterectomy received were similar across rural and urban groups (Table 1). A lower proportion of patients in rural areas received lymph node assessment compared with patients from urban areas (43% vs. 49%).

Logistic regression results from analyses conducted on the overall sample are presented in Table 2. Figures 2–4 highlight notable findings from analyses stratified by stage. Full results for the stage-stratified analyses and associations between demographic characteristics and receipt of AT modalities are reported in Supplementary Tables S2–S6. We performed likelihood ratio tests and found no statistically significant interactions between rurality and travel time or census tract SES for any of the AT modalities.

Table 2.

Adjuvant treatment outcomes and associations with contextual factors.

	Received		Did not receive		OR and 95% CI
	n	(%)	n	(%)	aORa (95% CI)	aORb (95% CI)
Any AT
Rurality
Urban	4,710	(72.0)	1,834	(28.0)	Ref	Ref
Rural	835	(70.5)	350	(29.5)	0.91 (0.79–1.05)	0.92 (0.80–1.06)
Travel time (hours)
Median (IQR)	0.66	(0.40–1.34)	0.72	(0.40–1.45)	0.95 (0.88–0.99)	0.94 (0.89–1.00)
Community SES
High SES	2,428	(73.0)	898	(27.0)	Ref	Ref
Mid SES	2,265	(71.2)	916	(28.8)	0.90 (0.80–1.01)	0.98 (0.87–1.10)
Low SES	852	(69.7)	370	(30.3)	0.85 (0.73–0.99)	1.00 (0.85–1.18)
VBT
Rurality
Urban	2,052	(31.9)	4,381	(68.1)	Ref	Ref
Rural	361	(31.7)	778	(68.3)	0.96 (0.83–1.11)	0.96 (0.83–1.11)
Travel time (hours)
Median (IQR)	0.61	(0.37–1.22)	0.67	(0.40–1.39)	0.89 (0.84–0.95)	0.89 (0.84–0.95)
Community SES
High SES	1,064	(32.7)	2,205	(67.3)	Ref	Ref
Mid SES	1,017	(32.8)	2,107	(67.2)	1.00 (0.90–1.12)	1.04 (0.94–1.18)
Low SES	332	(28.2)	847	(71.8)	0.81 (0.69–0.95)	0.91 (0.77–1.07)
EBRT
Rurality
Urban	3,193	(42.2)	4,369	(57.8)	Ref	Ref
Rural	573	(43.4)	747	(56.6)	1.00 (0.88–1.14)	1.01 (0.88–1.15)
Travel time (hours)
Median (IQR)	0.67	(0.40–1.39)	0.64	(0.40–1.22)	1.03 (0.97–1.09)	1.03 (0.97–1.09)
Community SES
High SES	1,626	(58.7)	2,308	(41.3)	Ref	Ref
Mid SES	1,512	(57.0)	2,008	(43.0)	1.08 (0.97–1.19)	1.09 (0.98–1.21)
Low SES	628	(56.0)	800	(44.0)	1.09 (0.95–1.25)	1.13 (0.97–1.31)
Chemotherapy
Rurality
Urban	2,914	(45.3)	3,519	(54.7)	Ref	Ref
Rural	499	(43.8)	640	(56.2)	0.89 (0.77–1.03)	0.90 (0.78–1.05)
Travel time (hours)
Median (IQR)	0.64	(0.39–1.22)	0.67	(0.40–1.44)	0.89 (0.84–0.95)	0.89 (0.84–0.95)
Community SES
High SES	1,547	(47.3)	1,722	(52.7)	Ref	Ref
Mid SES	1,362	(43.6)	1,762	(56.4)	0.81 (0.72–0.90)	0.87 (0.78–0.98)
Low SES	504	(42.7)	675	(57.3)	0.79 (0.67–0.92)	0.94 (0.79–1.11)

Abbreviation: Ref, reference.

^aLogistic regression models controlling for clinical features: age, stage, grade, histology, surgical treatment, and year of diagnosis.

^bLogistic regression models controlling for clinical features and additional individual level characteristics: race, ethnicity, comorbidities, and Medicaid eligibility.

Figure 2.

Forest plot diagram demonstrating associations between rurality, travel time, and receipt of any AT for patients with stage IB endometrial cancer. Models evaluating the association of rurality and travel time on AT receipt controlled for individual-level disease characteristics and demographic factors.

Figure 4.

Forest plot diagram demonstrating associations between travel time, community SES, and chemotherapy receipt for patients with stage III and IV endometrial cancer. Models evaluating the association of travel time and community SES on chemotherapy controlled for individual-level disease characteristics.

Any AT

In the overall sample, travel time and living in a low-SES community were associated with lower odds of AT receipt when controlling for clinical characteristics. However, when controlling for both demographic and clinical characteristics, we did not observe statistically significant associations between rurality, travel time, or census tract SES on the receipt of any AT. In analyses stratified by stage, rurality was associated with lower odds of receipt of AT (aOR2 = 0.62; 95% CI, 0.46–0.83) as well as increased travel time (aOR2 = 0.86; 95% CI, 0.75–0.98) for patients with stage IB disease (Fig. 2). No associations were observed between contextual factors and AT receipt for stages II, III, or IV.

VBT

In the overall sample, increasing travel time was associated with lower odds of receipt of VBT (aOR2 = 0.89; 95% CI, 0.84–0.95). In analyses adjusted for clinical factors, living in a low-SES community was associated with lower odds of VBT receipt; however, the relationship was no longer significant when controlling for demographic factors, including Medicaid eligibility. We observed no associations between rurality and VBT receipt.

Among both stage IB and II patients, travel time was associated with lower odds of VBT receipt in analyses when adjusting for clinical and demographic characteristics (aOR2 = 0.75; 95% CI, 0.58–0.97 and aOR2 = 0.84; 95% CI, 0.74–0.94; Fig. 3). Living in a low-SES community was associated with lower odds of VBT receipt for stage II and stage IV patients when only controlling for clinical characteristics, but we observed no statistically significant associations between community SES with VBT receipt when also controlling for demographics. No statistically significant associations were observed between the contextual factors and VBT receipt for patients with stage III disease.

Figure 3.

Forest plot diagram demonstrating associations between rurality, travel time, and VBT receipt for patients with stage IB and II endometrial cancer. Models evaluating the association of travel time on brachytherapy receipt controlled for individual-level disease characteristics and demographic factors.

EBRT

We observed no statistically significant associations between rurality, travel time, and EBRT receipt in the overall sample (Table 2) or the analyses stratified by stage (Supplementary Tables S2–S5).

Chemotherapy

In the overall sample, we observed a negative association of living in a low- and mid-SES community compared with living in a high-SES community with chemotherapy receipt (OR1 = 0.79; 95% CI, 0.67–0.92 and OR1 = 0.81; 95% CI, 0.72–0.90). The associations of community SES were no longer statistically significant when adjusting for additional demographic characteristics. Increased travel time was associated with lower odds of chemotherapy receipt (aOR2 = 0.89; 95% CI, 0.84–0.95). We observed no associations between rurality and chemotherapy receipt, both in the overall sample and in the analyses stratified by stage.

In the analyses stratified by stage, travel time was significantly associated with lower odds of chemotherapy receipt only among patients with stage III disease (aOR2 = 0.85; 95% CI, 0.78–0.93; Fig. 4). For patients with stage III disease, living in a low- or mid-SES community was associated with lower odds of chemotherapy receipt (aOR1 = 0.71; 95% CI, 0.56–0.91 and aOR1 = 0.79; 95% CI, 0.66–0.94; Fig. 4). However, similar to the analyses in the overall sample, this association was no longer statistically significant when controlling for demographic characteristics.

Clinical and sociodemographic characteristics

Associations between clinical and sociodemographic characteristics and receipt of each AT modality from the fully adjusted models are reported in Supplementary Table S6. Older age was associated with lower odds of receipt of VBT, EBRT, and chemotherapy. We observed a notable, significant relationship between Medicaid eligibility and AT receipt. Patients who had any months of Medicaid eligibility while enrolled in Medicare had lower odds of receipt of any AT, VBT, and chemotherapy.

Discussion

In this study, using a large, US population-based cohort, we observed that geographic disparities in the receipt of AT are more evident in patients with early-stage endometrial cancer. The most critical finding is that for both patients with stage IB and stage II disease, living in a county with longer average travel times to gynecologic oncology care was associated with lower odds of receipt of VBT.

Distance to care has differential and sometimes paradoxical effects on rural cancer access to care (27–29). Our analyses of countywide average travel time indicate that low availability of gynecologic oncology care may negatively affect the receipt of brachytherapy but not chemotherapy and EBRT. Our findings related to chemotherapy receipt are consistent with the existing literature in gynecologic oncology outcomes (28). However, this relationship may vary by cancer type as increased distance to care is associated with decreased likelihoods of receiving chemotherapy in other cancer populations (13). Additionally, although we found that increased distance to care was associated with lower odds of brachytherapy, there was no statistically significant interaction between distance and rurality—indicating that regardless of rural or urban residence, patients who lived further from gynecologic oncology care are less likely to receive VBT.

These results have two implications. First, distance to care likely does not have a uniform effect on treatment receipt across cancer populations, particularly for those residing in rural areas. Regardless, increasing travel distance is associated with increased financial and logistic burdens on patients, particularly rural patients who may be more likely to be socioeconomically vulnerable; thus, patients traveling far distances for treatment need access to supportive services to either help facilitate transportation to care to ensure the receipt of high-quality treatment or resources to help offset the costs of transportation. Our study contributes to the literature by providing evidence that travel time to gynecologic oncology care is associated with lower odds of receipt of VBT. Additionally, 46% of patients living in rural counties in this sample resided in counties with an average travel time to the nearest gynecologic oncologist of 1 to 2 hours and 41% lived in counties with an average driving time of more than 2 hours. We did not observe significant differences or interactions between travel time, rurality, and EBRT or chemotherapy receipt. Although this lack of disparity is positive, it does not capture the material burdens that may result from accessing AT.

The second implication is that although the association between travel time and VBT receipt but not EBRT receipt may seem inconsistent, these findings indicate a need to better understand delivery patterns and accessibility of brachytherapy services. In this study, countywide average travel time to the nearest gynecologic oncologist may overlap with access to VBT services given the centralization of gynecologic oncology care at high-volume facilities. Additionally, brachytherapy services require specialized training and equipment, and the availability of those services has been decreasing over the years (30–32). The reason for the reduction in brachytherapy services may be due to less training available for radiation oncology residents compared with training for EBRT or lower insurance reimbursement for brachytherapy services (30, 33).

Despite its critical role in the treatment of various cancers, including endometrial cancer, it is estimated that 23.3% of the US population does not have access to brachytherapy services and that almost half of radiation oncology residents do not feel confident establishing a brachytherapy practice (33, 34). These factors, combined with a shrinking radiation oncology workforce while cancer rates increase, could indicate that brachytherapy services may be less available due to the cost, training, and technology requirements needed to maintain a brachytherapy service (31, 32, 34, 35). Given the significance of VBT in endometrial cancer treatment and the benefits of VBT therapy compared with EBRT (e.g., improved quality of life and fewer adverse effects), facilitating equitable access to brachytherapy services is crucial (3). Future work should characterize the geographic distribution of radiation oncologists providing brachytherapy services with gynecologic oncologists to better understand the relation between differential access to these specialty providers and patient outcomes.

We found limited evidence of associations between community SES measured by the Yost index and receipt of AT. Where we observed associations between community SES and AT receipt, we saw that living in a lower-SES community was associated with lower odds of receipt of AT. This is consistent with other studies in endometrial cancer and other gynecologic oncology patient populations (36–41). However, those effects were no longer significant when accounting for Medicaid eligibility, a proxy for individual SES. We likely observed this pattern because individual and community SES are highly correlated. Although other work has shown distinct effects of individual SES compared with community level on health outcomes (42), we may not have observed the same effects in this study because we examined healthcare utilization, which may be determined more by factors like insurance and individual financial resources rather than exposure to the environment of a low-SES community. It may be that once an individual receives a cancer diagnosis and is connected with a gynecologic oncologist, community-based socioeconomic barriers to care that are correlated with low cancer screening and late-stage cancer diagnoses (notably observed in breast and cervical cancers; ref. 43) may play less of a role in disparities in access to care than individual SES indicators. It is still critical to separate individual and community SES as area-level measures of SES have been shown to be poor proxies for individual SES (44, 45). Investigating the differential roles of community SES, access to social capital, and individual SES on AT receipt in patients with endometrial cancer to disentangle these relationships may provide more clarity to the associations observed in this study.

Although the focus of this study was on contextual factors and AT receipt, a notable finding was the association between Medicaid eligibility and decreased odds of receipt of VBT and chemotherapy receipt. To our knowledge, prior studies using SEER–Medicare to evaluate endometrial cancer care patterns have either only used the indicator within the SEER of Medicaid coverage at diagnosis or not included Medicaid coverage or eligibility as a predictor in their analyses. Medicaid eligibility may better approximate individuals in more precarious financial situations and with limited access to care as people may not present at diagnosis with adequate insurance coverage. Additional research is needed to examine the role of Medicaid eligibility in endometrial cancer outcomes, trends in Medicaid eligibility and coverage for patients, and the impact of policies, such as Medicaid redetermination or reimbursement disparities, on access to endometrial cancer treatment (46).

Limitations

There are a few limitations of using the SEER–Medicare dataset. SEER–Medicare only contains claims and cancer registry data for individuals 65 years of age and older and those with fee-for-service Medicare. Thus, our results may not be generalizable to younger patients with endometrial cancer, private or uninsured patients, or the increasing proportion of rural- and urban-residing patients in the United States who have opted in to Medicare advantage (47, 48). Additionally, rural populations are underrepresented in SEER–Medicare: roughly 20% of the US population resides in rural areas compared with 15% of our sample in this study (47). We did not include variables indicating where someone received care or the specialty of their provider, which are associated with improved outcomes among patients with endometrial cancer, or other factors such as referral patterns and patient preference that may contribute to unmeasured confounding (12). However, our focus was primarily on the impact of where patients resided on the receipt of AT, and our findings provide evidence for future work to examine specific factors that put individuals living in communities far from care or with low SES at risk for potential inequities in care. Additionally, we measured utilization of services, but utilization is not always synonymous with access. Other factors such as affordability and care quality may contribute to a more holistic understanding of access to care (13, 20).

The most granular geographic identifier available for this analysis was a county identifier and thus our measure of distance to care is neither the most accurate estimate of individual travel time to care nor does it capture individual-level barriers to care such as access to transportation. Yet, our method of calculating countywide average travel time to gynecologic oncology care is an innovative and more robust measure of care availability compared with straight-line distance to care using county centroid or other existing methods (22, 49). Our use of a county-level identifier of rurality (RUCC) may have obscured more granular variations in between rural and urban areas (50). We also used the Yost index to measure neighborhood SES at the census tract level (21). Although census tracts are considered to be an acceptable geographic level to approximate neighborhood SES, it is a fixed boundary unit and does not necessarily reflect all the local contexts with which an individual interacts (14).

This study has many strengths. We utilized a robust population-based dataset of cancer registry data matched with Medicare claims to evaluate the impact of characteristics of where an individual lives on AT receipt among patients with endometrial cancer. Using medical claims allowed us to determine AT receipt among our sample, comorbidity history, and surgical treatment after diagnosis. The use of Medicare enrollment data allowed us to capture Medicaid eligibility as proxy for individual socioeconomic status.

Conclusion

This study contributes to the current limited evidence on the association of contextual factors, in particular travel time to care and census tract SES, with the receipt of AT. Our findings suggest that characteristics of an individual’s place of residence beyond rural/urban status are important indicators for predicting inequitable access to AT, particularly VBT and among patients with earlier-stage disease. This will inform future work on understanding and improving access to AT for patients with endometrial cancer.

Supplementary Material

Supplementary Table 1

Supplementary Table 1 provides the ICD and HCPCS codes used to identify treatment variables and outcomes in Medicare claims files.

Supplementary Table 2

Supplementary Table 2 shows the odds ratios and confidence intervals for associations between contextual factors and adjuvant therapy outcomes for patients with Stage IB grade 3 endometrial cancer.

Supplementary Table 3

Supplementary Table 3 shows the odds ratios and confidence intervals for associations between contextual factors and adjuvant therapy outcomes for patients with Stage II endometrial cancer.

Supplementary Table 4

Supplementary Table 4 shows the odds ratios and confidence intervals for associations between contextual factors and adjuvant therapy outcomes for patients with Stage III endometrial cancer.

Supplementary Table 5

Supplementary Table 5 shows the odds ratios and confidence intervals for associations between contextual factors and adjuvant therapy outcomes for patients with Stage IV endometrial cancer.

Supplementary Table 6

Supplementary Table 6 shows the odds ratios and confidence intervals for associations between patient-level factors and adjuvant therapy outcomes among the whole sample.

Authors’ Disclosures

V.M. Petermann reports grants from Rita and Alex Hillman Foundation and NIH during the conduct of the study. J.L. Lund reports employment of spouse with GSK and ownership of GSK stock. No disclosures were reported by the other authors.

Authors’ Contributions

V.M. Petermann: Conceptualization, formal analysis, funding acquisition, methodology, writing–original draft, project administration. S.B. Wheeler: Conceptualization, supervision, validation, methodology, writing–review and editing. J.L. Lund: Conceptualization, resources, supervision, methodology, writing–review and editing. A.L. Bryant: Supervision, methodology, writing–review and editing. B.E. Jackson: Resources, data curation, supervision, methodology, writing–review and editing. B.B. Albright: Supervision, writing–review and editing. T.J. Worm: Data curation, methodology, writing–review and editing. J. Leeman: Conceptualization, supervision, writing–review and editing.