Urban versus Rural Residence and Outcomes in Older Patients with Breast Cancer

Overall, individuals living in rural counties experience higher annual age-adjusted cancer-related mortality rates when compared with those residing in urban areas (180 vs. 158 deaths/100,000/year; ref. 1). In contrast to lung, prostate, and colon cancer, the impact of rurality on age-adjusted mortality rates for patients with breast cancer varies based on the level of metro or rural classification (1, 2). Current evidence suggests that the gap in cancer-related mortality rates associated with urban and rural residence is widening over time (1, 3). The outcome disparity may be in part explained by lack of access to healthcare resources in the rural areas (4). While 20% of the U.S. population resides in rural areas, only 3% of medical oncologists have rural practices (5, 6). Similarly, nononcology specialists tend to practice in urban areas (134 vs. 40 specialists/10,000; ref. 6). These workforce issues suggest that rural patients with cancer are likely to face challenges in accessing adequate cancer care, and this could explain the differences in outcome by rurality (7). Low socioeconomic status (SES) can further compound these outcome disparities (3). Differences in economic and demographic characteristics between rural and urban counties largely explain the disparity in age-adjusted mortality in the general population (3). However, there has been no previous systematic attempt to understand the impact of rural residence and SES on overall survival (OS) in older patients with breast cancer. Furthermore, the impact of healthcare utilization (HCU) in mediating the disparities in outcome due to residence (rural vs. urban) or SES is not known. We address this knowledge gap in a population-based sample of patients with breast cancer diagnosed over the age of 65, using Surveillance Epidemiology and End Results (SEER) data linked to Medicare administrative claims data.

SEER is comprised of 20 geographically diverse cancer registries, designed to represent U.S. cancer incidence and mortality. SEER data were linked to Medicare administrative claims, providing billing information for diagnosis codes from inpatient, outpatient, and noninstitutional settings (8). This study was deemed exempt by University of Alabama at Birmingham's Institutional Review Board (Birmingham, AL).

Females >65 years of age and enrolled in Medicare with incident breast cancer diagnosed between January 1, 2000 and December 31, 2011 were included. Medicare claims were captured 1 year prior to diagnosis to establish preexisting comorbidities and up to December 31, 2013 or death, whichever occurred first. We excluded those diagnosed with no information regarding stage of disease. Individuals enrolled in Medicare Advantage or those lacking Medicare coverage for inpatient care or doctors' services or outpatient care at any point during the eligible period were excluded because health services could not be captured in their entirety for such patients (n = 109,608; Supplementary Fig. S1).

OS was examined from the end of primary breast cancer treatment (not including hormone therapy) to death, 5 years, or end of study (December 31, 2013). End of treatment was used as a starting point to mitigate immortal time bias introduced from patients who survived long enough to receive treatment. Healthcare visits to primary care physicians (PCP; internal medicine, family medicine, geriatrician, and primary care), oncologists (hematology/oncology, radiation oncology, and surgical oncology), and nononcology specialists, as well as hospitalizations and emergency department (ED) visits were captured from end of treatment to 1 month prior to death or end of study. Physician specialties were identified using unique physician identification number and the national provider identifier codes linked to the American Medical Association Masterfile (9). HCU was described for two time periods. The first time period captured from end of treatment to end of year 1, 1 month prior to death, or end of study. The second time period captured from the beginning of year 2 to the end of year 5, 1 month prior to death, or end of study. We hypothesize that the first-year time frame may have increased HCU due to proximity to diagnosis and treatment and year 2 to year 5 may be more representative of long-term utilization trends.

Rural versus urban residence was defined using the Rural/Urban Continuum Codes (10). “Urban” included all metropolitan counties with ≥250,000 people and urban counties (>20,000) adjacent to metropolitan counties. “Rural” included urban counties not adjacent to metropolitan counties, completely rural counties, or urban counties with <2,500 people. SES was defined using the proportion of population in the census tract below Federal Poverty Level (FPL) based on residence at diagnosis (11). The cohort was categorized into low-poverty (0%–10% below FPL), medium-poverty (10%–20%), and high-poverty (20%–100%) groups (12, 13). A composite measure of rural/urban residence and SES was also created, where patients were categorized as urban/low poverty, urban/medium poverty, urban/high poverty, rural/low poverty, rural/medium poverty, and rural/high poverty.

Age at diagnosis was categorized as 66–70 years, 71–75 years, 76–80 years, and 80+ years. Stage at diagnosis was determined using the SEER summary staging and the 6th American Joint Committee on Cancer staging. Race/ethnicity was categorized as non-Hispanic white (NHW), Hispanic, non-Hispanic black (NHB), and Asian American/Pacific Islander. Comorbidity was measured using the Elixhauser Comorbidity index (14) in the 12 months prior to breast cancer diagnosis. Chemotherapy and radiation details were drawn from outpatient claims, physician carrier claims, inpatient claims, durable medical equipment Medicare claims, and from Part D data (chemotherapy only for 2007–2013; Supplementary Table S1). Patients were categorized as having received chemotherapy (yes/no) and having received radiation (yes/no). Surgical intervention was identified from inpatient claims and SEER file. Patients were categorized as having received no surgery, partial mastectomy, and total mastectomy.

OS was examined using Kaplan–Meier method and proportional subdistribution hazards model for multivariate analysis. OS was examined by urban versus rural residence, by SES categories (high, medium, and low poverty), and by the combined rural–urban/SES variable.

The relative rate of healthcare visits per person-time [incidence rate ratios (IRR)] was examined for PCP visits, oncology visits, nononcology specialist visits, hospitalizations, and ED visits. IRRs were modeled for visits during year 1 and years 2–5. Covariates included age at breast cancer diagnosis, race/ethnicity, preexisting comorbidities, stage at diagnosis, surgery, chemotherapy, and radiation. For each model, count-based models (Poisson, zero-inflated Poisson, negative binomial, and zero-inflated negative binomial) were examined to determine the best fitting model [overdispersion factors, Akaike information criterion (AIC), and Bayesian information criterion (BIC)]. Analyses were repeated including any subsequent malignant neoplasms (SMN) diagnosed during years 2–5 as a competing event.

We hypothesized that physician visits, as a proxy for access to healthcare, may either act as a mediator or effect modifier of the association between the residence/SES status and OS. To evaluate this, we estimated mediation and interaction simultaneously using effect decomposition (15, 16). “Exposure” was defined as the residence/SES combined measure; we made a specific two-category comparison: urban/low poverty versus rural/high poverty. A priori, the mediator and/or moderators of interest were healthcare visits for the three physician visit types (PCP, nononcology specialist, and oncology specialist), creating dichotomous variables that indicated whether the patient had at least one visit per year versus no visit. We examined HCU spanning from end of treatment to 1 month prior to death to avoid survival bias due to treatment and capturing end of life visits. We stratified the analyses by Elixhauser comorbidity score (0, 1, and ≥2) and by stage of breast cancer (stages 0–III and stage IV), adjusting for age at diagnosis, race/ethnicity, and treatment. We hypothesized that ≥1 visit/year with a nononcology specialist would mediate and/or moderate the association between residence/SES and mortality in the strata with 0 and 1 comorbidities. Furthermore, ≥2 visits/year with a nononcology specialist would mediate and/or moderate the association between residence/SES and mortality in the strata with ≥2 comorbidities. Final model selections were informed by the iterative process of constructing the model for each mediation (i.e., associations between residence/SES and healthcare visits, and associations between healthcare visits and mortality). We specified an accelerated failure time model with Weibull distribution since the event is nonrare by the end of follow-up. Results from this analysis allowed us to examine excess relative risk of mortality attributed to HCU. Supplementary Data S1 describes the decomposition of excess risk into controlled direct effects, interaction only effects, mediation only effects, and mediated interaction effects.

Table 1 describes patient characteristics overall, and by the residence/SES combined measure. Of the 109,608 patients with breast cancer included in this cohort, 12.5% were rural dwellers, 15.6% resided in high-poverty areas, 86% were NHW, and 7% were NHB. The median age at diagnosis of breast cancer was 76 years (66–106) and the majority (56%) were diagnosed with stage I–II disease. Supplementary Table S1 describes patient characteristics by rural/urban status and SES separately.

There was an overrepresentation of rural dwellers in the high-poverty group (30.7% vs. 13.5%; P < 0.001) and among NHW (91.4% vs. 85.3%; P < 0.001) patients. Rural dwellers were less likely to have stage 0 disease (13.7% vs. 15.6%; P < 0.001); all other stages were comparable. Rural dwellers were also more likely to undergo total mastectomy (47.6% vs. 34.1%; P < 0.001) and to receive chemotherapy (49% vs. 45.4%; P < 0.001).

High-poverty group patients were more likely to be diagnosed with stage III (26.1% vs. 21.7%; P < 0.001) or stage IV (7.3% vs. 5.2%; P < 0.001) disease. High-poverty group patients were also more likely to receive chemotherapy (50.2% vs. 44.1%; P < 0.001) and more likely to undergo total mastectomy (42.9% vs. 32.3%; P < 0.001). Furthermore, the high-poverty group patients were also more likely to have multiple (2+) comorbidities (27.4% vs. 19.1%; P < 0.001; Supplementary Table S1).

Rural/high-poverty patients were more likely to be diagnosed at younger ages (66–70 years: 29% vs. 25%–27%; P < 0.001), receive total mastectomies (50% vs. 31%–47%; P < 0.001), and receive chemotherapy (54.3% vs. 43.1%–49.2%; P < 0.001) compared with all other groups. Urban/high-poverty patients were more likely to be NHB (27% vs. 1%–10%; P < 0.001) and to have two or more comorbidities (28% vs. 17%–25%; P < 0.001) compared with all other groups (Table 1).

Five-year OS (5 y OS) was 71% for the urban dwellers versus 69% for the rural dwellers (P < 0.001). The 5 y OS was 73% for low-poverty, 69% for the medium-poverty, and 63% for the high-poverty group (P < 0.001; Supplementary Table S2; Supplementary Fig. S2A and S2B). The 5 y OS was comparable between the urban/low-poverty (73.5%) and rural/low-poverty (73.4%) patients, P = 0.9 (Fig. 1). Furthermore, the 5 y OS was lowest for the urban/high poverty (62.5%), and was significantly lower than that for the rural/high-poverty (65.9%) patients (Fig. 1; P < 0.001). Similar patterns were observed when including SMNs as competing events (Supplementary Table S2; Supplementary Figs. S2C and S2D and S3).

Adjusting for age at diagnosis, race/ethnicity, stage of breast cancer at diagnosis, treatment, and preexisting comorbidity, the urban/high-poverty patients were at a 1.2-fold increased hazard [95% confidence interval (CI), 1.16–1.24; P < 0.001] and the rural/high-poverty patients were at a 1.22-fold increased hazard (95% CI, 1.15–1.29; P < 0.001) of 5-year mortality when compared with the urban/low-poverty patients. The rural/low-poverty patients were not at an increased hazard of mortality when compared with urban/low poverty (HR, 0.98; 95% CI, 0.92–1.05; P = 0.7). These observations were consistent across stage-stratified analyses and when including SMNs as an endpoint (Supplementary Tables S3 and S4).

Rural dwellers had a higher rate of PCP visits (IRR, 1.25; 95% CI, 1.23–1.27) and ED visits (IRR, 1.62; 95% CI, 1.51–1.74), and a lower rate of nononcology specialist visits (IRR, 0.75; 95% CI, 0.74–0.76) when compared with urban dwellers, after adjusting for age, stage at diagnosis, race/ethnicity, comorbidity, and treatment (Table 2).

High-poverty patients had a higher rate of PCP visits (IRR, 1.07; 95% CI, 1.05–1.08), oncology visits (IRR, 1.06; 95% CI, 1.04–1.09), ED visits (IRR, 1.40; 95% CI, 1.30–1.51), and hospitalizations (IRR, 1.07; 95% CI, 1.04–1.10), but a lower rate of nononcology specialist visits (IRR, 0.87; 95% CI, 0.85–0.88) compared with the low-poverty patients, after adjusting for covariates listed in the previous model (Table 2).

Table 3 shows that rural patients had a higher rate of PCP visits across all SES categories (rural/low poverty: IRR, 1.33; rural/medium poverty: IRR, 1.27; and rural/high poverty: IRR, 1.23; P < 0.001). They also had a higher rate of ED visits (rural/low poverty: IRR, 1.66; rural/medium poverty: IRR, 1.68; and rural/high poverty: IRR, 1.82; P < 0.001), but significantly lower rates of nononcology specialist visits (rural/low poverty: IRR, 0.69; rural/medium poverty: IRR, 0.74; and rural/high poverty: IRR, 0.74; P < 0.001) compared with the urban/low-poverty group. There was no significant difference in hospitalization rates. Urban/high-poverty patients had higher rates of PCP visits (IRR, 1.08; P < 0.001), hospitalizations (IRR, 1.07; P < 0.001), and ED visits (IRR, 1.31; P < 0.001), but lower rates of nononcology specialist visits (IRR, 0.89; P < 0.001) compared with urban/low-poverty patients.

Table 2 shows that rural dwellers had a higher rate of PCP visits (IRR, 1.17; 95% CI, 1.15–1.19) and ED visits (IRR, 1.34; 95% CI, 1.29–1.41), but a lower rate of nononcology specialist visits (IRR, 0.71; 95% CI, 0.69–0.72) compared with urban dwellers, after adjusting for age, stage at diagnosis, race/ethnicity, preexisting comorbidities, and treatment (surgery, receipt of chemotherapy, and/or radiation).

High-poverty patients had a higher rate of PCP visits (IRR, 1.08; 95% CI, 1.07–1.10), hospitalizations (IRR, 1.14; 95% CI, 1.11–1.17), and ED visits (IRR, 1.21; 95% CI, 1.15–1.27), but a lower rate of nononcology specialist visits (IRR, 0.84; 95% CI, 0.83–0.86) compared with low-poverty patients after adjusting for covariates (Table 2).

Table 3 shows that rural patients had a higher rate of PCP visits across all SES categories (rural/low poverty: IRR, 1.21; rural/middle income: IRR, 1.17; and rural/high poverty: IRR, 1.19; P < 0.001) and ED visits (rural/low poverty: IRR, 1.30; rural/medium poverty: IRR, 1.42; and rural/high poverty: IRR, 1.43; P < 0.001), but lower rates of nononcology specialist visits (rural/low poverty: IRR, 0.63; rural/medium poverty: IRR, 0.69; and rural/high poverty: IRR, 0.71; all P < 0.001) compared with the urban/low-poverty patients. Rural/high-poverty patients had significantly higher rates of hospitalization compared with urban/low-poverty patients (IRR, 1.11; P < 0.001). Urban/high-poverty patients had higher rates of PCP visits (IRR, 1.08; P < 0.001), hospitalizations (IRR, 1.13; P < 0.001), and ED visits (IRR, 1.15; P < 0.001), but significantly lower rates of nononcology specialist visits (IRR, 0.85; P < 0.001) compared with urban/low-poverty patients. Supplementary Table S5 provides results including SMN as an endpoint.

Given the significantly higher rates of PCP visits and lower rates of nononcology specialist visits by the rural/high-poverty patients when compared with the urban/low-poverty patients, we explored the mediating/moderating effects of nononcology specialist visits on the association between residence/SES and mortality. Nononcology specialist visits are likely correlated with the presence of comorbidities and are also likely to be different for patients with stage IV disease when compared with stages 0–III disease.

As shown in Table 4, rural/high-poverty patients experienced a total excess mortality risk of 26%–47% when compared with urban/low-poverty patients across the various strata. Paucity of nononcology specialist visits mediated 23%–57% of the total excess mortality risk, while the interaction between residence/SES and paucity of nononcology specialist visits accounted for 49%–92% of the total excess mortality risk experienced by the rural/high-poverty patients. Supplementary Data S2 provide further interpretation of results for all strata.

The impact of rurality on OS in a population of older Medicare beneficiaries diagnosed with breast cancer was largely predicated on SES. Overall, the high-poverty patient populations of both urban and rural residence demonstrated the lowest 5-year survival rates from end of treatment (63% and 66%, respectively) compared with the 74% 5-year survival rates seen in both urban and rural low-poverty patients. Differences in HCU were most apparent when taking SES into consideration. Rural/high-poverty patients had higher rates of PCP and ED visits, but lower rates of nononcology specialist visits compared with their urban counterparts—both patients in low and high poverty. Furthermore, patients with breast cancer living in rural/high-poverty areas at diagnosis had up to 40% excess risk of mortality compared with those living in urban/low-poverty areas, and this excess risk was largely attributed to the interaction between residence/SES status and paucity of nononcology specialist visits (5, 6).

Previous studies have examined the impact of rural residence or SES on cancer-related mortality or on healthcare access in isolation (17). We find that urban and rural high-poverty older patients with breast cancer were more likely to be diagnosed with stage III disease (26%) compared with their low-poverty counterparts (22%). This contrasts with studies reporting that rural residence is associated with delays in diagnosis resulting in later stage at diagnosis (18). Urban/high-poverty patients were more likely to be NHB and to have more comorbidities than all other groups. Previous evidence demonstrates that NHB women with early-stage breast cancer are less likely to receive treatment, more likely to discontinue therapy, and have worse survival than NHW women (19). Taking stage at diagnosis into account, rural/high-poverty patients in our study were more likely to have total mastectomies and receive chemotherapy compared with both their urban/low-poverty and urban/high-poverty counterparts. Treatment-related differences are important to consider to understand the observed mortality disparity among patients with breast cancer from rural areas or with high poverty, compared with their urban and low-poverty counterparts (20–23). There was no significant difference in visits to oncologists between these two groups, which is in contrast to prior literature suggesting that rural patients may receive less cancer-related care, including radiation (20) and oncology services (21). This may be partially attributable to the fact that this patient population is consistently insured under Medicare and is nationally representative, while the prior studies are largely regionally focused and include a variety of insurance coverage.

Another important characteristic to consider in an older patient population is the high rate of comorbid conditions, which contribute to a need for access to nononcology specialists (7). High-poverty patients with breast cancer, both urban and rural, had the highest prevalence of comorbidities, yet had a lower rate of nononcology specialist visits. However, through the first year after diagnosis, rural patients had higher rates of PCP visits, irrespective of SES, when compared with their urban/low-poverty counterparts. This generally supports the estimates that while there are fewer PCPs in rural areas, there are even fewer medical specialists in the rural setting.

A large proportion of excess mortality risk in rural/high-poverty patients was attributable to the interacting effect of nononcology specialist visits across all levels of comorbidity, and disease stage, suggesting that this population has complex medical needs. However, the proportion of excess risk attributable to the mediating effect of nononcology specialists was lower, suggesting that visits to nononcology specialists were likely not the only explanatory factor in the pathway from residence/SES to mortality, and that other unmeasured mechanisms likely played a role. Visits to a nononcology specialist may be a proxy for other healthcare-seeking behavior, which is an interesting finding in light of the fact that this population is technically an insured population and, theoretically, should have access to receiving care.

Recent studies have asked the question whether providing insurance to patients with cancer actually mitigates negative health effects of living in disadvantaged communities (defined by low median household income, percent of population below FPL, and education achievement in the area; ref, 24). Abdelsattar and colleagues found that when examining uninsured, insured, and Medicaid patients diagnosed with cancer between the ages of 18 to 64 years, health insurance did appear to somewhat mitigate the relationship between a community's SES and cancer care and outcomes, but it did not completely resolve the disparities (24). This is echoed in our study where all patients assessed had continuous health care coverage, yet outcomes varied distinctively across poverty level. The relationship between patient-level and community-level characteristics and the health care policies and resources available in that area are highly complex, warranting a greater examination of the factors that should be targeted to facilitate greater healthcare parity.

This study needs to be examined in the context of its limitations. The urban/rural status and SES of the patient is anchored to when they were diagnosed with breast cancer; the study did not update either the residence or SES through the 5 years of follow-up. Census-tract poverty information does not capture patient-level measures of economic hardship, potentially introducing ecologic bias. Administrative claims are designed for billing purposes and present inherent limitations. In addition, our results only apply to patients with breast cancer, and future studies may examine the differences in the effect of residence/SES on HCU and mortality in other cancer types. Finally, it is possible that not all relevant confounders were captured in these analyses.

This population-based study of older patients with breast cancer finds that rural/urban disparities in survival were modest, but that the underlying SES largely accounts for the patterns of HCU, potentially explaining some of these disparities in outcomes.

G.B. Rocque is a consultant for Pfizer and Genetech and reports receiving commercial research grants from Pfizer, Genentech, and Carevive. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K.M. Kenzik, G.B. Rocque, S. Bhatia

Development of methodology: K.M. Kenzik

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.M. Kenzik

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.M. Kenzik, G.B. Rocque, W. Landier, S. Bhatia

Writing, review, and/or revision of the manuscript: K.M. Kenzik, G.B. Rocque, W. Landier, S. Bhatia

Study supervision: S. Bhatia

K.M. Kenzik was supported by a Mentored Research Scholar Grants in Applied and Clinical Research, MRSG-18-020-01-CPPB, from the American Cancer Society.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.