Adverse Health Outcomes among Rural and Urban Breast Cancer Survivors: A Population-Based Cohort Study

Approximately 19.3% of Americans with cancer live in rural areas (1). There is overwhelming evidence that rural cancer patients are challenged with higher risks of various adverse health outcomes than their urban counterparts (2,–5). A number of studies reported an increased risk of cardiovascular disease (6, 7), diabetes (8), anxiety, depression, suicide (9), or osteoporosis (10) in breast cancer survivors compared with women without cancer. In addition, treatment-induced ocular toxicity and ototoxicity were suggested in patients with breast cancer due to a change in estrogen level from the breast cancer treatments (11,–13). However, to our knowledge, fewer health outcome studies have been reported of breast cancer survivors in rural communities. Studies focusing on distance to healthcare reported a higher likelihood of patients receiving mastectomy compared with lumpectomy in rural areas (14), and a higher likelihood of patients forgoing radiation in part due to lack of rural treatment facilities (15). Other studies reported on higher stage at diagnosis in rural breast cancer survivors (16), and higher odds of heart failure (HF) among older rural cancer survivors (17).

Rural populations are older, with higher poverty levels, and lack access to insurance and health care resources (18, 19). Given that breast cancer survival continues to increase (20), and given the lack of large-scale population-based health outcome studies in rural breast cancer survivor populations, continued understanding of health outcomes in rural populations will result in support for the management of care in patients with breast cancer. Thus, the aim of our study was to assess the risk of adverse health outcomes among rural compared with urban breast cancer survivors and to evaluate potential risk predictors for the highest risk outcomes.

This study cohort included women identified in the Utah Cancer Registry (UCR) diagnosed with first primary breast cancer (primary site ICD-O-3 C50.0 to C50.9). Inclusion criteria were that the breast cancer survivor was a Utah resident, ages ≥18 years at cancer diagnosis, diagnosed between 1997 and 2017, and survived for at least 1 year after breast cancer diagnosis. Rural breast cancer survivors were matched on cancer diagnosis year (±1 year) and age at cancer diagnosis (±1 year) with up to 5 urban breast cancer survivors. A total of 27 rural and 3 urban survivors were excluded for unknown cancer stage. A total of 2,359 rural breast cancer survivors and 11,748 urban breast cancer survivors were included in this study.

The UCR is the statewide, population-based cancer registry for Utah. All cancer survivors from the UCR are linked to the Utah Population Database (UPDB; refs. 21, 22). The UPDB uses record linking IBM InfoSphere QualityStage software to perform probabilistic records linking to various databases, including the UCR. The UPDB records included demographic, Utah driver's license, statewide vital, and family history information linked to medical records. Variables from the UCR included race, ethnicity, residence at cancer diagnosis, birth year, age at cancer diagnosis, cancer treatment receipt histology, estrogen receptor (ER) status, progesterone receptor (PR) status, HER2 status (available starting in 2010), and cancer stage at diagnosis. Within the UPDB, the cancer data is based on the UCR, which collects data on first course cancer-related treatment. Variables from the UPDB included family history of breast cancer, family history of cardiovascular disease, and baseline body mass index (BMI). Baseline BMI was calculated from the height and weight provided in the driver's license records 1 year before the breast cancer diagnosis date. On the basis of the American Academy of Family Physicians coding guidelines, baseline tobacco users were identified 1 year before the breast cancer diagnosis based on the International Classification of Diseases ICD-9/ICD-10 diagnosis codes for tobacco cessation and tobacco addiction (23). The Charlson Comorbidity Index (CCI) was calculated at baseline, from the International Classification of Diseases or ICD-9/ICD-10 diagnosis codes for the year prior to the index date, excluding cancer diagnosis given that the CCI index was based on a cohort of patients with cancer. The CCI index score calculation was based on prevalent comorbidities, a year before cancer diagnosis, as a measure of baseline overall health based on previously established algorithm (24, 25). Patients with prevalent diagnosis for the outcome of the interest were excluded, to calculate incidence of the outcomes of interest. For example, patients with previous heart disease were excluded when estimating the HR for incident heart disease after breast cancer diagnosis.

Rurality of residence was classified according to the rural–urban commuting area (RUCA) codes based on the 2010 decennial census and the 2006–2010 American Community Survey (ACS; ref. 26). The RUCA codes classify US census tracts based on standard census measures of urbanization, population density, and daily commuting from the decennial census. The RUCA codes were aggregated into urban (1.0, 1.1, 2.0–2.2, 3.0, 4.1, 5.1, 7.1, 8.1, and 10.1) and rural (4.0–7.0, 7.2–7.4, 8.0, 8.2–8.4, 9.0–9.2, 10.0, 10.2–10.4, and 10.5) group based on suggested C categorization by the Rural Health Research Center's experts (27). In addition, median household income for each census tract was available from the US Census Bureau of Economic Analysis (28). The Yost socioeconomic status index (SES) at the census 2010 tract level was available for Utah from the Surveillance, Epidemiology, and End Results (SEER) registry. The Yost score is a composite index of neighborhood-level census measure of SES, which incorporates average education, occupation, median income, poverty rate, median housing value, median rent, and unemployment rate (29). Individual-level education information was available from the UPDB.

Outcome data included the following medical records data sources that link with the UPDB: (i) the state ambulatory surgery databases and services databases (SASD), (ii) the inpatient hospital claims database from the Utah Department of Health, and (iii) the electronic medical record (EMR) data from the University of Utah Health (UUH) and the Intermountain Healthcare (IHC). The UPDB-linked records for approximately 94.9% of patient EMR records were from the IHC, and 54.1% of patient EMRs were from the UUH. For patients encountered in both hospital health systems, the first diagnosis code after cancer diagnosis was considered in the analysis, which avoids the potential of overlapping diagnosis on the same day within each or across the two systems. For instance, some patients used both UUH and IHC, however the earliest diagnosis identified in the medical records was used as the incident diagnosis of the outcomes of interest. The primary outcomes measured were a newly diagnosed cardiovascular disease, mental disorders, diabetes, osteoporosis, cataracts, and hearing impairments identified by available ICD-9 and ICD-10 diagnosis codes from the outcome data. The list of ICD diagnosis codes for each outcome of interest was based on the Chronic Conditions Data Warehouse (CCW; ref. 30), which included 15 adverse health outcomes considered in the analysis that are treatment or cancer diagnosis-related.

All baseline descriptive demographic and clinical characteristics for the breast cancer survivors were stratified by urban and rural residence and compared using the Pearson chi-square (χ²). The HRs and 99% confidence interval (99% CI; ref. 31) for incident adverse health outcomes in rural compared with urban cancer survivors were estimated using the Cox proportional hazards model overall from 1 to 5 years and >5 years after the first year from breast cancer diagnosis. Breast cancer survivors with a record of an outcome of interest before the index date were considered prevalent cases, and these cases were excluded from the HR models. The exclusion of prevalent cases in the models allows for calculating incidence. The HR models were stratified to 1 to 5 years and >5 years following the first year from the cancer diagnosis in the interest of investigating the risk of adverse health outcomes following the initial 5 years after the initial year from cancer diagnosis. The HRs were fit using PHREG function. We used the STRATA statement on the matched identification number within the PHREG function to account for the matching factors in the model. On the basis of the three properties of a confounder, on the association between the diagnosis of breast cancer and the risk of adverse health outcomes, we considered race and ethnicity as potential confounders because they are risk factors for the outcomes evaluated, associated with rurality and do not act as mediators. BMI, CCI, and socioeconomic status (Yost) may be mediators since rurality may be a predictor of these factors and therefore adjustment for these variables is not needed. The proportional hazards assumption was tested by creating interaction terms as a function of log (time) and the predictor variables. Flexible parametric modeling with restricted splines was used and reported where estimates differed from the original model, indicating a violation of the proportional hazard assumption. The follow-up time was measured from the date of breast cancer diagnosis (index date), until the earliest occurrence of an event (adverse health outcome) or censoring time (i.e., no outcome, last date of follow-up, or death), whichever occurred first.

In addition, risk factors for adverse health outcomes of significantly higher risk among rural compared with urban breast cancer survivors were assessed using the Cox proportional hazard models and 95% CI. Risk factor models were adjusted for potential confounders, that is, covariates that are risk factors for a given adverse health outcome, associated with the risk factor in question, but unaffected by the risk factor itself (not a mediator). The Cochran's homogeneity test was used to assess HR differences for each risk factor in urban and rural breast cancer survivors.

Using the linear regression model, about 28.7% of the missing education values and 30.2% of the missing BMI values were imputed on the basis of baseline CCI, race and ethnicity, age at cancer diagnosis, and birth year and BMI for education. Further, to examine the differences between the effects of risk factors by rural and urban residence, we modeled interaction effects. Specifically, we assessed interaction terms of residence with ethnicity, SES, radiotherapy, and surgery (individually) for the outcome of HF. In addition, we modeled one interaction term of residence and HF diagnosis with death as an outcome. P values for interaction terms were calculated by the likelihood ratio test comparing the model with and without the product term. Crude and adjusted estimates for each component and joint effects were reported.

Statistical analyses were performed in SAS 9.4 (Statistical Analysis System, RRID:SCR_008567, version 9.4; SAS Institute, Inc.). For all statistical analyses, statistical significance was based on two-tailed tests at the a priori α level of <0.05 for the assessment of risk factors and <0.01 for the main outcomes of interest. The University of Utah Institutional Review Board (IRB) and the oversight committee for the UPDB, the Resource for Genetic and Epidemiologic Research (RGE), approved this study. Under the IRB regulations, this study received approval for waiver of informed consent. The study was conducted in accordance with the ethical guidelines of the Belmont Report.

Raw data for this study can be accessed by the approval of the Resource for Genetic and Epidemiologic Research Committee (RGE), the oversight committee for the UPDB and IRB.

In total, there were 2,359 (16.7%) rural breast cancer survivors and 11,748 (83.3%) urban breast cancer survivors. Rural breast cancer survivors were more likely to be non-Hispanic White and less likely to have at least a college education (Table 1, P < 0.0001). Baseline tobacco use, family history of any cancer, family history of breast cancer, or family history of cardiovascular diseases did not differ between rural and urban survivors. Rural breast cancer survivors were more likely to have had a mastectomy than urban breast cancer survivors (Table 2, P < 0.0001). Similarly, a larger proportion of rural breast cancer survivors did not have radiotherapy than urban breast cancer survivors.

For the overall follow-up, rural breast cancer survivors had a 27% higher (HR, 1.27; 95% CI, 1.06–1.53) risk of HF than urban breast cancer survivors, adjusting for the matching factors, race, and ethnicity (Table 3). Breast cancer survivors from rural areas had a 34% lower risk (HR, 0.66; 95% CI, 0.56–0.78) of cataracts, a 19% lower risk (HR, 0.81; 95% CI, 0.69–0.96) of hyperlipidemia, and a 15% lower risk of osteoporosis (HR, 0.85; 95% CI, 0.72–1.00, P = 0.008) than urban breast cancer survivors, adjusting for potential confounders (Table 3).

Within the 5 years of follow-up, rural breast cancer survivors had a 39% (HR, 1.39; 95% CI, 1.02–1.65) higher risk of HF than urban breast cancer survivors, adjusting for confounders (Table 3). Rural breast cancer survivors continued to have lower risks of cataracts and hearing impairments than rural breast cancer survivors after 5 years of follow-up. There was no increase in the risk for other cardiovascular outcomes or mental health outcomes overall, within 5 years and >5 years of follow-up.

Demographic and clinical risk factors were assessed for HF because this was the only outcome for which there was an increased risk in rural compared to urban breast cancer survivors. Family history of cardiovascular diseases, family history of breast cancer, lower education attainment (rural only), and higher baseline BMI and CCI were risk factors for HF in rural and urban breast cancer survivors, with similar levels of risk (Table 4). Advanced cancer stage and single-agent chemotherapy treatment were associated with an increased risk for HF following breast cancer diagnosis, however, no heterogeneities were found between rural and urban breast cancer survivors (Table 5). Interaction effects between residence and ethnicity, SES, surgery, and radiotherapy on the risk of HF were not statistically significant (Table 6). An interaction term between residence and HF diagnosis (Table 6, P = 0.024) on the risk of death was statistically significant, indicating differences in risks of death among rural and urban breast cancer survivors, favoring rural breast cancer survivors.

In this population-based cohort, we evaluated the burden of cardiovascular disease, diabetes, mental health disorders, cataracts, hearing impairment, and osteoporosis for rural compared with urban breast cancer survivors. Breast cancer survivors in rural compared with urban areas had a higher risk of HF overall and within 1 to 5 years after the initial year from breast cancer diagnosis. There was no increase in the risk for any other health outcomes evaluated in this study. The risk of cataracts and hearing impairments was lower for rural breast cancer survivors overall, within and >5 years of follow-up. Further, a higher baseline BMI and CCI, family history of CVD, family history of breast cancer, and advanced cancer stage were potential risk factors of incident HF risk, although with similar levels of risk for rural and urban breast cancer survivors.

With respect to demographics, differences in income and ethnicity are consistent with a previous study on disparities in urban and rural breast cancer survivors identified within the SEER database (11). However, our observation of no difference in baseline BMI and CCI between rural and urban breast cancer survivors in this study was not expected. It is possible that in Utah, in contrast with other parts of the country, rural and urban breast cancer survivors do not differ greatly for these baseline characteristics. Our study confirmed treatment disparities between rural and urban breast cancer survivors (32). Radiotherapy in particular is more challenging since daily treatment over several weeks are required (33, 34), and the difficult of traveling a long distance to treatment facilities is a predominant barrier to receiving extended treatments for rural communities. Unlike previous findings (35), we did not observe any differences for cancer stage between rural and urban patients with breast cancer. Utah's overall cancer screening rates are lower than the national average (36, 37), potentially resulting in higher rates of delayed cancer diagnosis than reported in states with a higher percentage of rural women who underwent mammography screening (38).

The increased HF risk in rural breast cancer survivors in this study may be attributed to more intensive forms of cancer treatments since they are diagnosed at a later stage, which could increase risk of late effects closer to breast cancer diagnosis. However, the increased risk of HF after 5 years of follow-up was not statistically significant; perhaps because we lacked the statistical power to detect an association. Conversely, because of lower treatment adherence, we hypothesize that rural cancer survivors may experience a lower incidence of treatment-related late effects. Therefore, the increased HF risk observed in this study may in part be due to noncancer factors, the increased risk of HF in women of low-income from rural communities compared with their urban counterparts had been previously reported (39). These findings highlight the need for primary preventive strategies for rural cancer patients at risk of cardiovascular outcomes, including increased cardiac surveillance and monitoring to help lessen potential barriers to heart health in rural communities.

In terms of HF risk factors, the higher HF risk in single-agent treated breast cancer patients in this study may be due to treatment toxicity leading to treatment discontinuation and patients receiving single rather than multiple needed treatments. However, it may be more plausible that these patients may have received multiagent treatments but were inaccurately categorized as receiving single instead of multiple treatments. Given that the data on patients who received chemotherapy from the cancer registry may be underrepresented, it is not surprising that we did not observe an association for patients receiving chemotherapy. Similarly, we may have lacked the statistical power needed to detect an association and may not have captured the long-term effects of chemotherapy on the heart.

The risks of other adverse health outcomes evaluated, such as hypertension, diabetes, anxiety, and depression, did not differ between rural and urban breast cancer survivors. Underutilization of healthcare for these health outcomes is likely due to screening barriers in rural areas, or some conditions may not have been severe enough to be captured in our study. Depression and anxiety are of great concern for cancer survivors (9); however, depending on screening assessment methods or healthcare seeking behaviors, these outcomes may be underdiagnosed among cancer survivors, regardless of residence. It is unclear why rural breast cancer survivors in this study had a lower risk of hyperlipidemia, cataracts, hearing impairment, or osteoporosis. However, due to lower treatment adherence or care management in rural communities, it is possible that rural cancer survivors may have a lower incidence of certain treatment or noncancer-related effects. To further investigate the effects of breast cancer treatment or cancer diagnosis on rural breast cancer survivors, future studies using a general population without cancer as a comparison group are needed.

The major strength of this study is that it is the first study to comprehensively assess adverse health outcomes through prolonged follow-up of a relatively large sample of rural breast cancer survivors. Further, all diagnoses are based on electronic medical and ambulatory discharge records from large state regional healthcare providers in the state and are not subject to recall bias, which is problematic for studies based on self-reported outcomes. Similarly, although electronic medical records may not capture less severe diagnoses, medical records allow for the inclusion of a wide range of available ICD diagnosis codes for the identification of evaluated adverse health outcomes.

There are limitations to consider for this study. Although our sample included approximately 4.6% rural and 9.9% urban Hispanic breast cancer survivors, it is unlikely that our findings can be generalizable to more diverse rural regions of the United States. Similarly, Utah's low alcohol drinking and cigarette smoking rates compared with the rest of the country (40) may contribute to a healthier cohort of breast cancer survivors, potentially resulting in lower comorbidity risk estimates compared with other breast cancer cohorts. In the first few years following a cancer diagnosis, cancer survivors undergo increased medical surveillance, including more frequent follow-up visits and medical screening. However, breast cancer survivors in rural communities may receive fewer follow-up visits in the early years following a cancer diagnosis, which may minimize the frequency of outcomes evaluated. Nevertheless, there was a higher risk of HF in the first 5 years of follow-up among rural breast cancer survivors. Given the number of patients with ER-positive breast cancer, endocrine therapy usage is likely underreported in the overall study sample. These missing data are likely to bias the results towards the null, regardless of residency.

In conclusion, we observed an increased risk of HF among rural compared with urban breast cancer survivors. Future studies are needed to investigate preventive approaches to identify patients at high risk of cardiovascular outcomes for whom preventive strategies are warranted and can be implemented in rural areas to reduce the comorbidity burden among rural breast cancer survivors. Although other adverse health outcomes did not differ for rural and urban breast cancer survivors in this study, investigating these outcomes remains essential for understanding the comorbidity burden across rural populations in the United States.

S. Lloyd reports personal fees from Cancer Study Group and Kipp and Christian outside the submitted work. V.G. Deshmukh is a co-founder of Backdrop Health Inc. and ownership of Backdrop Health Inc. stock. A. Date reports grants from NIH during the conduct of the study; other support from AstraZeneca outside the submitted work. C.A. Porucznik reports personal fees from McKesson Corporation outside the submitted work. B. Haaland reports other support from Pentara Corporation and personal fees from National Kidney Foundation outside the submitted work. N. Henry reports grants from National Cancer Institute during the conduct of the study; personal fees from Myovant Sciences and Up-To-Date outside the submitted work. M. Hashibe reports grants from NCI during the conduct of the study. No disclosures were reported by the other authors.

A. Koric: Conceptualization, formal analysis, methodology, writing–original draft. B. Mark: Formal analysis, writing–review and editing. C.-P. Chang: Writing–review and editing. S. Lloyd: Writing–review and editing. M. Dodson: Methodology, writing–review and editing. V.G. Deshmukh: Data curation, methodology. M. Newman: Data curation, methodology. A. Date: Data curation, methodology. L.H. Gren: Writing–review and editing. C.A. Porucznik: Writing–review and editing. B. Haaland: Writing–review and editing. N. Henry: Writing–review and editing. M. Hashibe: Conceptualization, supervision, funding acquisition, methodology, writing–review and editing.

This work was supported by grants from the NIH (R01 CA244326, R21 CA185811, R03 CA159357, M. Hashibe, PI), the Huntsman Cancer Institute, and the Cancer Control and Population Sciences Program (HCI Cancer Center Support Grant P30CA042014). This research was supported by the Utah Cancer Registry, which is funded by the NCI's SEER Program, Contract No. HHSN261201800016I, the U.S. Center for Disease Control and Prevention's National Program of Cancer Registries, Cooperative Agreement No. NU58DP007131–01, with additional support from the University of Utah and Huntsman Cancer Foundation. Partial support for all datasets within the Utah Population Database is provided by the University of Utah, Huntsman Cancer Institute and the Huntsman Cancer Institute Cancer Center Support grant, P30CA042014 from the NCI. The computational resources used were partially funded by the NIH Shared Instrumentation Grant 1S10OD021644–01A1.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.