Abstract
Background
Patients with cancer living in rural areas have inferior cancer outcomes; however, studies examining this association use varying definitions of “rural,” complicating comparisons and limiting the utility of the results for policy makers and future researchers.
Methods
Surveillance, Epidemiology, and End Results data (2000-2016) were used to assess risk of cancer mortality and mortality from any cause across 4 definitions of rurality: Urban Influence codes (UIC), National Center for Health Statistics (NCHS), Rural-Urban continuum codes (RUCC), and Index of Relative Rurality. Binary (urban vs rural) and ternary (urban, micropolitan, rural) definitions were evaluated. Multivariable parametric survival models estimated hazards of mortality overall and among 3 cancer groupings: screening related, obesity related, and tobacco related. Definition agreement was also assessed.
Results
Overall, 3 788 273 patients with an incident cancer representing 605 counties were identified. There was little discordance between binary definitions of rural vs urban and moderate agreement at the 3 levels. Adjusted models using binary definitions revealed 15% to 17% greater hazard of cancer mortality in rural compared with urban. At the 3 levels when comparing rural with metropolitan, RUCC and NCHS saw similarly increased hazard ratios; however, Index of Relative Rurality did not. Screening-related cancers saw the highest hazards of mortality and the largest divergence between definitions. Obesity-related and tobacco-related cancers saw similarly increased hazards of mortality at the binary and ternary levels.
Conclusions
Hazard of death is similar across binary definitions; however, this differed when categorized as ternary or continuous, especially among screening-related cancers. Results suggest that study purpose should direct choice of definitions and categorization.
US cancer mortality rates declined by 26% between 1991 and 2015 (1), but the decline has not been equal across all populations (2). Approximately 19% of the US population resides in rural areas; for over 3 decades, this population has experienced increasingly inferior cancer outcomes compared with their urban counterparts (3-5). Defining rurality remains a moving target. Studies use varying definitions of “rural,” potentially increasing the complexity in comparing studies and the utility of results for future policy or intervention (6,7). To fully evaluate rural disparities in cancer survival, any discordance between definitions and subsequent effect on survival estimates should be evaluated.
The choice of a definition can be due to constraints placed by data availability or for strategic reasons in applying the findings to geographic areas. Definitions using counties may be more heterogeneous compared with census tracts or zip code definitions, yet policy and funding decisions are often based on county designations (8). County-level data include the US Department of Agriculture’s Rural-Urban Continuum Codes (RUCC) (9) and Urban Influence Codes (UIC) (10), The National Center for Health Statistics (NCHS) Urban-Rural Classification Scheme for Counties (11), and the Purdue University Index of Relative Rurality (IRR) (12). Each definition varies in categorization of metropolitan vs nonmetropolitan (urban vs rural). The RUCC and UIC are similar in their definition of a metropolitan county, with RUCC having 3 categories and the UIC having 2 categories, both based on population size. However, the UIC further categorizes the nonmetropolitan counties into micropolitan and noncore; micropolitan is defined by the size of the adjacent metro area alone and noncore by the size of the adjacent metro or micro area and the town population. NCHS emphasizes metropolitan areas and is similar in definition to the RUCC; however, nonmetropolitan is defined by only 1 micropolitan and 1 rural code. The IRR is a continuous measure of rurality using population size, density, remoteness, and built-up area. When defined as a binary variable (metropolitan vs nonmetropolitan), the RUCC, UIC, and NCHS definitions are identical; however, when defined as ternary (metropolitan, micropolitan, or noncore or rural), RUCC differentiates from NCHS and UIC (9-12). These differences should be considered when using different measures of rurality for policy decisions related to allocation of funding for health-care resources, incentives for providers or insurers, or protection of resources (eg, rural health-care facilities).
To date, few studies have examined discordance between rurality definitions (13,14). Moss et al. (13) focused on the patient confidentiality associated with using census tract-level measures in the identification of urban vs rural incidence of breast and lung cancer patients in the Surveillance, Epidemiology, and End Results (SEER) data. Another study focused on the discordance in both county- and tract-level definitions in the Midwest and Wisconsin using SEER data but did not address cancer survival (14). There remain gaps in evidence, including the impact of different definitions on cancer mortality, an evaluation across the United States, and inclusion of multiple cancer types.
This study estimated cancer survival among individuals diagnosed with cancer by 4 county-level urban and rural definitions. We note varying use of terminology in previous studies (eg, metropolitan and urban); for brevity, we use the terms “rural” and “urban” in this manuscript although the distinction in definition should be noted.
Methods
Data sources
A program under the National Cancer Institute, SEER collects high-quality cancer data (incidence, prevalence, and mortality) from 21 cancer registries, capturing 35% of the US population. Data include patient demographics, primary tumor site, tumor morphology and stage at diagnosis, first course of treatment, and vital status. This study used the SEER 18 data released in April 2021 for SEER 8.4.
Sample population
We identified individuals with key incident cancers (exception: prior melanoma) diagnosed between 2000 and 2016 at age ≥20. We selected cancer types based on prior literature indicating mortality differences by urban-rural status. We created the following groups (Supplementary Table 1, available online): 1) cancers with screening recommendations: evidence of lower screening rates in rural locations (15) including breast, lung, cervical, colorectal, and prostate cancer; 2) obesity related: obesity accounts for 40% of cancers in the United States (16). Rural areas have a higher prevalence of obesity compared with urban counties (34% vs 28%) (17). These included cancers of breast, prostate, colon, rectal, endometrial, kidney, pancreas, ovarian, stomach, liver, esophagus, thyroid, and meningioma; 3) smoking- or tobacco-related cancers: rates of smoking or tobacco use are higher in rural areas (17% vs 12%) (18,19). These included cancers of lung, oral cavity, larynx, pharynx, esophagus, cervical, liver, pancreas, colorectal, and kidney.
Primary sample selection yielded 4 118 950 individuals. We excluded individuals with unmatched county codes (n = 79 498), missing stage information (n = 245 494), and male breast cancer (n = 5685), yielding a final sample of 3 788 273 individuals. Given the large sample size, a 20% random sample from the overall and 30% cancer type–stratified random sample from each subgroup was examined.
Urban and rural definitions
Supplementary Table 2 (available online) provides detailed definitions of rural designation. For UIC, RUCC, and NCHS we used the 2013 data; for IRR, we used the 2010 data (closest available). UIC included 12 categories, with 2 categories for metropolitan counties (counties with ≥1 urbanized areas with population ≥50 000) and 10 nonmetropolitan categories. The nonmetropolitan counties included micropolitan and noncore (remaining) counties for the 3-level definition. RUCC classification defined metropolitan counties by population size and subdivided nonmetropolitan counties by degree of urbanization and adjacency to urban areas, resulting in 9 categories (3 metro; 6 nonmetro). The NCHS codes focus on the refinement of metropolitan categories. In the 6 total categories, 4 are metropolitan and 2 nonmetropolitan (1 micropolitan and 1 rural). IRR is continuous from 0 (completely metropolitan) to 1 (completely rural or remote); we used a threshold of 0.5 for urban vs rural and 0.4 for specifying 3-level metropolitan (≤0.4), micropolitan (>0.4 to <0.6), and rural (≥0.6) (20).
Outcome
The primary outcome was hazard of cancer mortality at 5 years. Secondary outcomes included 1) hazard of cancer mortality among the 3 subcohorts, and 2) hazard of mortality from any cause across the binary and ternary definitions and the IRR continuous measure. We used the SEER indicator for vital status (all-cause) and Cause of Death Recode (using International Classification of Diseases 8th-10th edition).
Statistical analyses
Each definition was examined for the urban vs rural frequency. Overlap between designations was evaluated to determine if the number of definitions could be reduced. Cohen’s kappa with an ordinal weight was calculated to assess the level of agreement across definitions at both the 2-level and 3-level definitions (strong agreement: ≥0.7).
Using flexible parametric survival models, unadjusted and adjusted cancer-specific (competing risk model) (21) and all-cause hazard ratios (HRs) were estimated. Primary exposure was urban or rural status. Each rural definition was evaluated separately, with models for binary and ternary definition, and continuous measure of IRR. Multivariable models included categories of age at diagnosis (20-44 years, 45-54 years, 55-64 years, 65-74 years, 75-84 years, 85+ years), race and ethnicity, sex, stage (early or late), and year of diagnosis. Stage was dichotomized into early and late such that the SEER summary stages in situ, localized, and regional were considered early and distant was considered late. Data management used SAS 9.4 (22) and survival models used Stata V16 (23). This study was deemed nonhuman subjects research by the Boston University Institutional Review Board.
Results
County characteristics
Of the 605 counties captured in this study, 57.7% were rural according to the binary RUCC, UIC, and NCHS definition and 57.8% according to the IRR. There were no discordant pairs between the 2-level definitions of RUCC, UIC, and NCHS (Figure 1). We were able to remove the UIC as a separate analysis because it is represented by the RUCC binary definition and the NCHS ternary definition. NCHS had a high level of agreement IRR (Cohen kappa: 0.731). For the 3-level definitions, RUCC and NCHS had a Cohen weighted kappa of 0.846. RUCC and NCHS had moderate agreement with IRR (Cohen weighted kappa = 0.678, 0.58). Figure 2 depicts agreement between the 3-level definitions of rurality for the US SEER and non-SEER populations.
Figure 1.
Agreement between A) binary and B) ternary definitions of rurality using weighted Cohen kappa. The binary Rural-Urban continuum codes (RUCC), National Center for Health Statistics (NCHS), and Urban Influence Codes (UIC) had perfect agreement with each other and substantial agreement with the binary Index of Relative Rurality (IRR; κ = 0.73). Among the ternary definitions, NCHS and UIC had perfect agreement, RUCC and NCHS had the highest agreement (κ = 0.84), and the NCHS and IRR had the lowest agreement (κ = 0.58).
Figure 2.
Proportion rural in each county as defined by the ternary Rural-Urban continuum codes (RUCC) (A), Index of Relative Rurality (IRR) (B), and National Center for Health Statistics (NCHS) (C) definitions of rurality among the US Surveillance, Epidemiology, and End Results (SEER) and non-SEER populations.
Sample characteristics
Of the 3 788 273 individuals in the overall cohort, 2 658 723 were screening-related cancers (70.2%), 2 173 409 obesity-related cancers (57.4%), and 1 783 638 tobacco-related cancers (47%). All random sample cohorts were representative of the original sample (Supplementary Table 3, available online). Patients primarily resided in urban areas (88%-90%; Table 1). The majority of patients were non-Hispanic white (71%) diagnosed at age 55 years or older (77%) with stage I disease (52%). Trends were consistent across the random sample cohorts, except for median survival times (screening: 48 months, obesity: 44 months, and tobacco: 10 months).
Table 1.
Patient demographics and clinical characteristics overall and by cancer typea
| Overall (n = 757 655) | Screening related (n = 797 616) | Obesity related (n = 652 023) | Tobacco related (n = 535 091) | |
|---|---|---|---|---|
| No. (%) | No. (%) | No. (%) | No. (%) | |
| Age at diagnosis, y | ||||
| 20-44 | 58 733 (7.8) | 50 040 (6.3) | 73 105 (11.2) | 31 513 (5.9) |
| 45-54 | 120 253 (15.9) | 120 781 (15.1) | 125 968 (19.3) | 76 259 (14.3) |
| 55-64 | 202 772 (26.8) | 213 542 (26.8) | 165 286 (25.3) | 133 782 (25.0) |
| 65-74 | 207 518 (27.4) | 232 346 (29.1) | 148 500 (22.8) | 146 470 (27.4) |
| 75-85 | 130 059 (17.2) | 141 407 (17.7) | 102 941 (15.8) | 110 586 (20.7) |
| 85+ | 38 320 (5.1) | 39 500 (5.0) | 36 223 (5.6) | 36 371 (6.8) |
| Sex | ||||
| Female | 389 624 (51.4) | 395 161 (49.5) | 472 677 (72.5) | 237 115 (44.3) |
| Male | 368 031 (48.6) | 402 455 (50.5) | 179 346 (27.5) | 297 976 (55.7) |
| Race and ethnicity | ||||
| Hispanic | 75 755 (10.0) | 71 646 (9.0) | 75 042 (11.5) | 52 004 (9.7) |
| Non-Hispanic Black | 90 297 (11.9) | 98 149 (12.3) | 71 575 (11.0) | 63 481 (11.9) |
| Non-Hispanic White | 537 966 (71.0) | 574 773 (72.1) | 453 367 (69.5) | 381 920 (71.4) |
| Non-Hispanic Other | 53 637 (7.1) | 53 048 (6.7) | 52 039 (8.0) | 37 686 (7.0) |
| Stage | ||||
| In situ | 4604 (0.6) | 5991 (0.8) | 6444 (1.0) | 1458 (0.3) |
| Localized | 392 708 (51.8) | 434 301 (54.5) | 333 509 (51.2) | 173 148 (32.4) |
| Regional | 196 635 (26.0) | 198 877 (24.9) | 196 835 (30.2) | 161 739 (30.2) |
| Distant | 163 708 (21.6) | 158 447 (19.9) | 115 235 (17.7) | 198 746 (37.1) |
| Treatment | ||||
| Surgery | 289 788 (38.2) | 322 987 (40.5) | 122 224 (18.8) | 262 058 (49.0) |
| Chemotherapy | 219 801 (29.0) | 229 710 (28.8) | 225 183 (34.5) | 204 817 (38.3) |
| Radiation | 259 309 (34.2) | 302 893 (38.0) | 195 794 (30.0) | 149 913 (28.0) |
| Second cancer diagnosis | 78 077 (10.3) | 86 935 (10.9) | 71 580 (11.0) | 47 066 (8.8) |
| Cause of death | ||||
| Alive | 421 215 (55.6) | 459 484 (57.6) | 389 441 (59.7) | 186 588 (34.9) |
| Primary cancer | 199 359 (26.3) | 204 341 (25.6) | 149 044 (22.9) | 236 027 (44.1) |
| Secondary cancer | 44 836 (5.9) | 32 053 (4.0) | 40 836 (6.3) | 44 048 (8.2) |
| CVD | 38 847 (5.1) | 44 280 (5.6) | 30 162 (4.6) | 27 014 (5.1) |
| All other reasons | 53 398(7.0) | 57 448 (7.2) | 42 540 (6.5) | 41 414 (7.7) |
| RUCC | ||||
| Urban | 670 214 (88.5) | 704 420 (88.3) | 581 906 (89.3) | 465 925 (87.1) |
| Rural | 87 441 (11.5) | 93 196 (11.7) | 70 117 (10.8) | 69 166 (12.9) |
| UIC | ||||
| Urban | 680 598 (89.8) | 715 620 (89.7) | 590 419 (90.6) | 473 578 (88.5) |
| Rural | 77 057 (10.2) | 81 996 (10.3) | 61 604 (9.5) | 61 513 (11.5) |
| NCHS | ||||
| Urban | 670 214 (88.5) | 704 420 (88.3) | 581 906 (89.3) | 465 925 (87.1) |
| Rural | 87 441 (11.5) | 93 196 (11.7) | 70 117 (10.8) | 69 166 (12.9) |
| IRR | ||||
| Urban | 683 746 (90.2) | 718 926 (90.1) | 593 208 (91.0) | 476 667 (89.1) |
| Rural | 73 909 (9.8) | 78 690 (9.9) | 58 815 (9.0) | 58 424 (10.9) |
Overall is based on a 20% sample of the total population. Screening, obesity, and tobacco related are based on a 30% random sample of the total population. CVD = cardiovascular disease; IRR = Index of Relative Rurality; NCHS = National Center for Health Statistics; RUCC = Rural-Urban Continuum Codes; UIC = Urban Influence Codes.
Hazard of cancer death at 5 years
All cancers—binary definition
After adjusting for age, sex, race, year of diagnosis, and stage, patients residing in rural areas as defined by the binary RUCC, UIC, and NCHS definition experienced a 15% (HR = 1.15, 95% confidence interval [CI] = 1.13 to 1.16) greater hazard of cancer mortality at 5 years compared with those in urban areas (Table 2; Figure 3). Using the binary IRR, rural patients experienced a 15% (HR = 1.15, 95% CI = 1.13 to 1.17) greater hazard of mortality. Thus, the binary definitions of RUCC, UIC, and NCHS and IRR yielded 5-year hazard ratio estimates within a 1% to 2% range. Unadjusted analyses are presented in Supplementary Tables 4 and 5 (available online).
Table 2.
Adjusted hazard of 5-year cancer-specific mortality between rural and urbana
| Overall | Screening related | Obesity related | Tobacco related | |
|---|---|---|---|---|
| HR (95% CI) | HR (95% CI) | HR (95% CI) | HR (95% CI) | |
| Binary | ||||
| RUCC | 1.15 (1.13 to 1.16) | 1.19 (1.17 to 1.20) | 1.07 (1.05 to 1.09) | 1.08 (1.07 to 1.10) |
| IRR | 1.15 (1.13 to 1.17) | 1.19 (1.17 to 1.20) | 1.08 (1.07 to 1.10) | 1.09 (1.07 to 1.10) |
| Ternary | ||||
| RUCC (ref: metropolitan) | ||||
| Micropolitan | 1.14 (1.13 to 1.16) | 1.18 (1.16 to 1.19) | 1.07 (1.05 to 1.09) | 1.08 (1.07 to 1.10) |
| Rural/noncore | 1.17 (1.13 to 1.22) | 1.25 (1.21 to 1.29) | 1.09 (1.05 to 1.14) | 1.11 (1.07 to 1.15) |
| IRR (ref: metropolitan) | ||||
| Micropolitan | 1.16 (1.14 to 1.17) | 1.19 (1.17 to 1.21) | 1.09 (1.07 to 1.11) | 1.09 (1.07 to 1.10) |
| Rural/noncore | 0.99 (0.89 to 1.09) | 1.01 (0.92 to 1.11) | 0.97 (0.87 to 1.09) | 1.05 (0.95 to 1.16) |
| NCHS (ref: metropolitan) | ||||
| Micropolitan | 1.13 (1.11 to 1.15) | 1.16 (1.14 to 1.18) | 1.08 (1.05 to 1.10) | 1.07 (1.06 to 1.09) |
| Rural/noncore | 1.17 (1.14 to 1.19) | 1.22 (1.20 to 1.24) | 1.12 (1.09 to 1.14) | 1.10 (1.08 to 1.11) |
| Continuous | ||||
| IRR continuous | 1.60 (1.54 to 1.66) | 1.79 (1.73 to 1.85) | 1.29 (1.23 to 1.34) | 1.38 (1.33 to 1.43) |
HR = hazard ratio; IRR = Index of Relative Rurality; CI = confidence interval; NCHS = National Center for Health Statistics; RUCC = Rural-Urban Continuum Codes.
Figure 3.
Hazard of cancer-related mortality among the A) overall, B) screening-related, C) obesity-related, and D) tobacco-related groups across ternary definitions of rurality. IRR = Index of Relative Rurality; NCHS = National Center for Health Statistics; RUCC = Rural-Urban Continuum Codes.
All cancers: ternary definition
Patients residing in micropolitan or rural areas as defined by RUCC had a 14% (HR = 1.14, 95% CI = 1.13 to 1.16) and 17% (HR = 1.17, 95% CI = 1.13 to 1.22) higher hazard of 5-year cancer mortality than those residing in urban areas. Micropolitan areas defined by the IRR demonstrated a 17% higher cancer death hazard (HR = 1.17, 95% CI = 1.14 to 1.17). Rural was not statistically significantly different from urban in the IRR. Finally, using the NCHS and UIC definition, micropolitan and rural patients had 13% (HR = 1.13, 95% CI = 1.11 to 1.15) and 17% (HR = 1.17, 95% CI = 1.14 to 1.19) higher hazard of cancer death compared with urban or metropolitan, respectively. Overall, ternary definitions of rurality yielded estimates within 1% to 3%.
All cancers: continuous
For each unit increase in IRR, hazard of death increased by 60% (95% CI = 1.54 to 1.66).
Screening-related cancer: binary definition
Adjusted hazard ratio estimates were 1.19 1.19 for the RUCC, UIC, and NCHS and IRR binary definitions (Table 2).
Screening-related cancer: ternary definition
Micropolitan residence defined by the RUCC, IRR, and NCHS/UIC yielded hazard ratio estimates of 1.18, 1.19, and 1.16, respectively (Figure 3, A-D). For those defined as rural under the RUCC and NCHS/UIC, estimates were 1.25, and 1.22. The IRR hazard ratio estimate for rural patients under the 3-level definition was not statistically significant. Ternary definitions of rurality yielded estimates within 1% to 7% of each other among those with screening-related cancers.
Screening-related cancer: continuous
For each unit increase in IRR, hazard of death increased by 79% (HR = 1.79, 95% CI = 1.73 to 1.85).
Obesity-related cancer: binary definition
Hazard ratio estimates for the obesity-related cancer cohort were 1.07 for the RUCC, NCHS, and UIC and 1.08 for the IRR binary definitions (Table 2).
Obesity-related cancer: ternary definition
Micropolitan residence defined by the RUCC, IRR, and NCHS/UIC resulted in statistically significant hazard ratio estimates of 1.07, 1.09, and 1.08, respectively (Figure 3, A-D). For those defined as rural using RUCC and NCHS/UIC, the hazard ratio estimates were 1.09 and 1.12. The IRR hazard ratio estimate for rural patients was not statistically significant. Thus, the ternary definitions of rurality yielded estimates within 1%-4% of each other among those with obesity-related cancers.
Obesity-related cancer: continuous
For each unit increase in IRR, hazard of death increased by 29% among the obesity-related cancer cohort (HR = 1.29, 95% CI = 1.23 to 1.34).
Tobacco-related cancer: binary definition
Hazard ratio estimates for the tobacco-related cancer cohort were 1.08 and 1.09 for the RUCC, NCHS, and UIC and IRR binary definitions, respectively (Table 2).
Tobacco-related cancer: ternary definition
Micropolitan residence defined by the RUCC, IRR, and NCHS/UIC resulted in hazard ratio estimates of 1.07, 1.07, and 1.06, respectively (Figure 3, A-D). For rural, the hazard ratio estimates were 1.11 (RUCC) and 1.10 (NCHS/UIC). The IRR hazard ratio estimate for rural patients was not statistically significant.
Tobacco-related cancer: continuous
For each unit increase in IRR, the hazard of death increased by 38% in the tobacco-related cancer cohort (HR = 1.38, 95% CI = 1.33 to 1.43).
Hazard of death from any cause
Screening-related cancers had the largest discrepancies between the ternary definitions (2%-25% difference) (Table 3). For micropolitan in the ternary definition, obesity-related cancers had the largest differences across definitions (2%-15% across definitions). Continuous IRR estimates ranged from 1.37 for obesity and tobacco related to 1.75 for screening related.
Table 3.
Adjusted hazard of 5-year all-cause mortality between rural and urbana
| Overall | Screening related | Obesity related | Tobacco related | |
|---|---|---|---|---|
| HR (95% CI) | HR (95% CI) | HR (95% CI) | HR (95% CI) | |
| Binary | ||||
| RUCC | 1.17 (1.16 to 1.18) | 1.20 (1.18 to 1.21) | 1.11 (1.09 to 1.12) | 1.10 (1.09 to 1.11) |
| IRR | 1.17 (1.15 to 1.18) | 1.19 (1.18 to 1.20) | 1.11 (1.09 to 1.12) | 1.10 (1.09 to 1.11) |
| Ternary | ||||
| RUCC (ref: metropolitan) | ||||
| Micropolitan | 1.16 (1.15 to 1.18) | 1.19 (1.17 to 1.20) | 1.10 (1.09 to 1.11) | 1.09 (1.08 to 1.11) |
| Rural/noncore | 1.21 (1.18 to 1.25) | 1.27 (1.23 to 1.30) | 1.14 (1.09 to 1.17) | 1.13 (1.10 to 1.16) |
| IRR (ref: metropolitan) | ||||
| Micropolitan | 1.17 (1.16 to 1.18) | 1.19 (1.18 to 1.21) | 1.11 (1.10 to 1.13) | 1.10 (1.09 to 1.11) |
| Rural/noncore | 0.98 (0.91 to 1.06) | 1.02 (0.95 to 1.10) | 0.99 (0.91 to 1.08) | 1.04 (0.96 to 1.13) |
| NCHS (ref: metropolitan) | ||||
| Micropolitan | 1.15 (1.14 to 1.17) | 1.17 (1.16 to 1.19) | 1.09 (1.07 to 1.11) | 1.09 (1.07 to 1.10) |
| Rural/noncore | 1.19 (1.17 to 1.20) | 1.22 (1.21 to 1.24) | 1.12 (1.10 to 1.14) | 1.11 (1.10 to 1.13) |
| Continuous | ||||
| IRR continuous | 1.60 (1.55 to 1.65) | 1.75 (1.70 to 1.80) | 1.37 (1.33 to 1.41) | 1.37 (1.34 to 1.41) |
HR = hazard ratio; IRR = Index of Relative Rurality; LCI = lower confidence interval; NCHS = National Center for Health Statistics; Rural-Urban Continuum Codes; UCI = upper confidence interval.
Discussion
The use of various measures of rurality has created uncertainty about the validity of comparing results across studies. In this study, patients residing in rural areas had a greater hazard of death from cancer and from any cause, regardless of definition, with the exception of the ternary definition of rural in the Index of Relative Rurality. The binary definitions of the RUCC, NCHS, UIC, and IRR were comparable such that adjusted hazards of cancer death differed only by 1% to 3%, and the ternary definitions varied by up to 4%. A continuous IRR measure varied widely between groups, with hazard ratios ranging from 1.30 to 1.80. Findings indicate some discrepancies between rurality definitions; minimal when categorized at binary or ternary levels compared with continuous with additional variation among the cancer groups assessed.
Sixty percent of all counties in the United States are nonmetropolitan (674 micropolitan and 1378 rural). This study included 605 counties, of which 57% were rural according to the RUCC, UIC, and NCHS and 58% according to the IRR. Although all definitions consider some degree of population size and density, remoteness, and proximity to urban areas, each definition uses different cutoffs for “urban” or “metropolitan.” RUCC and UIC consider counties with populations less than 250 000 and 1 million residents, respectively, as urban or metropolitan, whereas IRR incorporates degree of rurality relative to other places. Thus, because the RUCC and UIC consider only extreme scores as “rural,” using these definitions will yield slightly larger urban or rural differences in survival compared with the IRR categorized at 2 or 3 levels. With respect to rural health-care policies, choice of measure and number of categories may result in the unintentional exclusion of high-risk individuals from government programs and other interventions intended to mitigate disparities.
We found consistency between the RUCC, UIC, and NCHS with broad binary definitions but differences within ternary definitions for RUCC and NCHS/UIC. Specifically, the NCHS/UIC and RUCC revealed similar estimates comparing micropolitan and metropolitan, but the RUCC definition yielded larger differences between the rural or noncore and metropolitan populations. Although the UIC was not separately evaluated given the overlap with RUCC and NCHS, it is important for researchers and policy makers to consider the makeup of the UIC rural categories. The NCHS was specifically designed to study health outcomes that vary with rurality. It focuses on delineating metropolitan areas and therefore may have greater ability to identify important health differences within metropolitan areas. Alternatively, the RUCC’s focus is on delineating the most rural groups from urban populations, which may lead to larger effect sizes among the rural groups compared with the NCHS in this study. Although Schiefelbein et al. (14) did not examine mortality, the authors found that binary RUCC, UIC, and NCHS designations largely identified similar populations in the Midwest and Wisconsin. This study extends these findings with the similar hazard of death among rural populations identified by the binary definitions. Further, prior studies suggest that the continuous measures of rurality may provide a better representation of the variation of rurality (24,25).
An additional novel component is the comparison of differences within cancer groupings with known rural disparities in cancer survival. The largest differences in survival among the definitions were seen in the screening-related cancers, with hazard ratios from models using the RUCC definition being 1% to 13% higher than the others. Overall, the ternary definitions yielded similar estimates of the difference between urban or metropolitan and micropolitan, but the RUCC definition demonstrated 35% increased cancer mortality hazard for the micropolitan designation, followed by NCHS (22%) and IRR, which was not statistically significant. These large discrepancies may be partially attributable to the greater hazards of mortality associated with rurality among screening-related group. Rural populations have substantially lower cancer screening rates compared with urban populations (26,27). Additionally, rural patients are at greater risk for late-stage diagnosis, potentially due to low physician density and greater distance to screening facilities compared with urban areas (27-29). Interventions should be focused on or developed to increase access to care among rural populations. In comparison, obesity-related groups consist of more cancers with high mortality rates (1) (eg, pancreas, liver, esophagus); therefore, the effect of rurality may be diminished. Larger urban and rural differences in hazard of mortality in the screening-related cancer group may be driven by urban and rural differences in cancer screenings, and smaller differences in the obesity-related group may be due to the cancer no-screening recommendations. Prior works demonstrated urban and rural differences in cancer screening (26) and mortality (30), whereas this study identifies a gap in knowledge of whether lower rates of cancer screening among rural populations directly contribute to larger urban and rural differences in survival compared with nonscreening-related cancers.
Although this study presents the first comparison of survival estimates across various definitions of rurality for patients with cancer, several limitations should be considered. First, we acknowledge there are additional patient-level (eg, comorbidity, socioeconomic status) factors associated with cancer mortality (5). However, the purpose of this study was to provide estimates controlling for readily available factors that will allow others to estimate potential differences in their own work. Census tract and zip code–level data may reveal further disparities in survival. However, county-level data are often readily accessible to researchers given privacy rules and are widely used in research. Finally, SEER data have been found to underrepresent rural populations (31). However, SEER includes detailed sociodemographic and clinical characteristics and is more readily available compared with other datasets (32,33).
The hazard of mortality was largely consistent across multiple definitions in the binary form. Differences in survival emerged when definitions were categorized into 3 levels or considered as a continuous variable. Additionally, screening-related cancers were most subject to these differences. Findings are meant to provide baseline comparisons; there is no “correct” definition of rurality because the rationale behind each individual definition or categorization has a specific use. We recommend that if there is an option of rurality definition, care should be taken to identify the definition meeting the needs of the study (eg, NCHS emphasizes urban environments). Future work should examine more granular categorizations and evaluate the screening-related cancer cohort in further detail. Current findings may be useful to researchers and policy makers in selecting definitions for conducting analyses, comparing across studies, and creating policies that affect rural populations.
Supplementary Material
Contributor Information
Jeffrey A Franks, Division of Hematology and Oncology, University of Alabama at Birmingham, Birmingham, AL, USA.
Elizabeth S Davis, Department of Surgery, Boston University, Boston, MA, USA.
Smita Bhatia, Division of Pediatric Hematology, Oncology and Blood or Marrow Transplant, Department of Pediatrics, University of Alabama at Birmingham, Birmingham, AL, USA; Institute for Cancer Outcomes and Survivorship, University of Alabama at Birmingham, Birmingham, AL, USA.
Kelly M Kenzik, Department of Surgery, Boston University, Boston, MA, USA; Slone Epidemiology Center, Boston University, Boston, MA, USA.
Data availability
All datasets were derived from sources in the public domain: https://seer.cancer.gov/seerstat/https://www.ers.usda.gov/data-products/rural-urban-continuum-codes.aspxhttps://www.ers.usda.gov/data-products/urban-influence-codes/https://www.cdc.gov/nchs/data_access/urban_rural.htmhttps://purr.purdue.edu/publications/2960/1.
Author contributions
Jeffrey A. Franks, MSPH (formal analysis; writing—original draft; writing—review and editing); Elizabeth S. Davis, MSPH (formal analysis; methodology; writing—original draft; writing—review and editing); Smita Bhatia, MD MS (conceptualization; methodology; writing—original draft; writing—review and editing); Kelly M. Kenzik, PhD (conceptualization; formal analysis; funding acquisition; methodology; supervision; writing—original draft; writing—review and editing).
Funding
Research reported in this publication was supported by the National Cancer Institute of the National Institutes of Health under Award Number R37CA266193 (PI: Kenzik). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Conflicts of interest
None.
Acknowledgements
The funder did not play a role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to submit the manuscript for publication.
References
- 1. Miller KD, Nogueira L, Devasia T, et al. Cancer treatment and survivorship statistics, 2022. CA Cancer J Clin. 2022;72(5):409-436. doi: 10.3322/caac.21731. [DOI] [PubMed] [Google Scholar]
- 2. Alcaraz KI, Wiedt TL, Daniels EC, Yabroff KR, Guerra CE, Wender RC.. Understanding and addressing social determinants to advance cancer health equity in the United States: a blueprint for practice, research, and policy. CA Cancer J Clin. 2020;70(1):31-46. doi: 10.3322/caac.21586. [DOI] [PubMed] [Google Scholar]
- 3. Han X, Yabroff KR, Ward E, Brawley OW, Jemal A.. Comparison of insurance status and diagnosis stage among patients with newly diagnosed cancer before vs after implementation of the patient protection and affordable care act. JAMA Oncol. 2018;4(12):1713-1720. doi: 10.1001/jamaoncol.2018.3467. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Henley SJ, Jemal A.. PMC7076627; rural cancer control: bridging the chasm in geographic health inequity. Cancer Epidemiol Biomarkers Prev. 2018;27(11):1248-1251. doi: 10.1158/1055-9965.EPI-18-0807. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Bhatia S, Landier W, Paskett ED, et al. Rural-urban disparities in cancer outcomes: opportunities for future research. J Natl Cancer Inst. 2022;114(7):940-952. doi: 10.1093/jnci/djac030. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Ghazarian AA, Martin DN, Lam TK.. PMC6219612; opportunities and challenges in rural cancer research: an epidemiologic perspective. Cancer Epidemiol Biomarkers Prev. 2018;27(11):1245-1247. doi: 10.1158/1055-9965.Epi-18-0962. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Levit LA, Byatt L, Lyss AP, et al. Closing the rural cancer care gap: three institutional approaches. J Clin Oncol Oncol Pract. 2020;16(7):422-430. doi: 10.1200/OP.20.00174. [DOI] [PubMed] [Google Scholar]
- 8. Health Resources and Services Administration. Defining rural population. https://www.hrsa.gov/rural-health/about-us/what-is-rural. Updated 2022. Accessed July 10, 2022.
- 9. Economic Research Service. Rural-urban continuum codes. 2013. https://www.ers.usda.gov/data-products/rural-urban-continuum-codes/. Updated 2020. Accessed May 20, 2022.
- 10. Economic Research Service. Urban influence codes. 2013. https://www.ers.usda.gov/data-products/urban-influence-codes/. Updated 2013. Accessed May 20, 2022.
- 11. National Center for Health Statistics. Urban rural classification scheme for counties. 2017. https://www.cdc.gov/nchs/data_access/urban_rural.htm. Accessed August 10, 2022.
- 12. Waldorf B, Kim A. The Index of Relative Rurality (IRR): US county data for 2000 and 2010. 2018. https://purr.purdue.edu/publications/2960/1. doi: 10.4231/R7959FS8. Accessed May 20, 2022. [DOI]
- 13. Moss JL, Stinchcomb DG, Yu M.. Providing higher resolution indicators of rurality in the Surveillance, Epidemiology, and End Results (SEER) database: implications for patient privacy and research. Cancer Epidemiol Biomarkers Prev. 2019;28(9):1409-1416. https://www.ncbi.nlm.nih.gov/pubmed/31201223. doi: 10.1158/1055-9965.EPI-19-0021. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Schiefelbein AM, Taylor AK, Krebsbach JK, et al. Same people, different results: categorizing cancer registry cases across the rural-urban continuum [Preprint published online December 29, 2021]. Res Sq. doi: 10.21203/rs.3.rs-1200114/v1. Accessed August 11, 2022. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Kurani SS, McCoy RG, Lampman MA, et al. Association of neighborhood measures of social determinants of health with breast, cervical, and colorectal cancer screening rates in the US Midwest. JAMA Netw Open. 2020;3(3):e200618. doi: 10.1001/jamanetworkopen.2020.0618. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Steele CB, Thomas CC, Henley SJ, et al. Vital signs: trends in incidence of cancers associated with overweight and obesity - United States, 2005-2014. MMWR Morb Mortal Wkly Rep. 2017;66(39):1052-1058. doi: 10.15585/mmwr.mm6639e1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Lundeen EA, Park S, Pan L, O'Toole T, Matthews K, Blanck HM.. Obesity prevalence among adults living in metropolitan and nonmetropolitan counties - United States, 2016. MMWR Morb Mortal Wkly Rep. 2018;67(23):653-658. doi: 10.15585/mmwr.mm6723a1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Doogan NJ, Roberts ME, Wewers ME, et al. A growing geographic disparity: rural and urban cigarette smoking trends in the United States. Prev Med. 2017;104:79-85. doi: 10.1016/j.ypmed.2017.03.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Buettner-Schmidt K, Miller DR, Maack B.. Disparities in rural tobacco use, smoke-free policies, and tobacco taxes. West J Nurs Res. 2019;41(8):1184-1202. doi: 10.1177/0193945919828061. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Waldorf B, Kim A.. Defining and Measuring Rurality in the US: From Typologies to Continuous Indices. Department of Agricultural Economics, Purdue University; 2015.
- 21. Mozumder SI, Rutherford MJ, Lambert PC.. Stpm2cr: a flexible parametric competing risks model using a direct likelihood approach for the cause-specific cumulative incidence function. Stata J. 2017;17(2):462-489. [PMC free article] [PubMed] [Google Scholar]
- 22. SAS Institute. SAS 9.4. SAS documentation. 2013;9.4.
- 23. StataCorp. Stata Statistical Software: Release 16. Updated 2019.
- 24. Hall SA, Kaufman JS, Ricketts TC.. Defining urban and rural areas in U.S. epidemiologic studies. J Urban Health. 2006;83(2):162-175. doi: 10.1007/s11524-005-9016-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. McLafferty S, Wang F.. Rural reversal? Rural-urban disparities in late-stage cancer risk in Illinois. Cancer. 2009;115(12):2755-2764. doi: 10.1002/cncr.24306. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Tran L, Tran P.. US urban-rural disparities in breast cancer-screening practices at the national, regional, and state level, 2012-2016. Cancer Causes Control. 2019;30(10):1045-1055. doi: 10.1007/s10552-019-01217-8. [DOI] [PubMed] [Google Scholar]
- 27. Chandak A, Nayar P, Lin G.. Rural-urban disparities in access to breast cancer screening: a spatial clustering analysis. J Rural Health. 2019;35(2):229-235. doi: 10.1111/jrh.12308. [DOI] [PubMed] [Google Scholar]
- 28. Eberth JM, Zahnd WE, Josey MJ, Schootman M, Hung P, Probst JC.. Trends in spatial access to colonoscopy in South Carolina, 2000-2014. Spat Spatiotemporal Epidemiol. 2021;37:100414. doi: 10.1016/j.sste.2021.100414. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Caldwell JT, Ford CL, Wallace SP, Wang MC, Takahashi LM.. Intersection of living in a rural versus urban area and race/ethnicity in explaining access to health care in the United States. Am J Public Health. 2016;106(8):1463-1469. doi: 10.2105/AJPH.2016.303212. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Kenzik KM, Rocque GB, Landier W, Bhatia S.. Urban versus rural residence and outcomes in older patients with breast cancer. Cancer Epidemiol Biomarkers Prev. 2020;29(7):1313-1320. doi: 10.1158/1055-9965.Epi-19-1414. [DOI] [PubMed] [Google Scholar]
- 31. National Cancer Institute. SEER Linked Databases. Surveillance, Epidemiology, and End Results Program. https://seer.cancer.gov/data-software/linked_databases.html. Updated 2022. Accessed May 20, 2022.
- 32. Zahnd WE, Jenkins WD, James AS, et al. Utility and generalizability of multistate, population-based cancer registry data for rural cancer surveillance research in the United States. Cancer Epidemiol Biomarkers Prev. 2018;27(11):1252-1260. doi: 10.1158/1055-9965.EPI-17-1087. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Duggan MA, Anderson WF, Altekruse S, Penberthy L, Sherman ME.. The Surveillance, Epidemiology, and End Results (SEER) program and pathology: toward strengthening the critical relationship. Am J Surg Pathol. 2016;40(12):e94-e102. doi: 10.1097/PAS.0000000000000749. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
All datasets were derived from sources in the public domain: https://seer.cancer.gov/seerstat/https://www.ers.usda.gov/data-products/rural-urban-continuum-codes.aspxhttps://www.ers.usda.gov/data-products/urban-influence-codes/https://www.cdc.gov/nchs/data_access/urban_rural.htmhttps://purr.purdue.edu/publications/2960/1.