Beyond composite measures of regional vulnerability: Rural–urban colorectal cancer mortality disparities mediated by area-level characteristics

Abstract

Background:

Rural–urban disparities in colorectal cancer (CRC) mortality have been attributed to individual- and population-level factors. Population-level composite measures such as the Social Vulnerability Index (SVI) have been used to investigate rural–urban disparities. However, the contribution of each SVI component to CRC mortality disparities among rural and urban counties is unknown.

Methods:

Via CRC mortality data for all US counties from 1999 to 2020 from the Centers for Disease Control and Prevention, rural–urban differences in county demographics and in 14 SVI factors were examined. Four-way effect decomposition was used to estimate the proportion of the rural–urban CRC mortality disparity mediated by the SVI.

Results:

Among 2927 counties (rural, 60.7%; urban, 39.3%), rural counties had 11.8% higher CRC mortality than urban counties; 18.6% of that disparity was mediated by low socioeconomic status (SES), 8.8% by household characteristics, and 2.7% by racial/ethnic minority status. Among all counties, poverty (incidence rate ratio [IRR], 3.2; 95% confidence interval [CI], 2.8–3.5), unemployment (IRR, 5.4; 95% CI, 4.4–6.8), lacking a high school diploma (IRR, 2.7; 95% CI, 2.4–2.9), household crowding (IRR, 3.3; 95% CI, 2.5–4.4), and lacking a vehicle (IRR, 3.4; 95% CI, 3.0–3.8) had the greatest impact on CRC mortality. Compared to urban counties, rural counties with a higher proportion of people without a vehicle had a higher risk of CRC mortality (IRR, 4.3; 95% CI, 3.7–5.0; vs. IRR, 1.9; 95% CI, 1.5–2.4).

Conclusions:

The rural–urban CRC mortality disparity is largely driven by low SES—higher poverty and unemployment, and lower income and education. The relationship between CRC mortality and the SVI is nuanced, and evaluating each component of the SVI may allow for more targeted area-level interventions than evaluating the SVI alone.

INTRODUCTION

Colorectal cancer (CRC) is the second most common cause of cancer-related death in the United States. Social determinants of health (SDoHs), the conditions under which people live and work, contribute to CRC mortality at both the individual and population levels.^1,2 Rural and urban communities each face unique disparities attributable to SDoHs but rural populations experience higher CRC incidence and mortality.^3–5 From 2016 to 2020, CRC mortality among rural populations was 21%–23% higher than among urban populations.⁶ There are several SDoHs that contribute to the CRC mortality disparity between rural and urban communities, including lower screening rates, transportation challenges, limited provider availability, and underinsurance among rural populations.^4,7–12 Additionally, individual-level CRC risk factors including obesity, smoking, and alcohol consumption are higher among rural populations.^13,14

Identifying and characterizing geographic areas with high CRC mortality have implications beyond a descriptive purpose. Because federal and state funding is designated at the county level, informed geographic distribution of risk may better focus interventions and resources. Methods to characterize regions with high CRC mortality risk include the use of composite area-level measures of disadvantage.^15–18 One frequently used measure in CRC research is the Social Vulnerability Index (SVI), which is composed of 14 SDoHs. The SVI was originally developed as a measure of regional ability to withstand a natural disaster but has since been adapted for use in health outcomes research.^19–22 Prior studies have associated a worse SVI with emergent presentations of CRC, a lower likelihood of guideline-concordant care, and higher mortality after CRC resection.^23–25 However, because composite measures such as the SVI are multifactorial, insight into the specific, intervenable area-level CRC mortality risk factors that exist within the SVI may be limited.

Potentially modifiable SDoHs within the SVI include having a vehicle, household crowding, educational attainment, unemployment, and income, among several others. Some SDoHs of the SVI, such as the racial and ethnic makeup of a county and county-level education, have been shown to be associated with CRC mortality. However, the relationships between other SDoHs within the SVI, such as vehicle availability and household crowding, and CRC mortality are not known. Furthermore, mediation by county-level SDoHs of the rural–urban CRC mortality disparity has not been studied. We evaluated the contributions of the 14 SVI SDoHs (subsequently referred to as “SVI components”) to CRC mortality differences between rural and urban populations and the extent of mediation by SVI components of the rural–urban CRC mortality disparity. We hypothesized that components such as income and having a vehicle would affect CRC mortality in rural and urban populations differently, and that the SVI would mediate the relationship between CRC mortality and rurality.

MATERIALS AND METHODS

Data sources

We obtained pooled colon and rectal cancer mortality data across all US counties from 1999 to 2020 from the Centers for Disease Control and Prevention Wide-Ranging Online Data for Epidemiologic Research (CDC WONDER) database.²⁶ Mortality rates per 100,000 were age adjusted (age 15 years and older), and counties with missing mortality data were excluded. County-level demographic data including age, race, language, education, employment, and income were obtained from the US Census Bureau.²⁷ This study used publicly available data, and thus did not require institutional review board approval.

SVI

To create the SVI, the CDC obtained data from the US Census Bureau (Census 2010 data) and the American Community Survey (2006–2010 data), and subsequently developed these data into a composite measure of vulnerability representing the relative vulnerability of a county relative to other counties.²⁷ The SVI is a composite-level measure composed of the following 14 components within four domains; socioeconomic status: population living below the poverty line, unemployment, per capita income, and individuals without a high school diploma; household characteristics: individuals aged 65 years and older, individuals aged 17 years and younger, and single-parent households; racial and ethnic composition: racial and ethnic minority individuals (non-White) and those with limited English proficiency (self-reported as speaking English “less than well”); and housing and transportation: housing density (more than 10 units in a residential building), mobile homes, household crowding (more individuals than rooms), lacking a vehicle, and group home living.²⁸ These figures are reported as percentiles relative to other counties; higher percentiles correspond to worse regional disadvantage.²⁹ SVI domains and factors are noted in Table 1. SVI data were linked with county mortality data via Federal Information Processing System codes. We used SVI data from 2010 because it represented a midpoint in our mortality data window from 1999 to 2020. Our exposures of interest included the overall SVI, SVI domains, and 14 SVI components.

TABLE 1.

County characteristics and social vulnerability percentiles by rural–urban county status.

Characteristic	All counties (N = 2927), median % (IQR)	Rural counties (n = 1776; 60.7%), median % (IQR)	Urban counties (n = 1151; 39.3%), median % (IQR)	p
CRC mortality rate per 100,000	26.3 (23.4–29.8)	27.6 (24.5–31)	24.7 (22.1–27.4)	<.0001
Sociodemographic factors
Age 65+ years	15.4 (13.1–17.9)	16.6 (14.6–19.2)	13.6 (11.5–15.6)	<.0001
Non-White race	14.4 (5.9–33.0)	11.1 (4.8–30.9)	19.8 (9.4–34.8)	<.0001
Non-English speaking	0.8 (0.3–2.0)	0.6 (0.3–1.6)	1.2 (0.5–2.5)	<.0001
Without a high school diploma	15.5 (11.6–21.6)	17.1 (12.3–23.6)	13.7 (10.3–18.6)	<.0001
Unemployment	7.4 (5.6–9.3)	7.5 (5.4–9.6)	7.3 (6.0–8.9)	.9987
Income, US $	21,728 (18,974–24,776)	20,439 (18,150–22,812)	24,311 (21,533–27,727)	<.0001
Percentile rankings for SVI components and by domaina
Overall SVI (composite of all domains)	51.6 (27.3–75.8)	54.9 (29.4–79.4)	47.8 (24.6–70.0)	<.0001
Socioeconomic status	51.6 (26.3–75.9)	61.0 (33.7–81.8)	38.9 (17.8–61.5)	<.0001
Below the poverty line	51.1 (25.7–75.5)	59.0 (33.1–81.4)	38.4 (16.3–63.0)	<.0001
Unemployed	51.9 (28.1–75.5)	52.6 (24.9–78.5)	51.2 (32.9–71.6)	.9987
Income	50.4 (25.2–75.4)	63.0 (40.7–82.3)	28.2 (12.4–52.6)	<.0001
No high school diploma	50.7 (26.2–75.1)	58.2 (31.9–80.8)	40.3 (12.1–63.7)	<.0001
Household characteristics	51.4 (26.4–75.7)	53.9 (27.3–79.8)	48.9 (25.2–70.6)	<.0001
Age 65+ years	48.7 (24.4–72.8)	61.8 (29.5–80.7)	28.9 (13.2–50.4)	<.0001
Age ≤17 years	50.8 (25.8–74.8)	45.1 (22.2–70.6)	58.8 (34.5–79.8)	<.0001
Single-parent households	52.0 (27.9–75.8)	47.7 (24.1–73.5)	57.9 (34.9–78.3)	<.0001
Racial and ethnic minority status	50.6 (26.0–74.9)	42.5 (20.5–68.5)	62.4 (38.2–81.2)	<.0001
Racial or ethnic minority	50.4 (25.6–75.0)	42.5 (19.7–72.7)	59.2 (28.1–77.3)	<.0001
Limited English proficiency	50.6 (26.3–74.8)	42.5 (20.5–68.7)	61.9 (37.7–79.9)	<.0001
Housing type and transportation	52.0 (27.6–75.9)	51.1 (28.7–75.7)	53.9 (25.8–76.0)	.5238
Living in dense housing	52.3 (27.6–76.3)	42.8 (22.6–63.9)	72.5 (42.5–89.1)	<.0001
Living in a mobile home	49.9 (24.3–75.5)	58.2 (34.9–79.5)	33.5 (12.8–64.7)	<.0001
Household crowding	50.0 (26.0–74.6)	50.9 (25.4–76.3)	48.7 (26.7–71.8)	.6748
Lacking a vehicle	51.7 (27.7–75.5)	53.8 (30.0–76.6)	47.9 (24.7–64.0)	.0004
Living in a group home	50.9 (26.5–75.8)	50.8 (23.3–72.3)	50.9 (28.6–78.3)	.0005

Abbreviations: CRC, colorectal cancer; IQR, interquartile range; SVI, Social Vulnerability Index.

^aHigher SVI percentile indicates more regional disadvantage.

Rural–urban exposure

Counties were categorized as rural or urban on the basis of US Department of Agriculture Rural–Urban Continuum Codes (RUCCs); rurality increases as the RUCC increases, such that a lower RUCC corresponds to a more populous, metropolitan county, whereas a higher RUCC corresponds to a more rural county. RUCCs 1–3 were categorized as urban, and RUCCs 4–9 were categorized as rural.^30,31

Statistical analysis

County-level sociodemographic data, including the proportion of the county aged 65 years and older, of non-White race, non-English speaking, and without a high school diploma, unemployment, income, CRC mortality rate, and SVI percentiles, were analyzed with Wilcoxon rank sum tests. The effects of the SVI on CRC mortality per 100,000 were assessed via univariate and multivariable Poisson regression models with a log link. Results are reported as incidence rate ratios (IRRs) and 95% confidence intervals (CIs). Effect estimates represent incident CRC mortality for each one-unit increase in percentile rankings. All models were stratified by rural–urban status. All SVI components were modeled as continuous variables.^32,33 Each SVI value incorporated into our models represented the proportion of the county experiencing that specific SDoH.

We separately examined the effect of SVI components on CRC mortality, including age (proportion of the county aged 65 years and older), race (proportion of racial minorities in the county), education (proportion of the county without a high school diploma), and income (median annual household income of the county), via regression models. For example, in this analysis, the effect estimate for single-parent households includes age, race, education, and income as covariates.

Effect decomposition

The proportion of the mortality difference between rural and urban counties mediated by specific SVI domains was calculated via four-way effect decomposition in SAS, version 9.4. Causal assumptions for the effect decomposition include (1) conditional exchangeability, (2) positivity, (3) consistency, and (4) noninterference. These assumptions were met. The decomposition provided the following estimates: (1) controlled direct effect (proportion of the rural–urban mortality difference not mediated by the SVI), (2) reference interaction (interaction between rurality and the SVI but not mediation), (3) mediated interaction (effect due to both mediation and interaction), and (4) pure indirect effect (proportion of the rural–urban mortality difference attributable to the SVI alone). The pure indirect effect most closely resembles a causal mediation pathway. Consistent with prior works, we categorized each SVI domain into high (>80th percentile) and low (<80th percentile), and the SVI represented the mediator.^32,33 Covariates included the other SVI domains; for example, when modeling the relationship between domain 1 and CRC mortality, domains 2–4 were included as covariates. Effect decomposition models identified no meaningful difference when including provider densities (primary care providers, gastroenterologists, general surgeons, and colorectal surgeons per 100,000 population) as a covariate, which resulted in the exclusion of this covariate from the final effect decomposition models. All analyses were conducted with SAS, version 9.4 (Cary, North Carolina).

RESULTS

County-level characteristics

Among the 3147 counties included in the CDC WONDER database, 215 (6.8%) were excluded because of missing CRC mortality data, four (0.1%) were excluded because of missing RUCCs, and one was excluded because of missing provider density data. Among the 2927 counties included in the analysis (rural: n = 1776; 60.7%; urban: n = 1151; 39.3%), rural populations were older (16.6% vs. 13.6% aged 65+ years), less educated (17.1% vs. 13.7% without a high school diploma), had a lower median annual household income ($20,439 vs. $24,311), and had higher CRC mortality than urban counties (27.6 vs. 24.7 per 100,000). Urban counties had higher proportions of non-White (19.8% vs. 11.1%) and non–English-speaking individuals (1.2% vs. 0.6%). Unemployment did not differ on the basis of rurality. Rural counties had a higher (worse) SVI overall and for socioeconomic status (SES) (i.e., lower SES) and household characteristics. Urban counties had a higher SVI for racial and ethnic minority status. The SVI for housing type and transportation did not differ across rural and urban counties (Table 1; Figure 1).

FIGURE 1.

Unadjusted linear regression for colorectal cancer mortality and SVI domains among all counties and stratified by rural and urban counties. SVI indicates Social Vulnerability Index.

CRC mortality and the SVI among all counties

In multivariable analyses, all counties had a higher CRC mortality risk with a higher SVI for SES (IRR, 1.32; 95% CI, 1.29–1.35), household characteristics (IRR, 1.25; 95% CI, 1.22–1.28), and housing type and transportation (IRR, 1.1; 95% CI, 1.07–1.12) and a lower CRC mortality risk with a higher SVI for racial and ethnic minority status (IRR, 0.91; 95% CI, 0.89–0.93) (Figure 1). SVI components that were associated with higher CRC mortality included living in poverty, areas with high unemployment, lower income, low education, more individuals aged 65 years and older and 17 years and younger, a higher proportion of single-parent households, a greater proportion reporting racial minority status, a higher density of mobile homes and group homes, household crowding, and lacking a vehicle. Effect estimates and 95% CIs are presented in Table 2 and can be interpreted as follows, with the SVI component of unemployment as an example: each one-unit increase in the percentile ranking of unemployment was associated with a 4.62-fold higher incidence of CRC mortality among rural counties and a 7.1-fold higher incidence of CRC mortality among urban counties. Effect estimates adjusted for county-level age, race, education, and income are presented in Table S2. As compared to unadjusted models, adjusting for the aforementioned covariates generally attenuated effect estimates (i.e., poverty, unemployment, lacking a high school diploma, living in dense housing, and living in a group home), although some remained unchanged (income), and some effect estimates increased in magnitude after the inclusion of covariates (single-parent households and limited English proficiency) (Table 2; Table S2).

TABLE 2.

Poisson linear regression results for colorectal cancer mortality and SDoHs comprising the SVI, overall and stratified by rural and urban counties.

SVI domains and components	All counties, IRR (95% CI)	Rural counties, IRR (95% CI)	Urban counties, IRR (95% CI)
Overall SVI	1.23 (1.2–1.26)	1.21 (1.17–1.25)	1.19 (1.14–1.25)
Socioeconomic status	1.32 (1.29–1.35)	1.26 (1.22–1.3)	1.31 (1.25–1.37)
Below the poverty line	3.17 (2.84–3.53)	2.62 (2.3–2.99)	2.97 (2.39–3.69)
Unemployed	5.44 (4.37–6.77)	4.62 (3.62–5.89)	7.1 (4.38–11.5)
Income	0.84 (0.83–0.85)	0.83 (0.81–0.85)	0.88 (0.86–0.90)
No high school diploma	2.66 (2.42–2.92)	2.18 (1.94–2.45)	2.8 (2.34–3.36)
Household characteristics	1.25 (1.22–1.28)	1.22 (1.19–1.26)	1.23 (1.18–1.29)
Age 65+ years	1.42 (1.2–1.69)	0.57 (0.45–0.72)	1.7 (1.23–2.36)
Age ≤17 years	1.61 (1.3–2)	3.13 (2.4–4.09)	0.83 (0.57–1.21)
Single-parent households	7.59 (6.2–9.3)	7.94 (6.29–10)	9.26 (6.16–13.9)
Racial and ethnic minority status	0.91 (0.89–0.93)	0.99 (0.96–1.02)	0.87 (0.84–0.91)
Racial or ethnic minority	1.12 (1.08–1.16)	1.23 (1.18–1.28)	1.02 (0.95–1.08)
Limited English proficiencya	0.21 (0.16–0.28)	0.34 (0.24–0.48)	0.21 (0.14–0.31)
Housing type and transportation	1.1 (1.07–1.12)	1.1 (1.07–1.14)	1.07 (1.03–1.12)
Living in dense housingb	0.32 (0.28–0.37)	0.18 (0.13–0.25)	0.54 (0.45–0.63)
Living in a mobile home	1.46 (1.36–1.58)	1.18 (1.07–1.3)	1.51 (1.33–1.71)
Household crowding	3.31 (2.52–4.35)	4.4 (3.3–5.86)	0.51 (0.26–1.01)
Lacking a vehicle	3.36 (2.97–3.8)	4.31 (3.72–4.99)	1.92 (1.53–2.4)
Living in a group home	1.42 (1.21–1.68)	1.27 (1.05–1.52)	1.18 (0.82–1.7)

Note: The SDoHs within the SVI (SVI components) were modeled as the proportions of the county experiencing a given SDoH. All variables were modeled as continuous.

Abbreviations: CI, confidence interval; IRR, incidence rate ratio; SDoH, social determinant of health; SVI, Social Vulnerability Index.

^aProportion of individuals speaking English “less than well” in a county.

^bDefined as >10 units in a residential building.

Two components, a higher percentage of the county speaking English less than well (IRR, 0.21; 95% CI, 0.16–0.28) and a higher density of multiunit housing structures (IRR, 0.32; 95% CI, 0.28–0.37), were associated with a decreased CRC mortality rate (Table 2).

CRC mortality and the SVI by rural–urban status

In multivariable analyses stratified by rural–urban status, a larger population of individuals aged 65 years and older was associated with increased CRC mortality among urban populations (IRR, 1.7; 95% CI, 1.23–2.36) but decreased mortality among rural populations (IRR, 0.57; 95% CI, 0.45–0.72), whereas more individuals aged 17 years and younger was associated with increased mortality among rural populations (IRR, 3.13; 95% CI, 2.4–4.09) but was not significant in the urban setting (IRR, 0.83; 95% CI, 0.57–1.21). A higher proportion of racial and ethnic minorities was associated with increased mortality in rural populations (IRR, 1.23; 95% CI, 1.18–1.28) but was not significant in urban counties (IRR, 1.02; 95% CI, 0.95–1.08). Household crowding was associated with higher mortality in rural (IRR, 4.4; 95% CI, 3.3–5.86) but not urban counties (IRR, 0.51; 95% CI, 0.26–1.01).

Although associated with higher mortality among all counties, lacking a vehicle was associated with much higher CRC mortality among rural counties (IRR, 4.31 vs. 1.92). Alternatively, unemployment, lacking a high school diploma, and single-parent households were associated with much higher CRC mortality in urban populations (IRR, 7.1 vs. 4.62; IRR, 2.8 vs. 2.18; and IRR, 9.26 vs. 7.94; respectively, for urban vs. rural) (Table 2).

Effect decomposition

Compared to urban counties, rural populations had 11.78% higher CRC mortality. Each domain of the SVI mediated the association between rurality and CRC mortality with different magnitudes. The overall SVI mediated 8.82% of the rural–urban difference in CRC mortality, SES mediated 16.27%, household characteristics mediated 8.12%, racial and ethnic minority status mediated 11.09%, and housing type and transportation mediated −0.53% of the rural–urban CRC mortality difference (Table 3; Table S1).

TABLE 3.

Effect decomposition of the higher colorectal cancer mortality among rural populations as mediated by the SVI and SVI domains.

SVI category	Mediated, %a	Controlled direct effect, %b	Mediated interaction, %c	Pure indirect effect, %d
Overall SVI	8.82	89.82	0.85	7.96
Socioeconomic status	16.27	83.02	1.22	15.05
Household characteristics	8.12	91.00	0.61	7.51
Racial and ethnic minority status	11.09	90.46	0.83	10.26
Housing type and transportation	−0.53	100.1	−0.04	−0.49

Note: Covariates for each model included the other SVI domains.

Abbreviation: SVI, Social Vulnerability Index.

^aPercent mediated indicates the percentage of the relationship between rurality and colorectal cancer mortality mediated by the SVI.

^bPercent controlled direct effect indicates the percentage of the rural–urban mortality difference not mediated by the SVI.

^cPercent mediated interaction indicates the percentage of the interaction between rurality and the SVI not via a mediation pathway.

^dPercent pure indirect effect indicates the percentage of the rural–urban mortality difference attributable to the SVI alone.

DISCUSSION

In this population-level study of rural–urban disparities in CRC mortality, county-level social determinants mediated the association between rurality and mortality, with the greatest mediation by socioeconomic factors including poverty, education, and unemployment.

This is the first study to investigate the relationship between CRC mortality and the 14 components that comprise the composite SVI, as well as the effects of each SVI category on rural–urban disparities in CRC mortality. Several studies investigating the role of neighborhood characteristics and CRC mortality have used the overall SVI to represent regional disadvantage, and have demonstrated that a higher SVI is associated with worse CRC mortality and worse survival after surgical resection.^{19,22,25,34,35} However, the composite nature of the SVI score does not provide insight into specific disparities that could be targeted for intervention. The present study identifies several specific SDoHs involved in the rural–urban CRC mortality disparity.

Factors associated with increased mortality among both rural and urban counties

Although studies have used poverty, education, income, and employment as proxies for SES, we found that the magnitude of the relationship between unemployment and mortality was greater for urban counties compared to rural counties (IRR, 7.1 vs. 4.6). Unemployed individuals have more limited insurance access, which is associated with presentation at later stages of disease and thus worse survival.³⁶ Because rural populations tend to be older, unemployment may confer retirement benefits, Social Security funds, and Medicare access.³⁷ However, in urban settings with higher concentrations of younger people, unemployment may correspond to underinsurance and a subsequent lack of colon cancer screening. This is supported by findings that counties with higher unemployment have higher rates of late-stage CRC presentation.³⁶ Additionally, urban areas have a higher cost of living and more housing insecurity, which may also contribute to the impact of unemployment on CRC mortality in urban compared to rural areas.^38,39

Low income, another aspect of SES, has been associated with increased CRC incidence and mortality, a relationship that strengthens with increasing county rurality.⁴⁰ Although we too found that income was associated with mortality, the effects were similar among rural and urban counties (IRR, 0.83 vs. 0.88 with increasing income). In 2010, the poverty line for a one-person household was $10,830, which is significantly lower than the median per capita income among our counties of $21,728. This may explain why poverty was associated with high CRC mortality risk among both rural and urban counties to a greater degree than income (IRR, 3.17).⁴¹ County-level poverty has been associated with presentation at later stages of disease as well as higher CRC mortality risk in prior studies.⁴²

Whereas low income increased CRC mortality risk, low educational attainment increased CRC mortality risk to an even greater degree among both urban and rural counties. We found that counties with higher proportions of individuals without a high school diploma had double the CRC mortality risk for both urban and rural counties, although the risk was slightly higher among urban counties (IRR, 2.8 vs. 2.2). Across measures of educational attainment, lower education at both the individual and population levels has been associated with higher CRC mortality.^43,44 Weir et al. estimated that one in five CRC deaths in counties with low educational attainment was potentially avoidable when compared to high–educational attainment counties.⁴⁴ Education may play a multifactorial role in a community’s economic and social development, and a low countywide level of education may lead to lower health literacy and hinder health care access, infrastructure, and development. Promoting education in such counties may have the downstream effect of reducing CRC mortality. Although effect sizes were greater for unemployment, education, and single-parent households among urban counties, these SVI components remain associated with high mortality risk among rural counties, which indicates a need for intervention in all counties affected by these SVI components.

A higher proportion of single-parent households was associated with higher CRC mortality among all counties. Single parents have been found to have worse self-rated health, more medical problems, more hospital admissions, and higher mortality than partnered parents long term.^45–47 The increased mortality risk experienced by single parents is thought to be related to higher stress, increased social isolation, and fewer household resources, including lower household income.^45,46 Counties with more single-parent households may thus have more individuals with psychological, financial, and medical burdens, which may have the spillover effects of reduced CRC screening, presentation at later stages, and subsequently higher mortality. However, the relationship between single-parent households and CRC mortality at both the individual and regional levels warrants future study.

Although lacking access to a vehicle increased CRC mortality risk among all counties, as expected, rural counties were affected more significantly than urban counties (IRR, 4.3 vs. 1.9). Individuals undergoing chemotherapy, radiation, and/or surgery as part of cancer care require reliable transportation, which can be a significant barrier for rural patients with cancer.⁴⁸ At least 4 million individuals in rural counties do not have a vehicle, and transportation insecurity has been associated with decreased cancer screening, declining or delaying cancer care, and missing follow-up appointments after cancer treatment.^49–54 Rural individuals have limited public transportation options and travel farther distances to access National Cancer Institute (NCI) cancer centers, which are more likely to be in urban areas.^55,56 Although NCI centers are associated with better oncologic outcomes and reduced mortality compared to low-volume centers, patients tend to seek treatment at the hospital nearest their home, which increases the likelihood that rural patients receive care at low-volume centers.^57–60 Our finding that lacking a vehicle disproportionately increases CRC mortality risk among rural populations contributes to the understanding of ways in which social determinants affect rural and urban populations differently. Individuals with CRC from rural counties may benefit from transportation insecurity screening and resources such as nonemergency medical transportation.

Factors with different impacts on mortality between rural and urban counties

By uncovering the nuance within each social determinant of the SVI, we were able to identify factors affecting rural and urban counties differently. Among urban counties, a larger population of individuals aged 65 years and older was associated with increased CRC mortality, whereas the opposite was true in rural counties. Alternatively, a larger proportion of individuals aged 17 years and younger, which is potentially representative of younger families in a given area, was associated with increased mortality among rural but not urban populations. Overall, younger people in rural areas and older people in urban areas may be marginalized with respect to health-related resources specific to their age group.

Additionally, increased racial and ethnic diversity was associated with increased mortality in rural populations but was not significant in urban populations. By using a Medicare data set, Ramkumar et al. found that compared to their urban counterparts, Black patients in rural areas had higher odds of CRC mortality, whereas White patients in rural areas had lower odds of mortality.⁶¹ Racial minorities residing in rural areas may experience more social isolation and structural barriers to accessing health care compared to urban communities with larger enclaves of racial/ethnic minorities, which are associated with increased social and family ties and are hypothesized to be protective against health hazards.^62,63 Although regional differences in CRC incidence among racial minorities have been identified across the United States, prior studies have demonstrated that racial/ethnic minorities living in rural areas have worse access to primary care interventions such as cholesterol screening and cervical cancer screening, along with a higher risk of chronic diseases, compared to their urban counterparts.^64–66 This disparity has been attributed to reduced access to care and higher allostatic load among rural minorities for whom health-related resources are more dispersed, as compared to urban racial/ethnic minorities.⁶⁶

Similarly, we found that household crowding was associated with higher CRC mortality in rural but not urban counties. Urban crowding is more common and includes multigenerational living among racial and ethnic minorities, where the household can support older family members diagnosed with CRC.^67,68 In rural settings, where crowding is less common, crowding may be related to poverty rather than cultural reasons.

Factors where a higher SVI was associated with lower mortality, where we found the SVI to “fall short”

A higher density of multiunit housing structures was associated with decreased CRC mortality among all counties. Multiunit housing structures can represent a wide variety of residences, ranging from public housing to luxury high rises in urban centers.⁶⁹ As such, multiunit housing may not represent social vulnerability but may represent urbanity, which we demonstrate portends a survival advantage.

Similarly, a higher percentage of individuals with limited English proficiency was associated with a decreased mortality risk among both rural and urban populations. People who are not fluent in English may be immigrants who were potentially raised outside the United States, and may have thus avoided carcinogenic exposures, habits, and diets of individuals raised in the United States. Two robust studies of intensive care unit patients found decreased mortality and improved outcomes among non–English-speaking patients compared to English-speaking patients.^70,71 Immigrants to the United States have lower rates of chronic diseases and mortality compared to US-born individuals, and increasing length of time since immigrating to the United States is associated with higher rates of obesity, diabetes, hypertension, and smoking.^72–75

Although limited English proficiency and residing in multiunit housing may increase vulnerability in some social and environmental circumstances, their role in CRC mortality is not clear. We demonstrate that a higher SVI does not portend worse CRC mortality across all domains, and examining distinct social determinants provides more nuance and may lead to more tailored interventions specific to a community’s needs.

Limitations

We used the 2010 Social Vulnerability Index (SVI) to minimize temporal incongruity with our mortality series (1999–2020). Subsequent SVI iterations have expanded the composite to include health insurance coverage, civilian disability status, and housing cost burden; additionally, English proficiency has been reclassified from the racial/ethnic minority domain to the household characteristics domain.²⁷ Some marginalized populations, including individuals with limited English proficiency, people with unstable housing, and those with low literacy are generally undercounted in US Census data, and thus may be underrepresented in our data.⁷⁶ Additionally, we excluded 220 counties (7%) because of missing data, although the distribution of rural/urban counties remained the same despite exclusion. We also could not account for patients crossing county lines for treatment, which is not uncommon among rural populations seeking cancer care.

In conclusion, the relationship between CRC mortality and the SDoHs of the SVI is more nuanced than the composite SVI suggests. Rural and urban counties have different needs, and evaluating the contribution of each component of the SVI to CRC mortality may allow for more targeted, effective interventions than by evaluating the SVI alone.

Supplementary Material

Additional supporting information can be found online in the Supporting Information section at the end of this article.

DATA AVAILABILITY STATEMENT

The data are publicly available from the Centers for Disease Control and Prevention Wide-Ranging Online Data for Epidemiologic Research (CDC WONDER) database.