Definition and categorization of rural and assessment of realized access to care

Abstract

Objective

To examine how three measures of realized access to care vary by definitions and categorizations of “rural”.

Data sources

Health Information National Trends Survey (HINTS) data, a nationally representative survey assessing knowledge of health‐related information, were used. Participants were categorized by county‐based Urban Influence Codes (UICs), Rural–Urban Continuum Codes (RUCCs), and census tract‐based Rural–Urban Commuting Area (RUCAs).

Study design

Three approaches were used across categories of UICs, RUCCs, and RUCAs: (1) non‐metropolitan/metropolitan, (2) three‐group categorization based upon population size, and (3) three‐group categorization based on adjacency to metropolitan areas. Wald Chi‐square tests evaluated differences in sociodemographic variables and three measures of realized access across three of Penchansky's “A's of access” and approaches. The three outcome measures included: having a regular provider (realized availability), self‐reported “excellent” quality of care (realized acceptability), and self‐report of the provider “always” spending enough time with you (provider attentiveness–realized accommodation). The average marginal effects corresponding to each outcome were calculated.

Data collection/extraction methods

N/A

Principal findings

All approaches indicated comparable variation in sociodemographics. In all approaches, RUCA‐based categorizations showed differences in having a regular provider (e.g., 68.9% of non‐metropolitan and 64.4% of metropolitan participants had a regular provider). This association was attenuated in multivariable analyses. No rural–urban differences in quality of care were seen in unadjusted or adjusted analyses regardless of approach. After adjustment for covariates, rural respondents reported greater provider attentiveness in some categorizations of rural compared with urban (e.g., non‐metropolitan respondents reported 6.03 percentage point increase in probability of having an attentive provider [CI = 0.76–11.31%] compared with metropolitan).

Conclusions

Our findings underscore the importance of considering multiple definitions of rural to understand access disparities and suggest that continued research is needed to examine the interplay between potential and realized access. These findings have implications for federal funding, resource allocation, and identifying health disparities.

What is known on this topic

• Conceptualization of rural differs based upon definition, geographic unit, and categorization.

• Different categorizations and definitions have relevance for resource allocation and describing disparities.

What this study adds

• Demographic characteristics of “rural” varied slightly across different definitions and categorizations of rural.

• Perceived provider attentiveness as a measure of realized accommodation was higher in some categorizations of rural compared with urban (e.g., respondents in non‐metropolitan counties reported 6.03 percentage point greater likelihood of an attentive provider).

• Realized access varies by rural definition and is an important area for future research.

1. INTRODUCTION

In the United States, between 46 and 59 million people (15%–20% of the population) live in rural areas. Population counts and proportions vary based upon the definition of “rural” and geographic unit used. ¹ , ² There are 15+ federal definitions of “rural”; most consider some combination of population size, proximity to urbanized or metropolitan areas, and commuting patterns, and use counties, ZIP codes, or census tracts as their geographic unit. ³ However, these definitions often suffer from measurement error, the modifiable areal unit problem (i.e., different results dependent upon shape and scale of geographic boundaries), and face validity concerns. ⁴ There is also drastic variation in the size of geographic units like counties, ranging from 13.2 to 147,805 square miles for county or county‐equivalents like Alaskan boroughs. ¹ Further, whether someone is considered to live in a rural or urban area varies based upon the geographic unit used (e.g., areas may be considered urban at the county level, but rural at the census tract level) and choice of categorization (e.g., binary metropolitan/non‐metropolitan variable, three‐level metropolitan/micropolitan/rural variable). ⁵

The definition and geographic unit utilized to measure rurality has important implications. First, various definitions and geographic units capture important aspects of rurality—population size and proximity to metropolitan areas—differently. For example, a county with a large population size may not have access to health care services. However, a county may have a small population, but be proximal to a larger area with available health care services. Second, different geographic units and definitions are used by federal agencies and others for resource allocation. For example, National Cancer Institute (NCI) funding opportunities have used Rural–Urban Continuum Codes (RUCC) reflective of non‐metropolitan counties and/or Rural–Urban Commuting Area (RUCA) codes reflective of non‐metropolitan census tracts. ⁶ Meanwhile, the Federal Office of Rural Health Policy considers non‐metropolitan counties as rural and eligible for certain grant mechanisms, but also considers census tracts in metropolitan counties that are 400+ square miles with fewer than 35 people/square mile as rural and eligible for funding. ⁷

Regardless of definition used, rural populations tend to have less potential access to health care. ⁸ , ⁹ Access to care can be operationalized across two domains: potential versus realized. ¹⁰ For example, potential access can be defined as access to care relative to population need, but does not consider whether an individual actually utilizes this care. Alternatively, realized access is characterized by the utilization of a given health care service. ¹⁰ Access to care can also be operationalized within the “5 A's of access” (i.e., availability, accessibility, affordability, acceptability, and accommodation) as proposed by Penchansky and Thomas. ¹⁰ , ¹¹ Potential availability is often characterized by the provider‐to‐population ratio within a specific area (e.g., physicians per 100,000 population within a county). Rural populations frequently have less availability of health care than their urban counterparts. ⁸ , ¹² Potential accessibility is often operationalized as distance or travel time to care, and rural populations often have longer distances or travel times to primary care, specialty care, and hospitals than urban populations. ¹³ , ¹⁴ , ¹⁵ Potential affordability is often characterized as whether an individual has health insurance. Rural populations are less likely to have health insurance than those in urban areas. ¹⁶ Broadly, acceptability is characterized by whether an individual is satisfied with available care, and accommodation is characterized by whether available care meets their needs (e.g., clinic hours, language interpretation services, provider attentiveness).

Rural–urban differences in realized availability and accessibility (often operationalized as utilization of care) have been studied extensively, but not across different rural–urban definitions or across components of the 5 A's framework. Studies of realized acceptability, for example, patient satisfaction or perception of care quality, have primarily been performed among beneficiaries of private insurers or Medicare, which has limited generalizability. ¹⁷ Studies examining realized accommodation are also limited.

The ability to surveil urban vs. rural differences in health care access varies across population‐based datasets, given the availability of measures and provision of geographic identifiers. For example, the Health Information National Trends Survey (HINTS) from NCI includes both county and census tract level rural–urban measures such as RUCC and RUCA codes, respectively, with both metrics including their full range of values (i.e., not collapsed into a smaller number of categories). ¹⁸ In contrast, the publicly available 2019 Behavioral Risk Factor Surveillance System (BRFSS) data include the county‐level National Center for Health Statistics Urban–Rural Classification System variable, but the six‐category metric is collapsed into two categories. ¹⁹ As a result, findings to a research question may vary across datasets, based upon the geographic unit and rural definition used. Previous methodological studies have examined how findings in cancer outcomes, such as incidence and staging, vary based upon geographic unit or rural–urban measure used. ⁴ , ⁵ However, few studies have examined how a self‐reported measure of realized health care access may vary across geographic units and/or rural–urban definitions. Identifying if and to what extent these differences exist may inform resource allocation, federal grant eligibility criteria, rural health disparities research, and policy and regulatory decision making. Many resource allocation and policy decisions are based upon health behavior and outcome data that rely on population‐based surveys, such as the one used in this study. Understanding the extent to which prevalence statistics (or disparities across groups) vary based upon how rurality is defined is important because findings can have downstream impacts on program funding and health care delivery. For example, if it is identified that rural and urban residents similarly report quality of care, then quality improvement programs or policies may not be implemented. However, if a different definition of rurality is used disparities in quality of care may be observed that identify program and policy needs. To this end, we examined how self‐report of having a regular provider (realized availability), quality of care (realized acceptability), provider attentiveness (realized accommodation) varied by definition and categorization of rural.

2. METHODS

2.1. Data sources

We used data from HINTS, ²⁰ a nationally representative survey that monitors trends in cancer communication, health behaviors, and health care access, among other variables. Although HINTS is focused on cancer, it includes broader health‐related questions and has a robust set of rural–urban measures at different geographic levels that enable exploration of methodological questions using health care access as an outcome. We used data from HINTS 5 Cycles 1–3, which includes surveys administered between 2017 and 2019. For all cycles, a two‐stage sampling approach was conducted to reach a nationally representative sample of non‐institutionalized adults aged 18 and older. For Cycles 1 and 2, mailed surveys were employed. In Cycle 3, a web‐based survey was piloted among some participants. Thus, multiple modalities were used: mail, web option (mail or web option), or a web bonus (mail or web option with a $10 incentive to complete via web). We used three cycles of data to ensure a robust sample size (N = 12,227), particularly of rural participants.

2.2. Definitions of rural

We considered three different U.S. Department of Agriculture (USDA) definitions to characterize rural: (1) Rural–Urban Continuum Codes (RUCCs; county‐based) (2) Urban Influence Codes (UICs; county‐based), and (3) RUCA (census tract‐based). RUCC is a county‐based classification scheme that categorizes counties based on population size for metropolitan counties, and both population size and adjacency to metropolitan area for non‐metropolitan counties. ²¹ Metropolitan areas are centered around an urbanized area of 50,000 or more persons, while non‐metropolitan areas are centered around areas of 10,000–49,999 population. Counties are categorized into nine different groups and traditionally dichotomized into rural or urban groupings (1–3 = metropolitan counties and 4–9 = non‐metropolitan counties). RUCCs were developed in 1974 to provide county‐based distinctions beyond a simple dichotomous metropolitan/non‐metropolitan designation by considering the non‐metropolitan counties' degree of urbanization and adjacency to metropolitan counties. UICs also categorize counties based upon population size and adjacency to metropolitan areas, but have a larger range of values (1–12) that further break down non‐metropolitan counties into micropolitan and non‐core counties based upon their adjacency. ²² Micropolitan areas are non‐metropolitan areas centered on population clusters of 10,000–49,999 persons, while non‐core areas are those that are neither part of metropolitan nor micropolitan areas. UICs were developed in 1994 in part to help identify differences in economic opportunities. We also considered the USDA's census tract‐based RUCA primary codes, which range from 1 to 10 and categorize census tract‐based upon their location in metropolitan, micropolitan, small town, or rural areas, and their commuting patterns relative to urban areas. ²³ RUCAs were developed in the 1990s to provide rural–urban characterizations at a sub‐county level (census tract). Following their original development, all three definitions have been updated every decennial census.

2.3. Rural definition and categorization approaches

Considering each definition's unique components, we examined three approaches: (1) binary (metropolitan and non‐metropolitan) at both county and census tract levels; (2) three categories based on urban population size (metropolitan, micropolitan, and non‐metropolitan/non‐micropolitan); and (3) three groups based on metropolitan status and adjacency to metropolitan areas as determined by either geographic (i.e., adjacency) or practical (i.e., commuting patterns) connectedness. Table 1 describes the approaches and rationale.

TABLE 1.

Rural definition and categorization approaches

Approach	Description/Rationale	Rural–urban continuum code designations (RUCC)	Urban influence code designations (UIC)	Rural–urban commuting area (RUCA) code designation
Approach 1: Metropolitan, non‐metropolitan	Binary categorization (RUCC, UIC are equivalent)	Metro = 1–3 (Population: 262,452,132) Non‐metro = 4–9 (Population: 46,293,406)	Metropolitan = 1–2 (Population: 262,452,132) Non‐metropolitan = 3–12 (Population: 46,293,406)	Metropolitan = 1–3 (Population: 257,810,493) Non‐metropolitan = 4–10 (Population: 50,935,045)
Approach #2: Metropolitan, micropolitan, non‐core	Three categories; non‐metropolitan groups based upon population size	N/A	Metropolitan = 1–2 (Population: 262,452,132) Micropolitan = 3,5,8 (Population: 27,154,213) Non‐core = 4,6,7,9–12 (Population: 19,139,193)	Metropolitan = 1–3 (Population: 257,810,493) Micropolitan = 4–6 (Population:27,788,720) Non‐core =7–10 (Population:13,906,208)
Approach #3: Metropolitan, adjacent to metropolitan, non‐adjacent to metropolitan	Three categories; non‐metropolitan groups based upon adjacency to metropolitan(RUCC, UIC are equivalent)	Metropolitan = 1–3 (Population: 262,452,132) Non‐metropolitan, adjacent = 4,6,8 (Population: 30,480,746) Non‐metropolitan, non‐adjacent = 5,7,9 (Population: 15,812,660)	Metropolitan = 1–2 (Population: 262,452,132) Non‐metropolitan, adjacent = 3–7,9–10 (Population: 33,050,993) Non‐metropolitan, non‐adjacent = 8,11–12 (Population: 13,242,413)	Metropolitan = 1–3 (Population: 257,810,493) Non‐metropolitan, high commuting or internal flow = 4,5,7,8 (Population:38,230,080) Non‐metropolitan, low commuting = 6,9,10 (Population:12,704,965)

Note: Population counts are based upon the 2010 census counts, which were used to develop these USDA measures.

2.4. Outcome variables and covariates

We considered three outcome variables on access to care available in all three HINTS five iterations, and thus facilitated a more robust rural sample size. These outcome measures are operationalized as measures of realized availability, acceptability, and accommodation. While these types of outcomes, especially self‐reported quality, have been explored in previous studies, they have not been operationalized as measures of “realized access”. The realized availability outcome was operationalized through whether a survey participant had a regular provider, that is, “Is there a doctor, nurse, or health professional that you see most regularly?” (Yes/No). Realized acceptability was operationalized as a self‐reported measure of the quality of care the survey participant had received in the last 12 months if they had a nonemergency room health care visit within the past 12 months. This Likert scale question was dichotomized into “excellent” and all other responses ranging from “poor” to “very good”. Realized accommodation was operationalized as a self‐reported measure of the participant's perception of the frequency of the provider's attentiveness, that is, “How often did [your provider] spend enough time with you?”, which was asked of respondents who had a visit to a health care provider or urgent care within the past 12 months. Response options included “always”, “usually”, “sometimes”, and “never”, but these were dichotomized as “always” compared with all other responses for our analyses. In our analysis, we also considered additional sociodemographic and health status variables including age, sex, race/ethnicity, insurance status, income, education, marital status, sexual orientation, U.S. Census region, and self‐reported health status. These variables were chosen as, in previous analyses, they commonly vary across rural and urban populations and thus may vary uniquely in statistical significance and magnitude based upon rural definition and categorization approaches and may have implications for resource allocation and policy decisions. ²⁴

2.5. Statistical analyses

Before merging all HINTS five iterations, following NCI analytic guidance, we assessed the HINTS 5 Cycle 3 data to determine if any of our rural–urban variables, covariates, or outcome variables varied across survey modalities to guide any survey weighting considerations in our analyses. ²⁵ Our variables of interest did not vary by survey modality (data not shown). We merged all three iterations of HINTS 5 and used jackknife replication variance estimation procedures to account for the complex survey design in all analyses in alignment with analytic recommendations from NCI. ²⁴

Sociodemographic characteristics were compared using Wald Chi‐square tests for each of the rurality definition approaches. Wald Chi‐square tests were also conducted to assess the prevalence of having a regular provider across rurality definition approaches. Unadjusted and adjusted logistic regression models were constructed to examine the association between each of three definitions across all three rural definition approaches and having a regular provider. Marginal effects and 95% confidence intervals (CIs) are reported.

3. RESULTS

Tables S1, S2,and S3 show sociodemographic characteristics for each respective approach. For all approaches, differences in socioeconomic characteristics across rural–urban measures were largely the same except for sexual orientation. However, there were notable differences in the proportion of participants living in different census regions and of different age groups in approaches 1 and 2 dependent upon the geographic unit used. In approach 1, there were statistically significant differences in sexual orientation across the county‐based definition, but not the census tract‐based definitions. In approach 2, in the county‐based grouping, 50.6% and 11.9% of non‐core participants resided in the South and West, respectively, compared with 35.3% and 16.1% in the census tract‐based grouping. In approach 3, in the county‐based grouping, 26.2% of participants in the non‐metropolitan/non‐adjacent county‐based grouping were aged 18–34 compared with 14.5% of those in the corresponding census tract‐based grouping.

3.1. Realized availability (having a regular provider)

Table 2 (n = 12,033) shows the prevalence of having a regular provider across approaches. In approach 1 (binary metropolitan/non‐metropolitan comparison), there were statistically significant differences in having a regular provider for the census tract‐based measure only with a higher proportion of non‐metropolitan participants reporting a regular provider compared with metropolitan (68.9% compared with 64.4%, respectively). Approach 2, which considers a three‐group categorization (metropolitan, micropolitan, and non‐core), showed statistically significant differences for both county and census tract measures. For approach 3, a three‐group categorization scheme based on metropolitan status and adjacency, statistically significant differences were observed only for the census tract measure, 64.4%, 66.5%, and 77.0% for metropolitan, non‐metropolitan adjacent, and non‐metropolitan adjacent, respectively.

TABLE 2.

Self‐Report of regular provider, quality of care, and provider attentiveness across rural–urban definition and categorization approaches

	Regular provider, yes (n = 12,033)		Quality of care, excellent (n = 9695)		Provider spent enough time with me, always (n = 10,422)
	County‐based measure (%)	Census Tract‐based measure (%)	County‐based measure (%)	Census Tract‐based measure (%)	County‐based measure (%)	Census tract‐based measure (%)
Approach #1		*			*	*
Metropolitan	64.8%	64.4%	36.2%	34.5%	47.9%	48.0%
Non‐metropolitan	67.4%	68.9%	34.3%	34.6%	54.6%	53.5%
Approach #2	*	*			*
Metropolitan	64.4%	64.4%	34.5%	34.6%	47.5%	48.0%
Micropolitan	66.2%	65.9%	34.4%	32.4%	52.6%	51.7%
“Non‐core”	72.6%	73.5%	35.9%	37.7%	55.4%	56.1%
Approach #3		*			*
Metropolitan	64.8%	64.4%	34.3%	34.6%	47.9%	48.0%
Non‐metropolitan, adjacent	68.8%	66.5%	36.1%	33.7%	56.4%	53.9%
Non‐metropolitan, not adjacent	65.0%	77.0%	36.2%	37.1%	51.7%	53.4%

^* p < 0.05.

In Table 3, we present the findings of unadjusted and adjusted logistic regression models indicating associations between our three rural approaches and having a regular provider. For approach 1, there were no differences in likelihood of having a regular provider across non‐metropolitan and metropolitan in either unadjusted and adjusted models for both the county‐based and census tract‐based measures. For approach 2, in both the county‐based and census tract‐based “non‐core” groups, the crude models showed greater likelihood of having a regular provider, but these associations were attenuated after adjustment. For approach 3, the unadjusted model for the census tract‐based measure showed that the non‐metropolitan, non‐adjacent group had 13.98 percentage point increase in probability of having a regular provider (95% CI = 4.41–23.55 percentage points) with attenuated results after adjustment. Full adjusted model findings are shown in Table S4.

TABLE 3.

Unadjusted and adjusted associations between rural measure and having a regular provider

	County‐based measure		Census tract‐based measure
	Unadjusted marginal effects	Adjusted marginal effects	Unadjusted marginal effects	Adjusted marginal effects
	(95% CI)	(95% CI)	(95% CI)	(95% CI)
Approach #1
Metropolitan	REF	REF	REF	REF
Non‐metropolitan	2.67 (−1.80–7.22)	−0.76 (−4.65–3.14)	4.61 (0.00–9.16)	−0.98 (−2.95–4.92)
Approach #2
Metropolitan	REF	REF	REF	REF
Micropolitan	1.18 (−2.75‐6.36)	−0.99 (4.75–2.78)	1.44 (−4.42–7.29)	−0.99 (−5.87‐3.88)
Non‐core	8.61 (1.48–15.74)*	3.03 (−2.67–8.72)	9.67 (3.49–15.85)*	4.43 (−0.87–9.74)
Approach #3
Metropolitan	REF	REF	REF	REF
Non‐metropolitan, adjacent	4.18 (−1.17–9.53)	−0.65 (−5.34–4.03)	2.07 (−2.99–7.14)	−0.46 (−4.65–3.73)
Non‐metropolitan, not adjacent	2.00 (−7.98–8.38)	−0.93 (−7.78–5.93)	13.98 (4.41–23.55)*	6.58 (−2.23–15.41)

Note: Marginal effects are presented as percentage points. Adjusted models accounted for age, sex, race/ethnicity, insurance status, income, education, marital status, sexual orientation, U.S. Census region, and self‐reported health status.

^* p < 0.05.

3.2. Realized acceptability (perceived quality of care)

We also found that regardless of approach and geographic unit used, there was no statistically significant rural–urban difference in the proportion of survey respondents who reported “excellent” quality of care. The percentage of respondents who indicated that they received “excellent” quality of care ranged from 32.4% to 37.7% (Table 2, n = 9695).

Table 4 displays the unadjusted and adjusted findings regarding self‐reported quality of care, that is, operationalized realized acceptability. Both unadjusted and adjusted analyses across all approaches and geographic units showed no evidence of associations between rurality and quality of care. Full adjusted model findings are shown in Table S5.

TABLE 4.

Unadjusted and adjusted associations between rural measure and perceived quality of care

	County‐based measure		Census tract‐based measure
	Unadjusted marginal effects	Adjusted marginal effects	Unadjusted marginal effects	Adjusted marginal effects
	(95% CI)	(95% CI)	(95% CI)	(95% CI)
Approach #1
Metropolitan	REF	REF	REF	REF
Non‐metropolitan	1.82 (−2.32‐5.95)	2.26 (−1.77–6.30%)	−0.13 (−4.10–3.85)	0.44(−3.35–4.24)
Approach #2
Metropolitan	REF	REF	REF	REF
Micropolitan	−0.15 (−4.30–4.00)	−0.02 (−4.37‐4.32)	−2.26 (−7.02–2.51)	−0.92 (−5.54–3.69)
Non‐core	1.34 (−4.40–5.59)	1.67 (−4.36–7.71)	3.01 (−2.90–8.91)	2.48 (−2.84–7.80)
Approach #3
Metropolitan	REF	REF	REF	REF
Non‐metropolitan, adjacent	1.80 (−4.20–7.81)	1.47 (−3.84–6.77)	2.42 (−6.51–11.35)	0.47 (−3.48–4.43)
Non‐metropolitan, not adjacent	1.84 (−4.83–8.52)	3.53 (−3.48–10.54)	−0.93 (−5.09–3.22)	0.35 (−7.35–8.05)

Note: Marginal effects are presented as percentage points. Adjusted models accounted for age, sex, race/ethnicity, insurance status, income, education, marital status, sexual orientation, U.S. Census region, and self‐reported health status.

3.3. Realized accommodation (provider attentiveness)

However, there were rural–urban differences in the proportion of respondents who indicated that their provider “always” spent enough time with them (Table 2, n = 10,422). For county‐based measures, respondents who lived in non‐metropolitan counties reported greater attentiveness from their provider compared with respondents who live in metropolitan counties. In approach 2 (metropolitan/micropolitan/non‐core), there was increasing reported attentiveness with decreasing urbanized population size: metropolitan (47.5%), micropolitan (52.6%), and non‐core (55.4%). In approach 3, both non‐metropolitan groups had higher percentage of respondents reporting attentiveness among their physicians compared with the metropolitan group (47.9%), with a higher percentage reported among those in non‐metropolitan counties adjacent to metropolitan (56.4%) compared with non‐metropolitan counties that are not adjacent to metropolitan counties (51.7%). With the census tract‐based measures, the only statistically significant difference in attentiveness was in the non‐metropolitan (53.5%) to metropolitan (48.0%) grouping, which showed the same dynamic of the county‐based comparison.

Table 5 shows the unadjusted and adjusted analyses of the association between rurality and self‐reported provider attentiveness. For all approaches and geographic units, except for the census tract‐based adjacency grouping (approach 2), unadjusted analyses showed that the more rural groupings tended to report increased probability that their provider always spent enough time with them. In adjusted analyses, two of the county‐based approaches showed continued increased probability of realized accommodation. Respondents who lived in non‐metropolitan counties had a 6.03 percentage point increased probability of noting that their provider spent enough time with them (95% CI = 0.76–11.31 percentage points) than their metropolitan counterparts. In the three‐group approach, which considered adjacency to metropolitan counties, those who lived in a non‐metropolitan county adjacent to metropolitan county had a 6.90 percentage point increased probability of noting that their provider spent enough time with them (95% CI = 0.92–12.89 percentage points) compared with urban respondents. Full adjusted model findings are shown in Table S6.

TABLE 5.

Unadjusted and adjusted associations between rural measure and self‐reported provider attentiveness

	County‐based measure		Census tract‐based measure
	Unadjusted marginal effects	Adjusted marginal effects	Unadjusted	Adjusted
	(95% CI)	(95% CI)	Marginal effects	Marginal effects
			(95% CI)	(95% CI)
Approach #1
Metropolitan	REF	REF	REF	REF
Non‐metropolitan	6.73 (1.55–11.90)*	6.03 (0.76–11.31)*	5.55 (0.41–10.60)*	4.87 (−0.28–10.02)
Approach #2
Metropolitan	REF	REF	REF	REF
Micropolitan	5.01(0.40–9.78)*	4.50 (0.30–9.29)	3.73 (−2.42–9.88)	4.07 (−2.07–10.21)
Non‐core	7.83 (0.98–14.68)*	6.36 (−0.77–13.48)	8.15 (1.19–15.10)	6.10 (−4.48–16.67)
Approach #3
Metropolitan	REF	REF	REF	REF
Non‐metropolitan, adjacent	8.54 (2.17–14.91)*	6.90 (0.92–12.89)*	5.38 (0.02–10.07)	3.65 (−6.36–13.66)
Non‐metropolitan, not adjacent	3.81 (−4.73–12.34)	4.63 (−4.39–13.64)	5.89 (−4.03–15.80)	5.25(−0.03–10.54)

Note: Marginal effects are presented as percentage points. Adjusted models accounted for age, sex, race/ethnicity, insurance status, income, education, marital status, sexual orientation, U.S. Census region, and self‐reported health status.

^* p < 0.05.

4. DISCUSSION

We used population‐based survey data from NCI to assess if and how definitions of “rural” and geographic units differed using a self‐reported measure of realized access to care. The sociodemographic characteristics of rural populations were similar for dichotomous definitions of rural regardless of a geographic unit. However, we found that for definitions based upon population size, census tract‐based measures had slightly higher proportions of non‐metropolitan, micropolitan, or non‐core participants in the West compared with county‐based measures. Definitions based upon adjacency to metropolitan showed descriptive differences in census region and age. Census tract‐based measures indicated a higher percentage of respondents in non‐adjacent and low commuting non‐metropolitan areas had a regular provider compared with other groupings. Non‐core respondents in both census tract and county‐based measures and non‐metropolitan, non‐adjacent respondents in census tract‐based measures had greater probability of having a regular provider, but these associations were attenuated to non‐significance in adjusted models. There were no differences in self‐reported “excellent” quality of care regardless of rural approach or geographic unit. However, we found that, in general, a higher proportion of rural participants reported that their provider “always” spent enough time with them with magnitude of differences varying by definition and categorization used. This association was maintained after accounting for covariates for some categorizations of rural.

There were mixed findings with the sociodemographic characteristics of the study sample dependent on the geographic unit, approach, and definition. When dichotomizing respondents into metropolitan and non‐metropolitan, the demographic and sociodemographic characteristics were similar regardless of geographic unit considered, suggesting that dichotomizing census tract‐based and county‐based measures may yield comparable results. This may be helpful for research that focuses on rural–urban health disparities, as often population‐based national registries and surveys provide only county‐based metrics for rural and urban or dichotomous versions of other metrics. ²⁵ This also corroborates previous research on breast cancer patients in Texas, which found a high level of agreement between binary rural–urban indicators as county and census tract levels. ⁵ However, for three‐group definitions based on either adjacency to metropolitan or population size, there were notable differences between the county and census tract‐based definitions related to U.S. Census region and age. For U.S. Census regions, this may be due to the large variation in geographic size of counties in different parts of the country. For example, in the West, counties are geographically much larger than in other parts of the country. ¹ Thus, when both geographic granularity (i.e., census tracts) and greater range in definitions (i.e., three groups) are used, this may provide a more accurate description of the study population.

We identified significant differences in the percentage of respondents who had a regular provider for the census tract‐based versions of all three approach types, but only for one of the county‐based approaches (the three‐level approach‐based on population size). In all these approaches, the most “rural” grouping had the highest percentage of respondents who reported a regular provider. These findings corroborate with older analyses of Medical Expenditure Panel Survey data that suggest rural patients may be more likely to report having a regular provider. ²⁶ Methodologically, this is consistent with previous studies showing that census tract‐based RUCA codes may be the best definition to use for health outcomes studies if census tract data are available as this measure draws the greatest distinctions between rural and urban groups and incorporates both population density and commuting patterns. ⁵ In our study, this metric showed consistent non‐metropolitan and metropolitan differences in a regular provider regardless of the approach of categorization, while there were mixed findings with the county‐based definition.

We also found that, across all approaches and groupings, there were no geographic differences in realized acceptability of care as operationalized by an assessment of care quality. This suggests that any available geographic unit or operationalization of rural–urban measures may provide similar results for assessing patient satisfaction with the quality of care. Previous studies among Medicare beneficiaries and other cohorts of older adults nationally and single state studies have yielded mixed findings. ²⁷ , ²⁸ , ²⁹ A single state survey and a national study of Medicare beneficiaries found similar levels of patient satisfaction as operationalized as a single question, but another study found that across different measures of quality (e.g., ease of care access) there were differences across metropolitan, micropolitan, and non‐core counties. ³⁰ In another study using the Surveillance Epidemiology and End Results‐Consumer Assessment of Healthcare Providers and Systems, researchers identified rural–urban differences in satisfaction with the timeliness and ease of care among Medicare beneficiaries with a cancer diagnosis with rural patients indicating lower satisfaction with timeliness. ³¹ In a study of hospice patients and their families, rural participants indicated better satisfaction with pain management as compared with urban participants. ³² Taken together, these studies show conflicting findings based upon number of geographic groupings, operationalization of realized acceptability, age of cohort, and treatment received. Future studies should examine these nuances among patients of a wider age range to better inform policy and clinical interventions to optimize realized acceptability of care. This is especially critical as many other assessments of quality are performed at the provider or hospital level as an aggregate of their patients' perceptions or other metrics, but rural or urban location of the provider or hospital is not always concordant with the rural or urban residency of the patient. ³³

Our findings showed that, even after adjustment for covariates, for some county‐based measures of rurality, rural participants had greater probability of self‐reported provider attentiveness (a measure of realized accommodation). These findings only occurred when rurality was considered as a dichotomous variable and in considering adjacency to urban, but greater probability of provider attentiveness was observed only among those living in non‐metropolitan counties that are adjacent to metropolitan. Of note, the questions regarding both quality of care (realized acceptability) and attentiveness of providers (realized accommodation) were only asked among those who had a physician visit within the past year, indicating that these respondents had access to a physician. This suggests that, among rural persons who visited a physician, they receive care that is comparable or favorable in the areas of acceptability and accommodation relative to their urban counterparts. This underscores the importance of considering both potential and realized access to care in evaluating rural–urban differences in access to care. To a certain extent, such studies can only be performed if data are collected outside of the health care system, such as population‐based surveys like HINTS, BRFSS, the National Health Interview Survey, and others, as it is important to capture whether potential access limitations inhibit utilization of care. To facilitate such analyses, mechanisms to obtain data at meaningful units, such as census tract and county, are needed to enable the linkage of potential access measures like distance to care or provider–population ratios. However, such data are often lacking or administratively or financially burdensome to obtain and analyze. Streamlined procedures and remote, secure access to data are needed to improve data availability. It is also important to consider the dynamic between potential and realized access to care to understand how patients make care decisions (i.e., among those with realized access to care). For example, studies show that some patients may bypass nearby care to obtain care that is of objective or subjective higher quality or that is more accommodating (e.g., cultural competency and language concordance). ³⁴ , ³⁵

Our study found that depending upon the outcome variable explored, census tract or county‐based measures identified greater nuance in the relationship between rurality and realized accessibility. Thus, this is an important area for continued research. First, some studies have shown notable discordance between federal categorizations of “rural” and “urban” and individual perceptions of whether where they live is urban, suburban, and rural. ³⁶ , ³⁷ This may also be an indication that perceptions of “rural” may go beyond what is quantified by federal measures, which are focused on population size, adjacency to metropolitan areas, and commuting patterns. Second, the Office of Management and Budget will continue to consider changing the definition of metropolitan/non‐metropolitan by examining factors beyond population thresholds, such as economic thresholds. ³⁸ If changes are made, this will subsequently affect definitions from the USDA, such as those examined in this study. Further, RUCCs, UICs, and RUCAs are updated in the years following each decennial census. Therefore, based upon population growth and decline as well as population dynamics, the codes for each census tract and county will likely change in the coming years.

Our study was not without limitations. First, we were only able to examine a self‐reported measure of having a regular provider, not how frequently they saw that specific provider or their ease in accessing the provider. Further, we were not able to assess how this might vary by the type of health care services (e.g., primary care or specialty care). While we were able to examine realized availability, acceptability, and accommodation, we were not able to assess realized accessibility or affordability. Questions on distance or travel time to a provider visited by a respondent would be required to assess realized accessibility. To assess realized affordability, questions would need to focus on the financial facilitators or barriers to accessing care, which were not available in the HINTS dataset. Additionally, we only examined two‐ or three‐group categories of rural and urban due to sample size, but examining more nuanced, continuous definitions of rurality (e.g., Index of Relative Rurality) is an important area of future research to better understand how access to care varies, especially in the most isolated rural areas. ³⁹ However, our study also had strengths. Notably, we used HINTS data that include different rural–urban metrics representing several geographic scales allowing us to explore different measures of rural and to consider different groupings of each. Many other federal population‐based survey data sources do not have rural–urban variables available at different levels to conduct similar analyses.

5. CONCLUSIONS

Our study found that for dichotomous rural–urban definitions, sociodemographic characteristics were similar regardless of geographic unit used, but some differences were identified across geographic units with three‐level groupings. Additionally, we found that relationships between rurality and measures of realized access to care varied depending upon the geographic unit and categorization approach used. As these measures are updated every 10 years, and as further changes to rural–urban measures evolve, this is an important area for continued research and evaluation given the important implications for funding, resource allocation, and identifying health disparities. Further, our findings suggest that continued research is needed to examine the interplay between potential and realized access across dimensions of access to better understand rural–urban disparities in access to care. This may help researchers and policy makers understand how greater availability and proximity to health care services translates into greater utilization of needed services and the perceived accommodation and acceptability of services.

Supporting information

Data S1 Supporting information.

Zahnd WE, Del Vecchio N, Askelson N, et al. Definition and categorization of rural and assessment of realized access to care. Health Serv Res. 2022;57(3):693-702. doi: 10.1111/1475-6773.13951