Spatial Access to Primary Care Providers in Appalachia

Abstract

Purpose: The goal of this research was to examine spatial access to primary care physicians in Appalachia using both traditional access measures and the 2-step floating catchment area (2SFCA) method. Spatial access to care was compared between urban and rural regions of Appalachia. Methods: The study region included Appalachia counties of Pennsylvania, Ohio, Kentucky, and North Carolina. Primary care physicians during 2008 and total census block group populations were geocoded into GIS software. Ratios of county physicians to population, driving time to nearest primary care physician, and various 2SFCA approaches were compared. Results: Urban areas of the study region had shorter travel times to their closest primary care physician. Provider to population ratios produced results that varied widely from one county to another because of strict geographic boundaries. The 2SFCA method produced varied results depending on the distance decay weight and variable catchment size techniques chose. 2SFCA scores showed greater access to care in urban areas of Pennsylvania, Ohio, and North Carolina. Conclusion: The different parameters of the 2SFCA method—distance decay weights and variable catchment sizes—have a large impact on the resulting spatial access to primary care scores. The findings of this study suggest that using a relative 2SFCA approach, the spatial access ratio method, when detailed patient travel data are unavailable. The 2SFCA method shows promise for measuring access to care in Appalachia, but more research on patient travel preferences is needed to inform implementation.

Introduction

Appalachia is largely rural, with 42% of its population classified as rural compared with the national average of 20%.¹ Socioeconomically, the region has a lower per capita income and a higher poverty rate than the national average.² Access to adequate health care is an ongoing concern in Appalachia, largely due to the mountainous terrain, rural population distribution, and socioeconomic disparities.³ Access to primary care services is especially important given primary care providers’ role as a gateway to health systems.⁴ In Appalachia, regular primary care encounters have been shown to increase early cancer detection and reduce mortality.^5,6 In Appalachia Ohio, children with irregular primary care visits had poorer general health outcomes, and the parents of those children reported that lack of access to primary care prevented regular contact.⁶ Regular, quality primary care encounters can also counteract the negative effects that economic disparities have on heath,⁷ a particularly important outcome given Appalachia’s generally reduced economic status.

Accurately measuring access to primary care is important for guiding interventions in Appalachia regions that often lack adequate resources. Spatial access is one component of access to care—distinct from nonspatial factors such as insurance status or level of education—and its status is often a population-wide indicator, with the US Department of Health and Human Services Health Professional Shortage Areas (HPSA) designation being the most common.^8,9 One traditional measure of spatial access is a county provider to population ratio, which is the methodology used for the HPSA designation.⁹ Another common strategy for measuring spatial access to care is to use geographic information systems (GIS) software to calculate the shortest travel time between population points and health care providers. Travel time is advantageous when geographic units are larger and more dispersed, as is the case in much of Appalachia.¹⁰ Nonetheless, there are noted limitations to these standard measures of spatial access to care.¹¹ Patients often cross county boundaries, and providers within a county boundary may not actually be accessible, especially in large, rural Appalachia travel time calculations overcome the latter problem of travel cost between patients and providers counties.¹⁰ However, travel time measures do not account for the supply and demand factors that affect health care.¹²

A more recent method for measuring spatial access to care is the 2-step floating catchment area (2SFCA) method.^8,12 The 2SFCA method begins by setting a catchment area (usually 30 or 60 minutes) around each health care provider and identifying all the populations within that provider’s catchment. A provider to population ratio is then calculated for each provider’s catchment. In step 2, each population becomes the center of a catchment, and the step 1 ratios associated with each provider in that population’s catchment are summed.^13-15 Owing to the uncertainty over how to weight travel within catchments, or how exactly to vary catchments, a relative approach called the spatial access ratio (SPAR) was created.¹⁶ This technique divides a population’s 2SFCA score by the mean 2SFCA score of all populations, which researchers demonstrated minimizes the differences in 2SFCA scores resulting from using different decay weights.¹⁶ Despite its theoretical advantages and its empirical validity in predicting clinical outcomes,^17,18 the 2SFCA method has never been used to study access to primary care in Appalachia. The focus of this study was to examine spatial access to primary care in Appalachia using both the traditional measures of provider to population ratios and closest travel time, as well as the 2SFCA method and its various parameters.

Study Area and Data

The study examined the Appalachia regions of Pennsylvania, Ohio, Kentucky, and North Carolina in the United States. The University of Michigan Human Subjects Board exempted this study from formal approval since completely de-identified patient data were used. Appalachia counties were determined using the Appalachia Regional Commission’s (ARC) county designations. Primary care physicians were derived from the 2008 American Medical Association (AMA) Physician Masterfile, using specialties of Family Practice, General Practice, Internal Medicine, and General Pediatrics.⁵ Office addresses at the street level were available for 8039 of the 9483 physicians in the study area. The population weighted centroid of the physician’s office census tract was used for the remaining 1444 physicians. Primary care physicians from neighboring states and neighboring in-state, non-Appalachia areas were not included, creating possible edge effects.

The 2010 US Census was used to derive population data at the census block group level. Block groups are ideally composed of 1500 people, with a range from 600 to 3000. The population weighted centroid¹⁹ of each block group was used as the geographic reference, similar to previous research.¹² Many previous studies measuring spatial access to care used the larger geographic areas of zip codes¹⁷ or census tracts,¹⁶ which provide less geographic specificity than the smaller census block groups. This study included neighboring block groups within 1 hour travel time of our Appalachia study area. There were 8721 populated block groups in Appalachia regions of the 4 states, resulting in a total population of 10 717 421 people.

Within Appalachia, populations were also dichotomized as urban and rural. Designation occurs at the census block level by assigning RUCU codes 7-10 as rural and RUCU codes 1-6 as urban.²⁰ Codes 1-6 also include suburban areas, but these areas were grouped as urban for ease of interpretation.²⁰ Descriptive statistics and t tests were used to examine the differences between rural and urban regions.

Method

Provider to Population Ratios

Primary care provider to population ratios were calculated at the county level. The total primary care providers in each county were divided by that counties total population. Although primary care data was current as of 2008, and population data was from the 2010 Census, the 2010 Census offered a closer approximation than 2000 Census data.

Travel Time

Primary care providers and census block groups were geocoded into ArcGIS (Version 10.1, ESRI Inc., Redlands, CA). The origin-destination (OD) cost matrix function of the ArcGIS Network Analyst extension was used to determine travel times between the population weighted centroid of each block group and the closest primary care provider. Similar to previous research,¹⁰ and corresponding with the maximum time used in the 2SFCA method, a maximum travel time of 60 minutes was set.

Two-Step Floating Catchment Area

The 2SFCA method begins by identifying all populations (k) within a service provider’s (j) maximum catchment (d₀), which we set as 60 minutes driving time. The rural characteristics of Appalachia necessitated the 60-minute maximum time, rather than a 30-minute maximum.^13,14,16,21 A total of 10 spatial access scores were applied to each census block group in the Appalachia regions on Pennsylvania (PA), Ohio (OH), Kentucky (KY), and North Carolina (NC): (1) ratio of county primary care providers to total county population, (2) travel time to closest primary care provider, (3) original 2SFCA score, (4) slow-decay 2SFCA score, (5) fast-decay 2SFCA score, (6) slow-decay and variable catchment 2SFCA score, (7) fast-decay and variable catchment 2SFCA score, (8) continuous-decay and variable catchment 2SFCA score, (9) slow-decay and variable catchment SPAR score, and (10) fast-decay and variable catchment SPAR score.

Results

The greater Pittsburgh area had the highest population density of our 4 state Appalachia region (Figure 1). The Youngstown area along the PA-OH border, the Erie area in northwest PA, and the Greensboro and Ashland areas in NC all also had among the most populated counties in our study region. Northcentral PA and eastern KY were least population areas.

Figure 1.

Population densities from the 2010 US Census in Appalachia counties of Pennsylvania, Ohio, Kentucky, and North Carolina.

All Spearman nonparametric correlations were significant between the 10 spatial access measures, except for the correlation between travel time and continuous decay, variable catchment scores (Table 1). For reference, the provider to population ratios, 2SFCA scores, and SPAR scores are all similar in that increasing values signify greater spatial access. Travel time is opposite, where lower times indicate greater spatial access, which is why travel time was negatively correlated with each of the other measures. There was a marked difference between the 2SFCA measures that included a variable catchment size function and those that did not. Several of the areas with the greatest population density also had the highest ratio of primary care physicians to county population, including the Pittsburgh, Ashland, and Greensboro areas (Figure 2a). There were also rural regions with higher ratios, such as in eastern KY, northeastern PA, and eastern OH. Many of the counties with ratios in the highest quintile were bordering counties with ratios in the lowest quintile, demonstrating the impact that county boundary lines can have when using provider to population ratios.

Table 1.

Spearman Correlations^a Between 10 Spatial Access to Primary Care Measures Across Appalachia Pennsylvania, Ohio, Kentucky, and North Carolina.

Provider–population ratio	1
Travel time	−0.381	1
Original 2SFCA	0.564	−0.426	1
Slow 2SFCA	0.765	−0.566	0.833	1
Fast 2SFCA	0.760	−0.626	0.719	0.970	1
Slow–Var. Catch. 2SFCA	−0.046	−0.045	−0.115	0.056	0.124	1
Fast–Var. Catch. 2SFCA	0.198	−0.342	0.150	0.374	0.463	0.874	1
Cont.–Var. Catch. 2SFCA	−0.117	0.017b	−0.174	−0.043	0.023	0.972	0.805	1
Slow–Var. Catch. SPAR	−0.046	−0.045	−0.115	0.056	0.124	1.000	0.874	0.972	1
Fast–Var. Catch. SPAR	0.198	−0.342	0.150	0.374	0.463	0.874	1.000	0.805	0.874	1
	Provider–population ratio	Travel time	Original 2SFCA	Slow 2SFCA	Fast 2SFCA	Slow–Var. Catch. 2SFCA	Fast–Var. Catch. 2SFCA	Cont–Var. Catch. 2SFCA	Slow–Var. Catch. SPAR	Fast–Var. Catch. SPAR

Abbreviations: 2SFCA, 2-step floating catchment area; SPAR, spatial access ratio. Slow–Var. Catch. 2SFCA, both slow-decay weightings and varying catchment size rules; Fast–Var. Catch. 2SFCA: both fast-decay weightings and varying catchment size rules; Cont.–Var. Catch. 2SFCA: both continuous-decay weightings and varying catchment size rules.

^aAll correlations significant at the .05 level, unless otherwise noted.

^bNot significant

Figure 2.

Spatial access to primary care providers by (a) county provider to population ratio, broken into quintiles, and (b) closest driving time from census block groups. 2SFCA, 2-step floating catchment area.

The majority of census block groups were within a 10-minute drive time of their closest primary care physician (Figure 2b). The census block groups with the highest original 2SFCA scores were located primary in the Pittsburgh area (Fig. 3a). After adding distance decay weights, 2SFCA scores were reduced in the suburban areas surrounding Pittsburg but increased in several other population centers of our study region, including near the Greensboro, Ashland, and Youngstown areas (Figure 3b and c). Overall, decay weights appear to have dispersed scores, rather than concentrating them across one large, adjoining area, as with the original 2SFCA scores. This was especially true for the fast-decay weighting scheme (Figure 3c). Adding variable catchment sizes with decay weights drastically altered the distribution of spatial access 2SFCA scores (Figure 3d-f).

Figure 3.

Spatial access of census block groups in Appalachia Pennsylvania, Ohio, Kentucky, and North Carolina to primary care providers by (a) original 2SFCA scores; (b) slow-decay 2SFCA scores; (c) fast-decay 2SFCA scores; (d) slow-decay and variable catchment 2SFCA scores; (e) fast-decay and variable catchment 2SFCA scores; (f) continuous-decay and variable catchment 2SFCA scores; (g) slow-decay and variable catchment SPAR scores; (h) fast-decay and variable catchment SPAR scores. Scores are broken into quintiles, with larger quintiles representing greater spatial access. 2SFCA, 2-step floating catchment area.

The difference between slow- and fast-decay variable catchment 2SFCA scores was similar to the difference when only decay weights and no variable catchment function were used. Fast-decay weights resulted in tighter groupings of higher access block groups at population centers, rather than the slow-decay weight scenario of high access block groups sprawled across population centers and surrounding suburbs (Figure 3d and e). When a continuous decay weighting scheme was used with varying catchment sizes, the result was nearly identical to the slow-decay, variable catchment size approach (Figure 3d and f). The slow- and fast-decay variable catchment SPAR scores demonstrated the same differences as their corresponding 2SFCA scores (Figure 3g and h). Figure 4 compares the change in actual 2SFCA scores across rural and urban areas when using slow- and fast-decay weights. Urban areas were responsible for the highest scores, regardless of the 2SFCA iteration used. After using only the distance decay technique, scores generally did not uniformly increase or decrease according to the weight used (Figure 4a and b). The exception was for the highest scoring urban block groups, where the fast-decay weights universally increased scores (Figure 4b). After adding variable catchment sizes and decay weights, however, a clear pattern emerged where fast-decay weights decreased scores for most block groups (Figure 4c and d). The results when using the decay-weighted and variable catchment SPAR scores were mixed (Figure 4e and f). For block groups classified as urban (7266 of the 8721 block groups were classified as urban), the SPAR technique achieved it intended effect of preventing any noticeable increase or decrease in scores based on decay weighting (Figure 4f). The SPAR technique was less effective for rural block groups, where fast-decay weights uniformly decreased scores (Figure 4e).

Figure 4.

Comparison of spatial accessibility to primary care providers across rural and urban census block groups of Appalachia Pennsylvania, Ohio, Kentucky, and North Carolina between (a, b) slow-decay and fast-decay 2SFCA scores; (c, d) slow-decay, variable catchment and fast-decay, variable catchment 2SFCA scores; (e, f) slow-decay, variable catchment and fast-decay, variable catchment SPAR scores. 2SFCA, 2-step floating catchment area; SPAR, spatial access ratio.

Table 2 compared the rural and urban mean spatial access scores across each state of the study region. Urban block groups in PA, OH, and NC had higher provider to population ratios, while higher ratios came from rural block groups in KY. Travel time to the closest primary care provider was lower for urban block groups across each state. The mean scores for the entire region demonstrate that the addition of a variable catchment size decreased scores considerably, regardless of the decay weighting scheme chosen. In PA and NC, urban areas had larger 2SFCA scores except for the slow and continuous decay, variable catchment scores, where rural areas performed better. Urban block groups in OH performed even better, scoring higher than rural block groups across each 2SFCA iteration. Scores in KY were largely reversed, where rural block groups performed better across each 2SFCA method except the fast-decay and variable catchment size approach.

Table 2.

Descriptive Statistics of Primary Care Spatial Access Scores Between Rural and Urban Census Block Groups of Appalachia Pennsylvania, Ohio, Kentucky, and North Carolina.^a

Spatial Access Measures	All States	Pennsylvania		Ohio		Kentucky		North Carolina
		Rural	Urban	Rural	Urban	Rural	Urban	Rural	Urban
Provider to population ratio	0.000880 (0.000542)	0.000710 (0.000347)	0.001031 (0.000637)	0.000481 (0.000209)	0.000708 (0.000301)	0.000636 (0.000304)	0.000553 (0.000257)	0.000747 (0.000307)	0.000970 (0.000467)
Travel time to closest mammography center (min)	5.81 (6.31)	7.50 (6.92)	4.03 (4.77)	8.26 (7.29)	5.88 (5.98)	11.89 (8.81)	9.54 (7.79)	10.62 (7.64)	6.122 (5.79)
Original 2SFCA	0.000706 (0.000382)	0.000507 (0.000238)	0.000936 (0.000351)	0.000321 (0.000195)	0.000465 (0.000272)	0.000470 (0.000232)	0.000410 (0.000174)	0.000370 (0.000160)	0.000583 (0.000265)
Slow-decay 2SFCA	0.000768 (0.000490)	0.000477 (0.000245)	0.001002 (0.000518)	0.000323 (0.000161)	0.000529 (0.000320)	0.000501 (0.000306)	0.000435 (0.000223)	0.000416 (0.000190)	0.000724 (0.000378)
Fast-decay 2SFCA	0.000804 (0.000573)	0.000472 (0.000260)	0.001032 (0.000629)	0.000342 (0.000186)	0.000579 (0.000389)	0.000528 (0.000398)	0.000451 (0.000265)	0.000455 (0.000254)	0.000807 (0.000480)
Slow-decay, variable catchment 2SFCA	0.000134 (0.000078)	0.000154 (0.000075)	0.000122 (0.000074)	0.000114 (0.000056)	0.000125 (0.000055)	0.000191 (0.000115)	0.000177 (0.000083)	0.000183 (0.000107)	0.000136 (0.000069)
Fast-decay, variable catchment 2SFCA	0.000116 (0.000071)	0.000107 (0.000062)	0.000116 (0.000072)	0.000076 (0.000045)	0.000108 (0.000054)	0.000119 (0.000090)	0.000132 (0.000082)	0.000133b (0.000095)	0.000126 (0.000069)
Continuous-decay, variable catchment 2SFCA	0.000147 (0.000086)	0.000177 (0.000082)	0.000130 (0.000080)	0.000134b (0.000068)	0.000142 (0.000061)	0.000206 (0000125)	0.000195 (0.000092)	0.000212 (0.000123)	0.000149 (0.000079)
Slow-decay, variable catchment SPAR	1.00 (0.58)	1.15 (0.56)	0.91 (0.55)	0.85 (0.42)	0.94 (0.41)	1.43 (0.86)	1.32 (0.62)	1.39 (0.79)	1.01 (0.52)
Fast-decay, variable catchment SPAR	1.00 (0.61)	0.92 (0.53)	1.00 (0.62)	0.66 (0.39)	0.93 (0.47)	1.03 (0.77)	1.14 (0.71)	1.14b (0.82)	1.08 (0.59)

Abbreviations: 2SFCA, 2-step floating catchment area; SPAR, spatial access ratio.

^a.Values are presented as mean (SD).

^bNot significant at P < .05; all other rural and urban comparisons were significant at P < .05.

Discussion

This research presented the first comparison in Appalachia of the most recently developed spatial access to care methods. Based on nonparametric correlations, the 2SFCA measures that included variable catchment sizes appeared distinct from the measures of provider to population ratios, closest travel time, and the 2SFCA method with and without decay weights. When comparing provider to population ratios across the study region, the most populated urban areas (eg, Pittsburgh, Greensboro) had the highest ratios. Several rural areas had high ratios as well, but abrupt county boundaries made interpretation difficult because several adjacent counties had ratios in the highest and lowest quintile groups. Traditionally, when the original 2SFCA method was applied to primary care physicians, urban areas often received the highest scores.^8,12 This also proved true with our results in Appalachia. Pittsburgh was the largest urban area in the study region, and it had the highest original 2SFCA scores. When distance-decay weights were added, Pittsburgh’s core maintained comparatively high access scores, but the surrounding suburbs were reduced to more mediocre scores. Those surrounding populations, especially those with a 30- to 60-minute drive from the center of Pittsburgh, were no longer able to count the abundance of central Pittsburgh physicians as accessible. This pattern matches the intended effect behind distance-decay functions, which was partly to limit accessibility at the boundaries of physician catchments.¹³ After adding distance-decay weights, the same effect occurred across other areas where original 2SFCA scores were high across a contiguous area, such as near Youngstown, OH and along the Williamsport to Scranton corridor in central and northeastern PA.

Adding variable catchment sizes to the slow and fast decay 2SFCA approach produced dramatic differences in the distribution of access scores. Pittsburgh’s scores ranked among the lowest of the study region. Smaller rural towns in eastern KY, southern OH, western NC, and central PA—who without variable catchments only had mediocre scores—had among the highest scores in the study region. The variable catchment approach we used capped the number of physicians a population could access at 100.¹⁴ Most towns of 15 000 people or greater were within a 1-hour drive of 100 primary care physicians. But, the population demand on those physicians was less in rural areas, resulting in the increased scores relative to the major urban area of Pittsburgh.

This research had several strengths. It represents the first application of the 2SFCA method to study access to primary care in Appalachia. Public health research in Appalachia is increasingly utilizing GIS techniques,²² which makes it imperative that the most current spatial access methods are evaluated. A methodological strength in this study was the increased geographic specificity as a result of using census block groups as the geographic population area, rather than the larger census tracts or zip codes that other research applying the 2SFCA method has employed.^16-18 Another methodological strength was the inclusion of variable catchment sizes, which allowed the comparison between identical distance decay 2SFCA measures either with or without varying catchments, instead of only evaluating decay weights.^23,24

There were also several limitations to this research. From a methodological perspective, not including neighboring primary care physicians likely created edge effects along the border of our study region. Another limitation, common in much research examining potential spatial access, was the lack of actual patient health care utilization behaviors. The application of distance-decay weights and variable catchment sizes is theoretically sound. Grounding distance decay and variable catchment functions in empirical data is more difficult. The specific parameters of these 2SFCA iterations need to be evaluated against patient health care behavior, preferably at the local level of a study region.

In summary, based on the findings of this study, the SPAR technique is recommended for use when patients’ health care traveling data is unavailable. Including variable catchment sizes is a theoretical improvement to the 2SFCA method, but choosing optimal parameters to vary catchment sizes by is difficult. The findings of this study also suggest that researchers compare several parameters (eg, capping the number of physicians a population can access at higher and lower values than the standard approach of 100) and contrast the effects with 2SFCA methods that only employ distance decay weights.

Author Biographies

Joseph Donohoe, PhD, is informatics and special projects lead at Mountain Pacific Quality Health Care in Helena, MT. His research interests are health informatics and medical demography.

Vince Marshall, MS, is research analyst at University of Michigan College of Pharmacy, Ann Arbor, MI. His interests are biostatistics and statistical analyses.

Xi Tan, PhD, is assistant professor of Pharmaceutical Systems and Policy at West Virginia University, Morgantown, WV. Her research interests are focused on medication use evaluation and health outcomes assessment.

Roger T. Anderson is professor of Public Health Sciences at the University of Virginia and associate director of the UVA Cancer Center. His research focuses on social epidemiology and health disparities.

Fabian T. Camacho, MS, MA, is senior biostatistician at the University of Virginia School of Medicine. His research focuses on applied statistical methodology and quantiative research methods.

Rajesh Balkrishnan, PhD, is professor of Public Health Sciences at the University of Virginia School of Medicine. His research focuses on health disparities assessment and applied statistical methodology in evaluating effectiveness of health care.