ORIGINAL ARTICLE
Rural Health Care Bypass Behavior: How Community and Spatial Characteristics Affect Primary Health Care Selection Scott R. Sanders, MPA, PhD;1 Lance D. Erickson, PhD;1 Vaughn R.A. Call, PhD;1 Matthew L. McKnight, MS;1 & Dawson W. Hedges, MD2 1 Department of Sociology, Brigham Young University, Provo, Utah 2 Department of Psychology and the Neuroscience Center, Brigham Young University, Provo, Utah
Abstract Purpose: (1) To assess the prevalence of rural primary care physician (PCP) Funding: Funding for this research comes from the Brigham Young University Gerontology Center. For further information, contact: Scott R. Sanders, MPA, PhD, 2039 JFSB, Department of Sociology, Brigham Young University, Provo, UT 84602; e-mail: [email protected]. doi: 10.1111/jrh.12093
bypass, a behavior in which residents travel farther than necessary to obtain health care, (2) To examine the role of community and non-health-carerelated characteristics on bypass behavior, and (3) To analyze spatial bypass patterns to determine which rural communities are most affected by bypass. Methods: Data came from the Montana Health Matters survey, which gathered self-reported information from Montana residents on their health care utilization, satisfaction with health care services, and community and demographic characteristics. Logistic regression and spatial analysis were used to examine the probability and spatial patterns of bypass. Results: Overall, 39% of respondents bypass local health care. Similar to previous studies, dissatisfaction with local health care was found to increase the likelihood of bypass. Dissatisfaction with local shopping also increases the likelihood of bypass, while the number of friends in a community, and commonality with community reduce the likelihood of bypass. Other significant factors associated with bypass include age, income, health, and living in a highly rural community or one with high commuting flows. Conclusions: Our results suggest that outshopping theory, in which patients bundle services and shopping for added convenience, extends to primary health care selection. This implies that rural health care selection is multifaceted, and that in addition to perceived satisfaction with local health care, the quality of local shopping and levels of community attachment also influence bypass behavior.
Key words access to care, health services research, rural health, satisfaction with care, utilization of health services.
Bypass is a behavior in which patients receive health care from providers located farther away than the nearest health care provider to their residence.1-3 In urban areas where there are larger clusters of health care providers, bypassing a provider is a matter of choice for residents, with little consequence for the providers being bypassed. However, in rural areas with fewer providers, individuals must expand the distance in which they are willing to travel to obtain similar levels of choice.4 Unfortunately, bypassing local providers can also erode the financial
viability of local health care services.5 Over time, the reduced demand that accompanies bypass can lead to the loss of rural health care services and result in a “health care desert.”1-3 Consequently, understanding the factors associated with bypass is essential to policy makers and health practitioners seeking to improve rural health care. Although there is an emerging literature on rural bypass, a number of shortcomings remain. Most studies of bypass use hospital records and emphasize why a particular hospital was chosen over others. The main
c 2014 National Rural Health Association The Journal of Rural Health 00 (2014) 1–11
1
Rural Health Care Bypass Behavior
conclusion of this type of research is that patients bypass local providers to pursue better quality care.2,3,6-12 Although these studies provide insights into provider selection, they seldom have detailed information on personal attitudes toward local health care services, actual physician choices, and the socioeconomic factors that influence bypass behavior. Without this information, there are substantial limitations to the current literature. First, it is difficult to determine whether travel patterns reflect patient preference or resource constraints.13 Second, there are few studies that examine primary care physician (PCP) selection and why rural residents choose a PCP in or outside their community.6 Third, there is little research on the role of community characteristics on bypass behavior.14 Non-health-related behaviors, such as purchasing goods and services outside one’s community, could influence bypass behavior if individuals bundle health care with other consumer needs.6
The Social Context of Health Care Bypass Current research on rural bypass focuses on which health care services are most attractive to patients. This focus on health care providers overlooks the influence of the broader social context of health care selection.15 We argue that the decision to bypass local PCPs occurs at the intersection of perceived satisfaction with local health care and perceived satisfaction with other local services, such as shopping. Consequently, rural health care selection should be viewed as multifaceted and part of a shift in consumer shopping habits in which consumers seek to bundle goods and services.16 We use outshopping theory to frame rural health care selection as part of a broader shift in consumer shopping patterns. Outshopping occurs when consumers leave their local communities to obtain retail goods and services even when those goods and services are available locally. Outshopping theory has primarily been used to understand how the introduction of big-box retailers affects rural consumption patterns. For example, by offering both dry goods and groceries, big-box stores affect the economic sustainability of retail and grocery stores in surrounding communities because their convenience attracts consumers away from local businesses.17,18 Furthermore, and critical for the understanding of health care bypass, dissatisfaction with a single type of local good or service is related to outshopping for multiple goods and services.19-23 Using outshopping as a framework for rural bypass allows a more nuanced analysis of rural health care selection. If rural residents consider health care to be a consumer service, they may bypass local PCPs in favor of those located where they engage in other consumer
2
Sanders et al.
activities. This increases convenience by consolidating travel for health care with traditional consumption. Furthermore, outshopping theory asserts that bypassing local goods and services is less likely when individuals feel strong community attachment. Strong levels of community attachment can produce a desire to “buy local” even when nonlocal options are perceived to be more satisfactory.24,25 Therefore, rural health care selection can be viewed as part of a social process in which higher levels of community attachment are expected to reduce the likelihood of bypass behavior. This research seeks to address the shortcomings in the bypass literature and provide new insights into how community characteristics affect rural bypass behavior. Specifically, this research explores (1) how dissatisfaction with local health care affects rural PCP bypass behavior among rural residents, (2) how the availability of community-level measures of dissatisfaction with local shopping and community attachment contribute to the current understanding of bypass behavior, and (3) which communities have significant concentrations of bypass behavior that can potentially result in the formation of health care deserts. In sum, we draw on outshopping theory to argue that a rural resident’s decision to bypass local health care is not solely dependent on dissatisfaction with local health care. Instead, bypass should be considered as a multifaceted phenomenon that is the result of a variety of “push” and “pull” factors. Dissatisfaction with local shopping can push rural residents to bypass local PCPs as they bundle health care with other consumer goods and services. Strong community attachment creates an opposite pull on individuals and can help to negate the push of outshopping. Guided by the outshopping and bypass literature, we expect that in addition to dissatisfaction with local PCPs, dissatisfaction with local shopping will also significantly increase the likelihood of bypass, while strong community attachment will decrease the probability of bypass. We also anticipate that having negative views of both local health care and shopping amplifies the likelihood of bypass, whereas strong community attachment coupled with a dissatisfaction with local health care will dampen the likelihood of bypass.
Methods/Data This research uses the 2010 Montana Health Matters (MHM) study to examine rural bypass behavior. The MHM study gathered self–reported information on health care utilization and perceived satisfaction with local health care. To understand how health care decisions are affected by the respondents’ social context, demographic characteristics and measures of community attachment were also collected.
c 2014 National Rural Health Association The Journal of Rural Health 00 (2014) 1–11
Rural Health Care Bypass Behavior
Sanders et al.
The sample for this study was drawn using the United States Postal Service’s computerized Delivery Sequence File. This file contains all known addresses in Montana and served as the sampling frame. Given the addressbased sample design, MHM used a multimethod, 5-wave mail/telephone survey protocol and a small honorarium in the first questionnaire mailing to maximize response to the survey.26 Almost 90% of those surveyed responded by mail—the remaining 353 (10.1%) responded to the same questions over the phone with interviewers after failing to complete the mail survey. We tested for mode effects in the analyses but do not report them because the tests were nonsignificant. The multimethod design resulted in 3,512 respondents with 1,498 in rural areas, 1,521 in highly rural areas, and 493 in the urban comparison group. A conservative estimate of the response rate for the study was 52% (urban-47%; rural-52%; highly rural-54%).27 The weighting scheme developed for these data takes into account ineligible households, the multistage cluster sampling design, and survey nonresponse, making the final sample representative of the population of Montana. Of the 3,019 respondents in rural or highly rural areas, 479 were ineligible for this study due to missing information about their PCP. Remaining missing data were addressed using multiple imputations with chained equations. Using the mi impute command in Stata 13 (StatCorp LP, College Station, TX, USA), we obtained 50 imputed data sets. Imputed data sets were separated by 100 iterations because graphical diagnostics from preliminary imputations suggested that the imputation model converged well before this point. The final sample size was 2,540. Analyses were performed separately on each imputed data set and results for each data set were combined using Rubin’s rules via Stata’s mi estimate command.
Measurements Defining Bypass Previous definitions of bypass vary. Some research identifies bypass as traveling outside of one’s ZIP code or if the distance between the resident ZIP code and the ZIP code where health care was obtained is greater than 15 miles.3,28 ˙ENREF˙29 ZIP code-based definitions are limited because ZIP code boundaries change and are based on mail delivery patterns rather than community boundaries.29 Other work uses a threshold based on the average distance traveled for health care within a county to determine if a patient receives local or distant care.11,30 We improved the precision of bypass measurements by utilizing information from the MHM data to map the
c 2014 National Rural Health Association The Journal of Rural Health 00 (2014) 1–11
exact location of each respondent and their PCP. Next, the location of all other known PCPs, including Veteran Affairs (VA) and rural health clinics, were mapped using data from ESRI business analysis and the Centers for Medicare and Medicaid Provider of Service files.31,32 By mapping the location of all PCPs in Montana, we were able to determine if a respondent is bypassing local PCPs to obtain care. We utilized a 15-mile straight-line threshold to define bypass behavior. Although it does not provide a varying and relative measure of bypass like the local/distant research,11,30 using the 15-mile straight-line definition maintains comparability and consistency with previous research which helps to highlight how the addition of outshopping variables contribute to the understanding of bypass behavior.3,28,33,34 In addition, the 15-mile straight-line definition can account for PCP selection occurring within and across ZIP code and county borders. This is particularly important when health care is highly concentrated, as is the case in Montana. Therefore, respondents were coded as bypassing local PCPs if they: (1) travel greater than 15 miles to receive primary health care, and (2) there are other PCPs located less than 15 miles away that respondents do not use.
Push & Pull Variables Community push variables included dissatisfaction with local health care and shopping. Each was based on a 7point scale from 1 “exceptional” to 7 “badly needs improvement.” Responses were reverse coded so that higher scores indicate greater dissatisfaction. To assess community pull forces, we used measures of community attachment that commonly appear in outshopping studies.24,25 The number of close friends in one’s community was based on an open-ended question, and then responses were scaled into groups of 10 ranging from 0-10 to 71+. The fit in community question asked, “How well do you fit into your community?” Responses ranged from 1 “poorly” to 7 “well.” Commonality with community residents was based on a question asking, “How much do you have in common with most people in your community,” with 1 being “nothing” and 7 being “everything.”24,25
Controls Demographic controls included age (in years), sex (male = 1), marital status (married = 1), veteran status (veteran = 1), and education. The income variable is the total annual household income in $1,000. Self-reported health was measured on a 5-point scale where 1 is “poor” and 5 is “excellent.” The health insurance variable identified if respondents have employer/union or direct purchase
3
Rural Health Care Bypass Behavior
health plans, Medicare or Medicaid, VA plans, other types of coverage, like Indian Health Services and federal retired employee health insurances, or no health insurance. To assess intercommunity ties, a variable specifying if a respondent lives in a rural community with high commuter flows to larger regional economic centers was included. Based on the Rural-Urban Commuting Area Codes, respondents were coded as living in a community with high commuting flows if they are located in a census track with a primary commuting flow of 30% or greater to a metropolitan, micropolitan, or small-town core.35 A community PCP gravity score was used to account for rural health care access. Earlier research indicates that individuals are often attracted to health care in larger economic centers with a greater number of PCPs. However, the attraction of these areas diminishes as the distance increases between the larger economic center and the individual’s residence.36 The gravity score was calculated by multiplying the number of PCPs in the ZIP code where respondents receive their health care by the number of PCPs in the ZIP code of residence. This product was then divided by the square of the distance traveled and multiplied by 100. This simple gravity model helps to account for the distance and number of PCPs in surrounding communities. Highly rural residence was identified as living in a county with a population density of 6 people or fewer per square mile. Rural areas were defined as all nonurban and nonhighly rural counties.
Sanders et al.
between dissatisfaction with local health care and bypass behavior.37 Sample weights and clustering of the data were accounted for in all analyses using Stata’s svy estimation prefix to ensure that point estimates were representative of Montana and significance tests were not biased. Finally, a hot-spot analysis addresses our third research question by identifying which communities have significant clusters of bypass behavior and are at the greatest risk of becoming health care deserts. Hot-spot analysis38 determines if a respondent is part of a spatially significant cluster of respondents with a high predicted probability of bypass. This is done by first using the results of a logistic regression model to calculate the predicted probability of bypass for each respondent. Then we use Moran’s I to determine the geographic distance between respondents that was optimal to identify clustering in the data.39 For these data, the peak value of Moran’s I is 3.5 miles. Third, the G i∗ is estimated for each respondent: n
Wi, j x j − X¯ j=1 ⎤ ⎡
G i∗ = ⎡ ⎢ ⎢ ⎣
⎥⎢ ⎥⎢ ⎦⎣
n n j−1 n
x 2j
−( X¯ )
n
n
n
Wi, j
j=1
W 2 i, j −
j−1
n j−1
n−1
2 ⎤ , Wi, j
⎥ ⎥ ⎦
2
After estimating the prevalence of bypass, we conducted bivariate analyses examining the relationship between community characteristics and rural bypass behavior. Next, a series of multivariate logistic regression models were estimated to explore rural bypass behavior. Model 1 addresses our first research question and includes community-level variables commonly used in existing bypass research, that is, dissatisfaction with local health care, demographics, and other control variables. Model 2 addresses our second research question by examining the effect of outshopping on bypass behavior by adding community push/pull variables. Further exploring bypass behavior and outshopping, Model 3 includes interactions between dissatisfaction with local health care (the main explanation of bypass in past literature) and the significant push and pull variables. These interactions tested whether (1) dissatisfaction with local shopping amplifies the relationship between dissatisfaction with local health care and the likelihood of bypass, and (2) if the pull of community attachment dampens the relationship
where i is the respondent in question, n is the total number of respondents within a 3.5-mile radius of i (based on Moran’s I), xj is the predicted probability of bypass for respondent j, wij is the geographic distance between respondents i and j, and X¯ is the mean probability of bypass for all j. The G i∗ is essentially a z-score that compares the observed density of the model-based probability of bypass to the expected density under the null hypothesis of no spatial correlation. If the G i∗ is greater than 1.96, the observed density is significantly different (ie, P < .05) than the expected value, resulting in the conclusion that the respondent is located in a spatial clustering of bypass. Consequently, groupings of respondents with significant G i∗ values indicate that there are spatial concentrations of respondents with high predicted probabilities of bypass, and the grouping is too large to be the result of random chance. This analysis used the location of each respondent instead of the aggregated political boundaries, like counties, which are often used in hot-spot analyses. By using point data, our final map shows the location of significant hot-spot clusters and nonsignificant respondents that are hidden in aggregate analyses. The results of the hot-spot analysis were then overlaid on a map illustrating the number of PCPs in each county and all the cities in Montana with a population of 20,000 or
4
c 2014 National Rural Health Association The Journal of Rural Health 00 (2014) 1–11
Analyses
Sanders et al.
greater. ArcGIS 10 (Esri, Redlands, CA, USA) was used to perform the hot-spot analysis and create the final map.
Rural Health Care Bypass Behavior
Table 2 presents the results of the logistic regression analyses. Consistent with previous research, Model 1 shows that dissatisfaction with local health care significantly increases the odds of bypass (OR = 1.15) net of other individual- and community-level controls. Living in communities with high commuting flows makes bypass more likely compared to living in communities with low commuting flows (OR = 2.22), but the gravity score is nonsignificant. Respondents age 65 or older are more likely to bypass compared to respondents 18-24 (OR = 1.57). There is a positive and significant finding for income (OR = 1.03) and self-reported health (OR = 1.08). Residents living in highly rural areas are more likely to bypass than those in rural communities (OR = 1.77). The other controls were nonsignificant.
Model 2 introduces the community push/pull variables to test whether outshopping theory can inform health care bypass. Dissatisfaction with local health care continues to have a significant relationship with bypass (OR = 1.12). Dissatisfaction with local shopping significantly increases the likelihood of bypass behavior (OR = 1.09), and the size of the relationship is similar to that of dissatisfaction with local health care. These relationships are positive, implying that both function as push variables with respect to bypass. The community pull variables are negatively related to bypass, as expected. Identifying more friends in the community reduces the odds of bypass (OR = 0.93). Commonality with community also reduces the odds of bypass (OR = 0.83). Although the fit in community variable is not significant, the significance of the community shopping, number of friends, and commonality with community variables suggests that the outshopping variables significantly affect bypass behavior. To further understand how outshopping affects health care selection, Model 3 includes interactions between dissatisfaction with local health care (push factor) and the significant push and pull variables from Model 2. The push-push interaction between dissatisfaction with local health care and shopping (OR = 1.14) is positive and significant, which indicates that the likelihood of bypass increases disproportionately as both push factors become greater. Figure 1 plots the predicted probabilities of bypass for the highest and lowest values for dissatisfaction with local shopping across the range of dissatisfaction with local health care. The divergence of the plots for high and low satisfaction with shopping as dissatisfaction with health care increases shows that the push effect from dissatisfaction with local health care on bypass increases with higher levels of dissatisfaction with local shopping. In other words, dissatisfaction with local health care becomes an even more important factor in health care bypass when rural residents are also dissatisfied with local shopping. The interaction of dissatisfaction with local health care (push factor) with number of friends in the community (pull factor) is nonsignificant. However, the interaction of dissatisfaction with local health care and commonality with community is negative and significant. The predicted probabilities of bypass for those with everything and nothing in common with their community are plotted across the range of dissatisfaction with local health care, revealing that those with everything in common with their community maintain a low probability of bypass even as the level of dissatisfaction with local health care increases (see Figure 2). However, the probability of bypass rises for those with nothing in common as dissatisfaction with local health care increases. Therefore,
c 2014 National Rural Health Association The Journal of Rural Health 00 (2014) 1–11
5
Results Table 1 presents the results of bivariate analyses of the community push and pull variables with bypass. Overall, 39% of the sample exhibited bypass behavior. Dissatisfaction with local health care is fairly evenly distributed in the sample, although slightly more respondents reported being relatively dissatisfied compared to being satisfied. The percentage of residents that bypass local health care increases with higher levels of dissatisfaction. Among respondents who are the most dissatisfied with local health care, over 45% bypass local PCPs. The other community push variable, dissatisfaction with local shopping, and the community pull variables all support outshopping as an important consideration for health care bypass. Dissatisfaction with local shopping is also fairly evenly distributed in the sample, and its relationship with health care bypass is similar to that of dissatisfaction with local health care. Over 41% of respondents who are most dissatisfied with local shopping bypass local health care, compared to only 25.22% for the most satisfied. Peaking with the 30% of respondents with 10 or fewer friends in their community, the overall trend of bypass decreases as the number of friends in a community increases. Only 14.29% of respondents with 71+ friends bypassed local PCPs. Among respondents reporting the strongest fit in their community, only 17.67% reported bypassing local health care, but 35% of those with the poorest fit did. The commonality with community variable followed this same pattern.
Results for Logistic Regression Models
Rural Health Care Bypass Behavior
Sanders et al.
Table 1 Community Push and Pull Characteristics and Rural Bypass Behavior
Community push characteristics Dissatisfaction with local health care 1-Satisfied 2 3 4 5 6 7-Dissatisfied Dissatisfaction with local shopping 1-Satisfied 2 3 4 5 6 7-Dissatisfied Community pull characteristics Number of friends in community 0-10 11-20 21-30 31-40 41-50 51-60 61-70 71+ Fit in community 1-Poor fit 2 3 4 5 6 7-Strong fit Commonality with community 1-Nothing 2 3 4 5 6 7-Everything
% Respondents
% Bypass
P Value
9.63 13.85 12.21 13.51 16.21 17.04 17.55
22.31 26.51 29.87 30.39 32.89 39.81 45.13
Rural Health Care Bypass Behavior: How Community and Spatial Characteristics Affect Primary Health Care Selection Scott R. Sanders, MPA, PhD;1 Lance D. Erickson, PhD;1 Vaughn R.A. Call, PhD;1 Matthew L. McKnight, MS;1 & Dawson W. Hedges, MD2 1 Department of Sociology, Brigham Young University, Provo, Utah 2 Department of Psychology and the Neuroscience Center, Brigham Young University, Provo, Utah
Abstract Purpose: (1) To assess the prevalence of rural primary care physician (PCP) Funding: Funding for this research comes from the Brigham Young University Gerontology Center. For further information, contact: Scott R. Sanders, MPA, PhD, 2039 JFSB, Department of Sociology, Brigham Young University, Provo, UT 84602; e-mail: [email protected]. doi: 10.1111/jrh.12093
bypass, a behavior in which residents travel farther than necessary to obtain health care, (2) To examine the role of community and non-health-carerelated characteristics on bypass behavior, and (3) To analyze spatial bypass patterns to determine which rural communities are most affected by bypass. Methods: Data came from the Montana Health Matters survey, which gathered self-reported information from Montana residents on their health care utilization, satisfaction with health care services, and community and demographic characteristics. Logistic regression and spatial analysis were used to examine the probability and spatial patterns of bypass. Results: Overall, 39% of respondents bypass local health care. Similar to previous studies, dissatisfaction with local health care was found to increase the likelihood of bypass. Dissatisfaction with local shopping also increases the likelihood of bypass, while the number of friends in a community, and commonality with community reduce the likelihood of bypass. Other significant factors associated with bypass include age, income, health, and living in a highly rural community or one with high commuting flows. Conclusions: Our results suggest that outshopping theory, in which patients bundle services and shopping for added convenience, extends to primary health care selection. This implies that rural health care selection is multifaceted, and that in addition to perceived satisfaction with local health care, the quality of local shopping and levels of community attachment also influence bypass behavior.
Key words access to care, health services research, rural health, satisfaction with care, utilization of health services.
Bypass is a behavior in which patients receive health care from providers located farther away than the nearest health care provider to their residence.1-3 In urban areas where there are larger clusters of health care providers, bypassing a provider is a matter of choice for residents, with little consequence for the providers being bypassed. However, in rural areas with fewer providers, individuals must expand the distance in which they are willing to travel to obtain similar levels of choice.4 Unfortunately, bypassing local providers can also erode the financial
viability of local health care services.5 Over time, the reduced demand that accompanies bypass can lead to the loss of rural health care services and result in a “health care desert.”1-3 Consequently, understanding the factors associated with bypass is essential to policy makers and health practitioners seeking to improve rural health care. Although there is an emerging literature on rural bypass, a number of shortcomings remain. Most studies of bypass use hospital records and emphasize why a particular hospital was chosen over others. The main
c 2014 National Rural Health Association The Journal of Rural Health 00 (2014) 1–11
1
Rural Health Care Bypass Behavior
conclusion of this type of research is that patients bypass local providers to pursue better quality care.2,3,6-12 Although these studies provide insights into provider selection, they seldom have detailed information on personal attitudes toward local health care services, actual physician choices, and the socioeconomic factors that influence bypass behavior. Without this information, there are substantial limitations to the current literature. First, it is difficult to determine whether travel patterns reflect patient preference or resource constraints.13 Second, there are few studies that examine primary care physician (PCP) selection and why rural residents choose a PCP in or outside their community.6 Third, there is little research on the role of community characteristics on bypass behavior.14 Non-health-related behaviors, such as purchasing goods and services outside one’s community, could influence bypass behavior if individuals bundle health care with other consumer needs.6
The Social Context of Health Care Bypass Current research on rural bypass focuses on which health care services are most attractive to patients. This focus on health care providers overlooks the influence of the broader social context of health care selection.15 We argue that the decision to bypass local PCPs occurs at the intersection of perceived satisfaction with local health care and perceived satisfaction with other local services, such as shopping. Consequently, rural health care selection should be viewed as multifaceted and part of a shift in consumer shopping habits in which consumers seek to bundle goods and services.16 We use outshopping theory to frame rural health care selection as part of a broader shift in consumer shopping patterns. Outshopping occurs when consumers leave their local communities to obtain retail goods and services even when those goods and services are available locally. Outshopping theory has primarily been used to understand how the introduction of big-box retailers affects rural consumption patterns. For example, by offering both dry goods and groceries, big-box stores affect the economic sustainability of retail and grocery stores in surrounding communities because their convenience attracts consumers away from local businesses.17,18 Furthermore, and critical for the understanding of health care bypass, dissatisfaction with a single type of local good or service is related to outshopping for multiple goods and services.19-23 Using outshopping as a framework for rural bypass allows a more nuanced analysis of rural health care selection. If rural residents consider health care to be a consumer service, they may bypass local PCPs in favor of those located where they engage in other consumer
2
Sanders et al.
activities. This increases convenience by consolidating travel for health care with traditional consumption. Furthermore, outshopping theory asserts that bypassing local goods and services is less likely when individuals feel strong community attachment. Strong levels of community attachment can produce a desire to “buy local” even when nonlocal options are perceived to be more satisfactory.24,25 Therefore, rural health care selection can be viewed as part of a social process in which higher levels of community attachment are expected to reduce the likelihood of bypass behavior. This research seeks to address the shortcomings in the bypass literature and provide new insights into how community characteristics affect rural bypass behavior. Specifically, this research explores (1) how dissatisfaction with local health care affects rural PCP bypass behavior among rural residents, (2) how the availability of community-level measures of dissatisfaction with local shopping and community attachment contribute to the current understanding of bypass behavior, and (3) which communities have significant concentrations of bypass behavior that can potentially result in the formation of health care deserts. In sum, we draw on outshopping theory to argue that a rural resident’s decision to bypass local health care is not solely dependent on dissatisfaction with local health care. Instead, bypass should be considered as a multifaceted phenomenon that is the result of a variety of “push” and “pull” factors. Dissatisfaction with local shopping can push rural residents to bypass local PCPs as they bundle health care with other consumer goods and services. Strong community attachment creates an opposite pull on individuals and can help to negate the push of outshopping. Guided by the outshopping and bypass literature, we expect that in addition to dissatisfaction with local PCPs, dissatisfaction with local shopping will also significantly increase the likelihood of bypass, while strong community attachment will decrease the probability of bypass. We also anticipate that having negative views of both local health care and shopping amplifies the likelihood of bypass, whereas strong community attachment coupled with a dissatisfaction with local health care will dampen the likelihood of bypass.
Methods/Data This research uses the 2010 Montana Health Matters (MHM) study to examine rural bypass behavior. The MHM study gathered self–reported information on health care utilization and perceived satisfaction with local health care. To understand how health care decisions are affected by the respondents’ social context, demographic characteristics and measures of community attachment were also collected.
c 2014 National Rural Health Association The Journal of Rural Health 00 (2014) 1–11
Rural Health Care Bypass Behavior
Sanders et al.
The sample for this study was drawn using the United States Postal Service’s computerized Delivery Sequence File. This file contains all known addresses in Montana and served as the sampling frame. Given the addressbased sample design, MHM used a multimethod, 5-wave mail/telephone survey protocol and a small honorarium in the first questionnaire mailing to maximize response to the survey.26 Almost 90% of those surveyed responded by mail—the remaining 353 (10.1%) responded to the same questions over the phone with interviewers after failing to complete the mail survey. We tested for mode effects in the analyses but do not report them because the tests were nonsignificant. The multimethod design resulted in 3,512 respondents with 1,498 in rural areas, 1,521 in highly rural areas, and 493 in the urban comparison group. A conservative estimate of the response rate for the study was 52% (urban-47%; rural-52%; highly rural-54%).27 The weighting scheme developed for these data takes into account ineligible households, the multistage cluster sampling design, and survey nonresponse, making the final sample representative of the population of Montana. Of the 3,019 respondents in rural or highly rural areas, 479 were ineligible for this study due to missing information about their PCP. Remaining missing data were addressed using multiple imputations with chained equations. Using the mi impute command in Stata 13 (StatCorp LP, College Station, TX, USA), we obtained 50 imputed data sets. Imputed data sets were separated by 100 iterations because graphical diagnostics from preliminary imputations suggested that the imputation model converged well before this point. The final sample size was 2,540. Analyses were performed separately on each imputed data set and results for each data set were combined using Rubin’s rules via Stata’s mi estimate command.
Measurements Defining Bypass Previous definitions of bypass vary. Some research identifies bypass as traveling outside of one’s ZIP code or if the distance between the resident ZIP code and the ZIP code where health care was obtained is greater than 15 miles.3,28 ˙ENREF˙29 ZIP code-based definitions are limited because ZIP code boundaries change and are based on mail delivery patterns rather than community boundaries.29 Other work uses a threshold based on the average distance traveled for health care within a county to determine if a patient receives local or distant care.11,30 We improved the precision of bypass measurements by utilizing information from the MHM data to map the
c 2014 National Rural Health Association The Journal of Rural Health 00 (2014) 1–11
exact location of each respondent and their PCP. Next, the location of all other known PCPs, including Veteran Affairs (VA) and rural health clinics, were mapped using data from ESRI business analysis and the Centers for Medicare and Medicaid Provider of Service files.31,32 By mapping the location of all PCPs in Montana, we were able to determine if a respondent is bypassing local PCPs to obtain care. We utilized a 15-mile straight-line threshold to define bypass behavior. Although it does not provide a varying and relative measure of bypass like the local/distant research,11,30 using the 15-mile straight-line definition maintains comparability and consistency with previous research which helps to highlight how the addition of outshopping variables contribute to the understanding of bypass behavior.3,28,33,34 In addition, the 15-mile straight-line definition can account for PCP selection occurring within and across ZIP code and county borders. This is particularly important when health care is highly concentrated, as is the case in Montana. Therefore, respondents were coded as bypassing local PCPs if they: (1) travel greater than 15 miles to receive primary health care, and (2) there are other PCPs located less than 15 miles away that respondents do not use.
Push & Pull Variables Community push variables included dissatisfaction with local health care and shopping. Each was based on a 7point scale from 1 “exceptional” to 7 “badly needs improvement.” Responses were reverse coded so that higher scores indicate greater dissatisfaction. To assess community pull forces, we used measures of community attachment that commonly appear in outshopping studies.24,25 The number of close friends in one’s community was based on an open-ended question, and then responses were scaled into groups of 10 ranging from 0-10 to 71+. The fit in community question asked, “How well do you fit into your community?” Responses ranged from 1 “poorly” to 7 “well.” Commonality with community residents was based on a question asking, “How much do you have in common with most people in your community,” with 1 being “nothing” and 7 being “everything.”24,25
Controls Demographic controls included age (in years), sex (male = 1), marital status (married = 1), veteran status (veteran = 1), and education. The income variable is the total annual household income in $1,000. Self-reported health was measured on a 5-point scale where 1 is “poor” and 5 is “excellent.” The health insurance variable identified if respondents have employer/union or direct purchase
3
Rural Health Care Bypass Behavior
health plans, Medicare or Medicaid, VA plans, other types of coverage, like Indian Health Services and federal retired employee health insurances, or no health insurance. To assess intercommunity ties, a variable specifying if a respondent lives in a rural community with high commuter flows to larger regional economic centers was included. Based on the Rural-Urban Commuting Area Codes, respondents were coded as living in a community with high commuting flows if they are located in a census track with a primary commuting flow of 30% or greater to a metropolitan, micropolitan, or small-town core.35 A community PCP gravity score was used to account for rural health care access. Earlier research indicates that individuals are often attracted to health care in larger economic centers with a greater number of PCPs. However, the attraction of these areas diminishes as the distance increases between the larger economic center and the individual’s residence.36 The gravity score was calculated by multiplying the number of PCPs in the ZIP code where respondents receive their health care by the number of PCPs in the ZIP code of residence. This product was then divided by the square of the distance traveled and multiplied by 100. This simple gravity model helps to account for the distance and number of PCPs in surrounding communities. Highly rural residence was identified as living in a county with a population density of 6 people or fewer per square mile. Rural areas were defined as all nonurban and nonhighly rural counties.
Sanders et al.
between dissatisfaction with local health care and bypass behavior.37 Sample weights and clustering of the data were accounted for in all analyses using Stata’s svy estimation prefix to ensure that point estimates were representative of Montana and significance tests were not biased. Finally, a hot-spot analysis addresses our third research question by identifying which communities have significant clusters of bypass behavior and are at the greatest risk of becoming health care deserts. Hot-spot analysis38 determines if a respondent is part of a spatially significant cluster of respondents with a high predicted probability of bypass. This is done by first using the results of a logistic regression model to calculate the predicted probability of bypass for each respondent. Then we use Moran’s I to determine the geographic distance between respondents that was optimal to identify clustering in the data.39 For these data, the peak value of Moran’s I is 3.5 miles. Third, the G i∗ is estimated for each respondent: n
Wi, j x j − X¯ j=1 ⎤ ⎡
G i∗ = ⎡ ⎢ ⎢ ⎣
⎥⎢ ⎥⎢ ⎦⎣
n n j−1 n
x 2j
−( X¯ )
n
n
n
Wi, j
j=1
W 2 i, j −
j−1
n j−1
n−1
2 ⎤ , Wi, j
⎥ ⎥ ⎦
2
After estimating the prevalence of bypass, we conducted bivariate analyses examining the relationship between community characteristics and rural bypass behavior. Next, a series of multivariate logistic regression models were estimated to explore rural bypass behavior. Model 1 addresses our first research question and includes community-level variables commonly used in existing bypass research, that is, dissatisfaction with local health care, demographics, and other control variables. Model 2 addresses our second research question by examining the effect of outshopping on bypass behavior by adding community push/pull variables. Further exploring bypass behavior and outshopping, Model 3 includes interactions between dissatisfaction with local health care (the main explanation of bypass in past literature) and the significant push and pull variables. These interactions tested whether (1) dissatisfaction with local shopping amplifies the relationship between dissatisfaction with local health care and the likelihood of bypass, and (2) if the pull of community attachment dampens the relationship
where i is the respondent in question, n is the total number of respondents within a 3.5-mile radius of i (based on Moran’s I), xj is the predicted probability of bypass for respondent j, wij is the geographic distance between respondents i and j, and X¯ is the mean probability of bypass for all j. The G i∗ is essentially a z-score that compares the observed density of the model-based probability of bypass to the expected density under the null hypothesis of no spatial correlation. If the G i∗ is greater than 1.96, the observed density is significantly different (ie, P < .05) than the expected value, resulting in the conclusion that the respondent is located in a spatial clustering of bypass. Consequently, groupings of respondents with significant G i∗ values indicate that there are spatial concentrations of respondents with high predicted probabilities of bypass, and the grouping is too large to be the result of random chance. This analysis used the location of each respondent instead of the aggregated political boundaries, like counties, which are often used in hot-spot analyses. By using point data, our final map shows the location of significant hot-spot clusters and nonsignificant respondents that are hidden in aggregate analyses. The results of the hot-spot analysis were then overlaid on a map illustrating the number of PCPs in each county and all the cities in Montana with a population of 20,000 or
4
c 2014 National Rural Health Association The Journal of Rural Health 00 (2014) 1–11
Analyses
Sanders et al.
greater. ArcGIS 10 (Esri, Redlands, CA, USA) was used to perform the hot-spot analysis and create the final map.
Rural Health Care Bypass Behavior
Table 2 presents the results of the logistic regression analyses. Consistent with previous research, Model 1 shows that dissatisfaction with local health care significantly increases the odds of bypass (OR = 1.15) net of other individual- and community-level controls. Living in communities with high commuting flows makes bypass more likely compared to living in communities with low commuting flows (OR = 2.22), but the gravity score is nonsignificant. Respondents age 65 or older are more likely to bypass compared to respondents 18-24 (OR = 1.57). There is a positive and significant finding for income (OR = 1.03) and self-reported health (OR = 1.08). Residents living in highly rural areas are more likely to bypass than those in rural communities (OR = 1.77). The other controls were nonsignificant.
Model 2 introduces the community push/pull variables to test whether outshopping theory can inform health care bypass. Dissatisfaction with local health care continues to have a significant relationship with bypass (OR = 1.12). Dissatisfaction with local shopping significantly increases the likelihood of bypass behavior (OR = 1.09), and the size of the relationship is similar to that of dissatisfaction with local health care. These relationships are positive, implying that both function as push variables with respect to bypass. The community pull variables are negatively related to bypass, as expected. Identifying more friends in the community reduces the odds of bypass (OR = 0.93). Commonality with community also reduces the odds of bypass (OR = 0.83). Although the fit in community variable is not significant, the significance of the community shopping, number of friends, and commonality with community variables suggests that the outshopping variables significantly affect bypass behavior. To further understand how outshopping affects health care selection, Model 3 includes interactions between dissatisfaction with local health care (push factor) and the significant push and pull variables from Model 2. The push-push interaction between dissatisfaction with local health care and shopping (OR = 1.14) is positive and significant, which indicates that the likelihood of bypass increases disproportionately as both push factors become greater. Figure 1 plots the predicted probabilities of bypass for the highest and lowest values for dissatisfaction with local shopping across the range of dissatisfaction with local health care. The divergence of the plots for high and low satisfaction with shopping as dissatisfaction with health care increases shows that the push effect from dissatisfaction with local health care on bypass increases with higher levels of dissatisfaction with local shopping. In other words, dissatisfaction with local health care becomes an even more important factor in health care bypass when rural residents are also dissatisfied with local shopping. The interaction of dissatisfaction with local health care (push factor) with number of friends in the community (pull factor) is nonsignificant. However, the interaction of dissatisfaction with local health care and commonality with community is negative and significant. The predicted probabilities of bypass for those with everything and nothing in common with their community are plotted across the range of dissatisfaction with local health care, revealing that those with everything in common with their community maintain a low probability of bypass even as the level of dissatisfaction with local health care increases (see Figure 2). However, the probability of bypass rises for those with nothing in common as dissatisfaction with local health care increases. Therefore,
c 2014 National Rural Health Association The Journal of Rural Health 00 (2014) 1–11
5
Results Table 1 presents the results of bivariate analyses of the community push and pull variables with bypass. Overall, 39% of the sample exhibited bypass behavior. Dissatisfaction with local health care is fairly evenly distributed in the sample, although slightly more respondents reported being relatively dissatisfied compared to being satisfied. The percentage of residents that bypass local health care increases with higher levels of dissatisfaction. Among respondents who are the most dissatisfied with local health care, over 45% bypass local PCPs. The other community push variable, dissatisfaction with local shopping, and the community pull variables all support outshopping as an important consideration for health care bypass. Dissatisfaction with local shopping is also fairly evenly distributed in the sample, and its relationship with health care bypass is similar to that of dissatisfaction with local health care. Over 41% of respondents who are most dissatisfied with local shopping bypass local health care, compared to only 25.22% for the most satisfied. Peaking with the 30% of respondents with 10 or fewer friends in their community, the overall trend of bypass decreases as the number of friends in a community increases. Only 14.29% of respondents with 71+ friends bypassed local PCPs. Among respondents reporting the strongest fit in their community, only 17.67% reported bypassing local health care, but 35% of those with the poorest fit did. The commonality with community variable followed this same pattern.
Results for Logistic Regression Models
Rural Health Care Bypass Behavior
Sanders et al.
Table 1 Community Push and Pull Characteristics and Rural Bypass Behavior
Community push characteristics Dissatisfaction with local health care 1-Satisfied 2 3 4 5 6 7-Dissatisfied Dissatisfaction with local shopping 1-Satisfied 2 3 4 5 6 7-Dissatisfied Community pull characteristics Number of friends in community 0-10 11-20 21-30 31-40 41-50 51-60 61-70 71+ Fit in community 1-Poor fit 2 3 4 5 6 7-Strong fit Commonality with community 1-Nothing 2 3 4 5 6 7-Everything
% Respondents
% Bypass
P Value
9.63 13.85 12.21 13.51 16.21 17.04 17.55
22.31 26.51 29.87 30.39 32.89 39.81 45.13
