Projecting the Impact of the Affordable Care Act Provisions on Accessibility and Availability of Primary Care Providers for the Adult Population in Georgia

Abstract

Objectives. To evaluate how met need for accessibility and availability of primary care among nonelderly individuals in Georgia will be affected by the Patient Protection and Affordable Care Act (ACA) over the next 10 years.

Methods. We used a stock-and-flow model to predict the number of available visits from 2013 to 2025, regression models to project needed visits, and an optimization model to estimate met need. The outputs of these models were used to estimate unmet need and the availability and accessibility of primary care.

Results. Our findings showed that the number of primary care providers will increase by 9.2% to 11.7% by 2025 and that the number of needed visits will increase by 20%. Under Medicaid expansion, the percentage of met need will increase from 67% to 80%. Accessibility will improve by 20% under expansion, and availability will decrease by 13% to 19% under expansion.

Conclusions. The ACAs’ provisions will reduce unmet need and positively affect accessibility while reducing availability in some communities. Increased need because of a larger Medicaid population under Medicaid expansion will not be a significant burden on the privately insured population.

Health care access has been on the US national policy agenda since the 1967 report of the National Advisory Commission on Health Manpower.¹ Several policy interventions have been implemented and evaluated to improve health care access,² most recently the provisions outlined in the Patient Protection and Affordable Care Act (ACA; Pub L No. 111–148). One primary emphasis of the ACA is to eliminate disparities in financial access to health care.

Although affordability (i.e., financial access) is an essential dimension of health care access, other dimensions such as availability and accessibility are equally relevant in eliminating health disparities. Availability refers to the opportunity patients have to choose among health care providers,³ whereas accessibility refers to the time or distance barriers faced by patients in reaching their providers. Spatial access, referring to availability and accessibility in combination,³ is critical in promoting preventive care and reducing severe health outcomes. Lack of spatial access to health care can lead to higher costs, more frequent emergency outbreaks, and inconsistency in health outcomes.

Because the ACA’s implementation will transform health care delivery in ways that are not fully understood, it is expected to have unintended consequences that could counteract its overall benefits. For example, although the ACA primarily focuses on financial access, it will also affect other forms of health care access. Availability and accessibility are functions of the provider network (supply) and of the patients accessing health care services (demand and need), and these elements will be affected by the main provisions of the ACA. A number of studies^4–8 forecast that the resulting increase in demand for care attributable to the ACA provisions may not be adequately supported by the supply of health care services, with, for example, an estimated need for approximately 20 400 primary care physicians by 2020.⁶ This situation is a concern because too few physicians or an inadequate supply could lead to services being delayed or forgone altogether. Such delays could worsen health conditions, eventually resulting in increases in severe health outcomes.

We sought to project how implementation of the ACA will affect spatial access among nonelderly adults geographically and over time. We addressed the following questions: What is the projected impact of the ACA, in the absence of Medicaid expansion, on the availability and accessibility of primary care providers? What is the projected burden of opting for Medicaid expansion on the availability and accessibility of primary care providers? Will a higher rate of Medicaid insurance acceptance by providers or an increase in the overall supply result in improvements in the met need for primary care services?

We used 3 models in making our projections: a supply projection model, a need projection model, and an optimization model. The supply and need models predicted the total numbers of available (supply) and needed primary care visits for each year from 2013 (the baseline year) until 2025. The optimization model estimated met need by matching projected supply and need under a series of access constraints. Availability and accessibility were projected at the census tract level. Although we based our study on the state of Georgia, the proposed models can be generalized and applied to other US states.

METHODS

The study population consisted of all adults 19 to 64 years of age living in Georgia. We excluded the Medicare population because the Medicaid and exchange markets primarily affect the insurance status of individuals younger than 65 years. We also excluded children because the Medicaid expansion does not apply to this group.

Data Sources

The data sources for our supply projection model included the 2013 National Plan & Provider Enumeration System,⁹ Georgia Board for Physician Workforce 2013 graduate medical education data¹⁰ and 2008 basic physician reports,¹¹ data on deaths from the Centers for Disease Control and Prevention,¹² the 2007 to 2011 American Community Survey (ACS) county-to-county migration flow educational attainment table,¹³ and the 2012 ACS educational attainment table.¹⁴

Data sources used for need projections included residential population projections from the Georgia Governor’s Office of Planning and Budget,¹⁵ the 2010 census and the 2012 ACS,¹⁶ estimates from the Kaiser Family Foundation and Urban Institute,¹⁷ and Congressional Budget Office estimates.¹⁸

Medicaid Analytic eXtract patient-level claims files from 2009 served as an additional data source for the optimization model.¹⁹

We used Texas A&M Geocoding Services data²⁰ to geocode provider addresses and the ArcGIS Network Analyst²¹ to compute street-network distances between census tract centroids and provider addresses.

Supply Projection Model

According to the Institute of Medicine’s definition,²² primary care includes general and family medicine and general internal medicine, and primary care providers include physicians, nurse practitioners (NPs), and physician assistants (PAs). The supply projection model provided county-level active workforce projections, projected supply distributions across census tracts, and available supply in terms of number of visits, all projected yearly.

We used a stock-and-flow model to compute the total number (stock) of active physicians yearly considering the current availability level of physicians in addition to the net flow; the latter represents the difference between new entrants into the workforce (new graduates from residency programs and immigrants) and those exiting the workforce as a result of retirement, death, profession change, or emigration.⁶ The model included 2 modules, the student and workforce modules. The student module was used to determine the number of students completing graduate medical education in Georgia^23,24 and entering the workforce. We used data from the Georgia Board for Physician Workforce¹⁰ to estimate the input parameters of the module, including the percentage of graduates practicing outside of Georgia or choosing to practice in specialties other than primary care (Appendix A, Table A, available as a supplement to the online version of this article at http://www.ajph.org).

The workforce module was used to determine the age and county of each physician currently working. Physicians 35 to 64 years of age were presumed to transition between different workforce areas within 5-year age intervals. We assumed that at the age of 75 years physicians either retire or enter administrative roles. New physicians entered the stock of physicians younger than 35 years after graduating, and new graduates were allocated to each county according to the county’s baseline distribution of primary care physicians. To incorporate physician attrition, each age stock included an additional outflow component according to which physicians departed the workforce. To incorporate immigration, each age stock included an additional inflow component under which providers entered the workforce (Appendix A, Table B).

A stock of NPs and PAs was added to the workforce on the basis of data derived from previously published studies. Workforce growth for PAs was based on national projections.²⁵ A constant number of PAs was added each year to obtain 2013 to 2025 percentage growth figures. We applied a similar procedure to calculate the total number of NPs each year according to the projection of registered nurses in Georgia.²⁶

We used a recursive sampling approach to distribute the projected supply surplus of new providers at the county level. For each selected tract, we assigned locations to new providers by sampling from the existing locations in the provider network. We converted workforce estimates at the census tract level to number of available visits by determining the average number of yearly visits per provider according to age group and gender. We considered 3 supply growth scenarios in our calculations: 0% (constant), 12% (medium), and 30% (high) growth in the number of enrolled graduate students.

We estimated provider acceptance of publicly insured patients according to the process outlined by Nobles et al.²⁷ We accounted for possible changes in physicians’ Medicaid participation attributable to the Medicaid parity program²⁸ by considering Medicaid acceptance ratio increases in only 2013 and 2014 (without parity) and increases over the entire study period (with parity). We estimated the increase in the Medicaid acceptance ratio due to the Medicaid parity program as 13.03%.²⁸ Complete details on the supply projection model are presented in Appendix A.

Need Projection Model

We projected need rather than demand for medical care to estimate potential spatial access to primary care. Demand depends on several factors such as income and education. According to the model outlined by Grossman,²⁹ highly educated people are expected to be more efficient producers of health. Our estimates of need were driven by demographic characteristics and health status.^30,31 Regression models were used to project the total insured and uninsured populations in Georgia at the census tract level yearly and by age and gender. We employed utilization ratios provided by Petterson et al.⁷ in converting projected populations to projected numbers of needed visits.^30,31

We projected need for 3 age groups (19–24, 25–44, and 45–64 years) defined according to insurance status and gender. Insured individuals were divided into those eligible for Medicaid and those with private insurance. Need projections were performed under 3 different scenarios: no ACA implementation, ACA implementation with expansion, and ACA implementation with nonexpansion of Medicaid eligibility. Under the no-ACA and nonexpansion of Medicaid scenarios, only adults with children younger than 18 years and a family income below 36% of the federal poverty level are eligible for Medicaid. Under Medicaid expansion, all adults with incomes below 138% of the federal poverty level are eligible for Medicaid. Details on the need projection model are provided in Appendix B, available as a supplement to the online version of this article at http://www.ajph.org.

Optimization Model

We applied the optimization model of Nobles et al.²⁷ to match need and supply under a series of access constraints such that total travel distance across all census tracts was optimized. Because the Medicaid-eligible and privately insured populations face different barriers to health care, the model considered the 2 groups separately in matching them to providers and accounting for their competing access.

Constraints included in the model reflected patient and provider trade-offs. From the patient’s perspective, the constraints ensured that obstacles faced in choosing a provider (e.g., distance, Medicaid acceptance) were taken into account in assigning patients to primary care providers. From the provider’s perspective, constraints were maximum caseload capacity for the study population, maximum Medicaid caseload, and the fact that different providers had different caseload capacities and different Medicaid acceptance levels.

The output of the model consisted of the optimal assignment of needed visits to providers in each census tract; needed visits within a census tract might be assigned to different providers, or a proportion of overall need might not be met. Hence, the model provided estimates of met need (also referred to as “served visits”). Unmet need was calculated as the difference between need and met need. A detailed description of the model is provided in Appendix C, available as a supplement to the online version of this article at http://www.ajph.org.

Availability and Accessibility Measures

We used the results of the optimization model to measure yearly spatial access to primary care among adults at the census tract level. We measured accessibility and availability by means of 2 indices varying between zero (worst value) and 1 (best value). Accessibility was calculated as 1 minus the ratio between the average travel distance to the assigned provider and the maximum distance according to our guidelines (25 miles).³² For example, if the distance was 10 miles, the corresponding accessibility index score was 0.6. Availability was calculated as 1 minus the level of congestion (i.e., the ratio of visits assigned to a provider and the provider’s maximum caseload) experienced at the assigned provider. For example, if the level of congestion was 80%, the availability index score was 0.2. We assumed that people who were not assigned to providers had the lowest accessibility and availability scores (i.e., scores of zero).

Employing the method described by Serban,³³ we used statistical inference to identify the specific tracts in which differences in accessibility, availability, or met need between 2 given scenarios were statistically significant. We considered multiple difference levels (ds = 0, 0.05, and 0.1) in an effort to understand how large differences were if they existed. The results are displayed in Tables 1 through 3 at the .05 significance level.

TABLE 1—

Numbers of Census Tracts Where Between-Scenario Differences in Percentages of Visits Are Significantly Positive or Negative, According to 3 Levels of Difference: 2013–2025 Projections for Georgia

	Difference Between Scenarios
Medicaid Expansion Status and Change	0%	5%	10%
Noa
Positive	1954	1472	0
Negative	0	0	0
No change	1	483	1955
Yesb
Positive	1955	0	0
Negative	0	0	0
No change	0	1955	1955

^aDifference between the percentage of served visits under the nonexpansion and no-Affordable Care Act scenarios (Appendix D, Table A, available as a supplement to the online version of this article at http://www.ajph.org).

^bDifference in the percentage of served visits under the expansion and nonexpansion scenarios (Appendix D, Table A).

TABLE 3—

Number of Census Tracts Where Between-Scenario Differences in the Accessibility and Availability Indexes for Publicly and Privately Insured Individuals Are Significantly Positive or Negative, According to 3 Levels of Difference: 2013–2025 Projections for Georgia

	Publicly Insured			Privately Insured
Parity Program Status and Changea	0%	5%	10%	0%	5%	10%
Accessibilityb
No
Positive	88	0	0	85	0	0
Negative	171	0	0	81	0	0
No change	1696	1955	1955	1789	1955	1955
Yes
Positive	82	0	0	79	0	0
Negative	61	0	0	156	0	0
No change	1812	1955	1955	1720	1955	1955
Availabilityb
No
Positive	0	0	0	0	0	0
Negative	1820	0	0	1955	0	0
No change	135	1955	1955	0	1955	1955
Yes
Positive	0	0	0	0	0	0
Negative	1955	1015	0	1955	0	0
No change	0	940	1955	0	1955	1955

^aDifference in the index under the expansion and nonexpansion scenarios (Appendix D, Table A, available as a supplement to the online version of this article at http://www.ajph.org).

^bSee the text for definitions of the accessibility and availability indexes.

RESULTS

We obtained results for 13 different implementation scenarios (Appendix D, Table A, available as a supplement to the online version of this article at http://www.ajph.org) related to Medicaid eligibility expansion (no ACA, nonexpansion, expansion), different supply growth rates (constant, medium, high), and different Medicaid acceptance ratios (without parity, with parity). The baseline scenario was no ACA, constant supply growth, and absence of parity. We assessed the effects of the ACA under nonexpansion on accessibility, availability, and served visits by comparing the nonexpansion and no-ACA scenarios. We measured the impact of Medicaid eligibility expansion by comparing the expansion and nonexpansion scenarios.

Supply Projection Model

Between 2013 and 2025, the number of primary care physicians is projected to increase by 9.2% (615 physicians in total) under the constant growth scenario, 10.1% (677 physicians in total) under the medium growth scenario, and 11.7% (784 physicians in total) under the high growth scenario. The numbers of NPs and PAs are projected to increase by 20% and 47%, respectively.

Need Projection Model

Under Medicaid expansion in 2025, an additional 980 000 people will be eligible for the Medicaid program and more than 700 000 fewer individuals will be uninsured. In 2013, the percentages of Medicaid-eligible adults and uninsured adults were 2% and 26%, respectively; under the expansion scenario, these percentages are projected to be 15% and 13% in 2025. Population aging and growth are projected to produce an increase of 20% in needed visits by 2025.

Optimization Model

Figure 1 shows the numbers of served visits in Georgia for all projected years according to different scenarios. In 2025, the number of served visits under the baseline scenario is projected at 7 044 866; an additional 876 124 visits will be served as a result of the exchange market. A negligible number of additional visits will be served owing to supply growth or implementation of the parity program, and an additional 439 087 visits will be served under Medicaid eligibility expansion.

FIGURE 1—

Total Numbers of Served Visits for All Projected Years and for Different Scenarios: Georgia, 2013–2025

Note. ACA = Patient Protection and Affordable Care Act.

In the 2013 baseline scenario, served visits represented 67% of needed visits. This percentage is projected to remain the same in 2025 both for the baseline scenario and under nonexpansion, and it is projected to increase to 80% under expansion regardless of supply growth (see Appendix D, Tables B–G).

At a difference (d) level of 0.0, almost all of the census tracts in 2025 will have a statistically significant increase in served visits for the entire population as a result of the ACA provisions, regardless of Medicaid expansion (Table 1). Our maps (Appendix D, Figures A and B) show that overall effects across the state are uniform. At d levels of 0.1 or above, differences are not statistically significant.

Availability and Accessibility Measures

Accessibility.

In the 2013 baseline scenario, median accessibility values (on a 0–1 scale, from worst to best) were 0.674 for the overall population, 0.919 for the publicly insured population, and 0.925 for the privately insured population. Appendix D (Figure C) shows median accessibility values at the state level for all projected years under different scenarios.

In 2025, the impact of the ACA under nonexpansion for the entire population ranges between −0.2 and 0.5 regardless of the implementation of the parity program, with positive values corresponding to improvements in accessibility under nonexpansion. At a difference (d) level of 0.0 (Table 2), assuming medium supply growth, 98% of the census tracts exhibit significantly positive differences regardless of parity program implementation. Our maps (Appendix D, Figure D1) show that the overall impact across the state is uniform. At d values of 0.1 or above, differences are not statistically significant.

TABLE 2—

Numbers of Census Tracts Where Between-Scenario Differences in the Accessibility and Availability Indexes Are Significantly Positive or Negative, According to 3 Levels of Difference: 2013–2025 Projections for Georgia

	Difference Between Scenarios
Parity Program Changea	0%	5%	10%
Accessibilityb
No
Positive	1934	1209	0
Negative	0	0	0
No change	21	746	1955
Yes
Positive	1935	1212	0
Negative	0	0	0
No change	20	743	1955
Availabilityb
No
Positive	579	0	0
Negative	109	0	0
No change	1267	1955	1955
Yes
Positive	538	0	0
Negative	104	0	0
No change	1313	1955	1955

^aDifference in the index under the nonexpansion and no-Affordable Care Act scenarios (Appendix D, Table A, available as a supplement to the online version of this article at http://www.ajph.org).

^bSee the text for definitions of the accessibility and availability indexes.

In 95% of the census tracts, assuming medium supply growth with parity, the impact of Medicaid expansion ranges between −0.095 and 0.095 in 2025 for publicly insured individuals; the impact ranges from −0.06 to 0.06 among privately insured individuals. At a d level of 0.0 (Table 3), 82 census tracts exhibit a significantly positive difference with respect to health care accessibility for the publicly insured population, and 61 exhibit a significantly negative difference; the corresponding totals for the privately insured population are 79 and 156. Our maps (Appendix D, Figures E1 and F1) show that both positive and negative effects are concentrated in urban areas (i.e., Atlanta and Columbus). At d levels of 0.05 or above, differences are not statistically significant.

Availability.

In the 2013 baseline scenario, median availability values (again on a 0–1 scale, from worst to best) were 0.202 for the overall population, 0.418 for the publicly insured population, and 0.271 for the privately insured population. Appendix D (Figure C) shows median availability values at the state level for all projected years under different scenarios.

In 2025, assuming medium supply growth, the impact of the ACA on availability ranges from −0.4 to 0.5 (with positive values corresponding to improvements in availability) under nonexpansion regardless of parity program implementation. At a d level of 0.0 (Table 2), 579 census tracts show a significantly positive difference in terms of health care availability without parity, and 109 show a significantly negative difference; the corresponding totals with parity are 538 and 104. At d levels of 0.05 or above, differences are not statistically significant.

In 95% of the census tracts, assuming medium supply growth with parity, the impact on availability of Medicaid eligibility expansion ranges from −0.48 to 0.48 for publicly insured individuals and from −0.28 to 0.28 for privately insured individuals. At a d level of 0.0 (Table 3), each census tract shows a significantly negative difference for both population groups regardless of parity program implementation. The corresponding maps are presented in Appendix D (Figures E2 and F2). At d levels of 0.1 or above, differences are not statistically significant.

DISCUSSION

In our study, we focused on the projected impact of implementation of the ACA on spatial access to primary care among nonelderly individuals in Georgia. In contrast to existing research,^6,7,34 we used met need, estimated via an optimization model, to evaluate the impact of the ACA (with or without Medicaid expansion) on health care access. This is an important contribution given that not all supply is available to patients in need of care owing to access barriers and system constraints.

Our results showed that the percentage of unmet needed visits decreases to 20% under expansion. This result is not consistent with those of existing studies, which projected that the supply shortage will worsen with the implementation of the ACA.^6,7,34 This discrepancy is explained by the fact that, unlike commonly used measures,^5–7 our shortage measure accounted for system interactions between supply and need. In addition, our findings reveal the importance of interventions designed to increase supply in targeted areas so that accessibility and availability barriers can be overcome when needed.

Overall, we project that the implementation of the ACA will have a positive impact on accessibility, which will improve by 20% under expansion, and a negative impact on availability, which will decrease by 13% to 19% under expansion. These findings reveal the importance of implementing interventions that simultaneously account for multiple dimensions of access.

However, the overall impact on spatial access will minimally affect geographic disparities. Few communities, particularly those whose populations are predominantly uninsured, will experience improved accessibility, and few will experience reduced accessibility. We also project that the ACA provisions excluding Medicaid expansion will affect availability, with the number of communities with projected lower availability larger than the number with projected lower accessibility.

Under Medicaid expansion, the burden on privately insured individuals in terms of primary care accessibility is not substantial; in only a few communities will those with private insurance experience lower accessibility. The availability of primary care providers for both the Medicaid-eligible and privately insured populations will decrease in most communities, although these decreases will not exceed 5%.

Limitations

A primary challenge in this study was limited data availability. However, we considered multiple data sources for different years in estimating our model parameters.

A second limitation was our reliance on several different model assumptions. The supply model assumed that provider productivity and geographic distribution will remain the same as in the baseline year. The supply of midlevel providers was underestimated because such providers are not captured in the National Plan & Provider Enumeration System if they do not directly bill for services. We may also have underestimated the projection of midlevel providers in that the supervision requirements in Georgia may be removed. The need projection model assumed that current patterns of health care use will remain the same as in the baseline year. Although our need estimates were based on utilization ratios, which may underestimate or overestimate need for some subpopulations, other approaches (e.g., those using care guidelines) might consistently underestimate need as well, particularly for populations with chronic conditions. In addition, in our optimization model, willingness to travel was assumed to be the same in rural and urban areas.

A third limitation is that our models accounted for changes in supply and need attributable to certain ACA provisions (i.e., Medicaid eligibility expansion, creation of the exchange insurance market, supply growth) but did not account for other major policy changes such as new payment models and technology-oriented health care provisions, the reason being that the impact of such changes on supply is still unclear. These issues can be addressed with simulations, which can generate behavioral responses to such changes.

Conclusions

Notwithstanding our study’s limitations, it has some important implications for health care providers and policymakers. Although several national-level studies highlight the potential lack of primary care providers to supply the increased need created by the ACA’s provisions,^4–7 we found that such provisions (i.e., Medicaid eligibility expansion and health exchange markets) will reduce total unmet need and have a positive impact on accessibility but a negative impact on availability. The increased burden of need associated with a larger Medicaid population is not significant in terms of accessibility among those who are privately insured; however, it may reduce the availability of primary care providers for both publicly insured and privately insured individuals.

If Georgia opts for Medicaid eligibility expansion, the level of met need across the state will increase substantially (from 67% to 80%) by 2025. Such a policy will have a positive impact on accessibility in some communities while reducing availability. Increases in supply will also positively affect met need and accessibility but will not uniformly overcome spatial access barriers. To be effective, interventions need to be targeted locally and need to simultaneously account for multiple dimensions of access.