Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 May 1.
Published in final edited form as: Med Care Res Rev. 2024 Nov 19;82(1):58–67. doi: 10.1177/10775587241296194

How Specialized are Special Needs Plans? Evidence from Provider Networks

Grace McCormack 1, Rachel Wu 2, Mark Meiselbach 3
PMCID: PMC12043026  NIHMSID: NIHMS2067740  PMID: 39560115

Abstract

Enrollment in Medicare Advantage (MA) Dual-Eligible Special Needs Plans (D-SNPs) among individuals dually eligible for Medicare and Medicaid has more than tripled over the past decade. Little is known about whether D-SNP plan design differs from standard MA plan design nor whether this design reflects the needs of dual-eligible enrollees. We characterize the degree to which D-SNPs specialize an important plan design dimension—provider networks. We find that in 2022, 46 percent of D-SNPs offer networks that are distinct from the insurer’s standard MA plan networks. Compared to D-SNP networks that are shared with standard MA plans, specialized D-SNP networks include more psychiatrists, Ob/Gyn’s, and neurologists, providers that specialize in treating conditions more common among dually eligible enrollees. Network specialization is more common among insurers participating in the local Medicaid market and less common in provider shortage areas, suggesting investment in Medicaid and reduced provider negotiation costs may facilitate specialization.

Keywords: Medicare Advantage, D-SNPs, provider networks, Medicaid

INTRODUCTION

Compared to beneficiaries only eligible for Medicare, beneficiaries dually eligible for both Medicare and Medicaid are over three times as likely to report being in poor physical health and 95 percent more likely to report having a mental health condition (Peña et al., 2023). As a means of addressing these complex needs, Dual-Eligible Special Needs Plans (D-SNPs), a special type of Medicare Advantage (MA) plan, were introduced to exclusively serve dually eligible beneficiaries. D-SNPs differ from standard MA plans in two important ways: first, D-SNPs are required to engage in some level of coordination with Medicaid and second, D-SNPs are required to design and implement care management models to specifically serve the complex needs of the dually eligible population (Medicare Payment Advisory Commission, 2019; Quality improvement program, 2018; Requirements relating to basic benefits, 2018; Scoring Guidelines, 2024).

Though enrollment in D-SNPs has more than tripled over the past decade (Ochieng, Biniek, Cubanski, & Neuman, 2023), and on average D-SNPs yield higher profit margins than standard MA plans (Centers for Medicare & Medicaid Services, 2016; Xu et al., 2023), little is known about whether and to what extent D-SNPs are designed to serve the specific, complex needs of the dually eligible population.

D-SNP enrollees may particularly benefit from specialized provider networks that are tailored to the specific needs of the dually eligible population. Past work has found that provider networks can have a profound impact on enrollee outcomes and satisfaction (Atwood & Lo Sasso 2016; Fortney et al. 2001; Gruber & McKnight, 2016; Haeder et al. 2016; Mazurenko, Taylor & Menachemi, 2022; Rose et al. 2021; Wallace, 2023). Given the higher incidence of mental and physical chronic conditions and the distinct socioeconomic composition of this population (Peña et al., 2023), dually eligible beneficiaries likely have a greater need for certain specialists or physicians with expertise serving communities with greater social need. However, while D-SNP insurers must justify how their provider network meets the specific needs of dually eligible beneficiaries to participate in the market (Medicare Payment Advisory Commission, 2019; Quality improvement program, 2018; Requirements relating to basic benefits, 2018; Scoring Guidelines, 2024), the Centers for Medicare and Medicaid Services (CMS) currently does not have different network adequacy requirements for D-SNPs compared to standard MA plans (US Government Accountability Office, 2015). Thus, insurers offering both a D-SNP and a standard MA plan in a given area can choose whether to use the same network for both types of plans or specialize their D-SNP network.

As enrollment in D-SNP MA plans continues to grow (Ochieng et al., 2023), the extent to which insurers D-SNP plans and D-SNP networks, specifically, have been designed to serve the needs of dually eligible beneficiaries remains an open question.

Conceptual Framework

There are potential costs and benefits to offering a specialized D-SNP network. Insurers may be able to attract more enrollees, garner higher quality metrics, or better contain costs if their providers are particularly desirable and skilled in serving the dually eligible population. However, the formation of specialized D-SNP networks also likely requires insurers to invest additional resources to identify which local providers would be valuable for the dually eligible population. Further, negotiating networks separately for dually eligible and non-dually eligible enrollees may also reduce the insurer’s bargaining power or, if there are few local providers, may make it difficult to form sufficient networks. These costs suggest that specialization may be more common in areas where the provider market is sufficiently large and unconcentrated and where insurers have already invested resources in identifying the needs of the local Medicaid population and forming networks.

Network specialization may also be affected by the degree to which a D-SNP is financially or operationally integrated with Medicaid. D-SNPs that are fully or highly integrated with Medicaid managed care plans (“FIDEs” and “HIDEs”, respectively) have a capitated Medicaid contract to provide a range of services, including long term care and/or behavioral health, and provide both Medicare and Medicaid benefits through a single managed care plan. By contrast, “coordination-only” D-SNPs bear limited financial risk for Medicaid-covered services, including behavioral health, and in most states are not required to engage in any coordination beyond notifying Medicaid of enrollee hospitalizations or SNF admissions (Medicare Payment Advisory Commission, 2019; Meyers, et al. 2023; Peña, Mohamed, & Burns, 2023; Requirements for dual eligible special needs plans, 2019). In comparison to coordination-only D-SNPs, FIDEs and HIDEs may be more likely to have specialized networks since they have already invested substantial resources into addressing the needs of the dually eligible population and additionally have a greater incentive to tailor networks to address behavioral healthcare and long term care needs that would otherwise not be covered under a coordination-only D-SNP contract.

New Contribution

Our work contributes significantly to the literature in two ways. First, we provide novel evidence on the extent to which plans are tailored to the needs of the dually eligible population. While recent work has called into question the quality of care provided under D-SNPs relative to traditional Medicare (TM) and non-D-SNP MA plans (Haviland et al., 2021; Johnston, Wen, & Pollack, 2022; Meyers, Offiaeli, Trivedi, & Roberts, 2023; Roberts & Mellor, 2022), very little is known about how D-SNP plans are designed nor the extent to which their design differs from standard MA plans. We address this gap in the literature by providing novel evidence on the design of D-SNP networks. Using national MA network data, we document the share of insurers who create distinct provider networks for their D-SNP plans (i.e., “specialized networks”) versus how often they use the same networks for both D-SNP and standard MA plans offered (i.e., “shared networks”). To better understand how well D-SNP networks are designed to meet the needs of the dually eligible population, we then document the average breadth and provider specialty composition of both specialized and shared D-SNP networks in comparison to each other and standard MA plans.

Second, we additionally contribute to the literature on network design by providing novel evidence on the determinants of network specialization. A robust literature has documented how various factors such as insurer size, provider market concentration, and enrollee population health status can impact the breadth and composition of provider networks in different insurance markets (Ho & Lee, 2017; Shepard, 2016). In this paper, we extend this literature by providing suggestive evidence on which factors impact an insurer’s decision to offer additional, distinct provider networks for different enrollee populations, an important question given the frequency with which insurers participate in multiple market segments within the same geography (Marr, Polsky, & Meiselbach, 2024).

METHODS

Data:

The primary dataset used in our analysis is 2022 Ideon provider networks data (formerly Vericred), which lists the set of providers in network for each MA plan offered in the calendar year. Ideon collects data through scraping insurers’ online provider directories. Ideon imposes internal quality assurance processes to ensure that accurate information is provided to its commercial clients, working directly with insurance carriers to limit potential inaccuracies in the network directories. These data have been used in prior work assessing provider networks in managed care (Graves et al., 2020; Meyers, Rahman, & Trivedi, 2021; Sen, Meiselbach, Anderson, Miller, & Polsky, 2021; Zhu, Zhang, & Polsky, 2017).

We matched each plan in the Ideon data to public county-level plan data from CMS to identify counties where plans were offered as well as the plan’s enrollment, affiliated insurer, and other characteristics. We used USDA data to characterize rural and urban status of counties (USDA ERS, 2023).

Data on the specialty, credentials, and location of practice of providers was provided by the 2021 OneKey data, a proprietary database of over 10 million healthcare providers. As in prior work, we limited to physicians (MD/DOs) who were active in the National Plan and Provider Enumeration System, born in 1940 or later, accepted Medicare payments, and practiced at an office, hospital, or residential facility (Zhu, Meiselbach, Drake, & Polsky, 2023). OneKey data is linked to MA networks in Ideon by National Provider Identifiers.

We used public 2021 enrollment data to characterize the overall size of the Medicare market in different counties as well as the 2021 DRG Interstudy data to characterize insurer participation in Medicaid Manage Care (MMC) in a county (Clarivate, 2022). These data are collected through a national proprietary census and are commonly used to study health insurance markets (Graves et al., 2020; Meiselbach et al., 2022).

Finally, we used data compiled by the Health Resources & Services Administration to identify primary care or mental health professional shortage areas (HPSA), regions determined as having insufficient providers to meet population needs.

Measures:

We classified all D-SNP networks as either specialized or shared. We considered a D-SNP network specialized if it was unique compared to the other (non-D-SNP) MA networks offered by that same insurer in the same state. Two networks were considered the same or “shared” if they either shared a network-identification number in the Ideon data or if both networks had at least 95 percent of the same providers found in the other network. Importantly, for a given D-SNP network and each comparator standard MA network, we only considered the networks shared if at least 95 percent of the D-SNP network’s providers are in the standard MA plan’s network and at least 95 percent of the standard MA plan’s network providers are in the D-SNP plan’s network. Our definition thus avoided classifying two networks as shared if, for example, the D-SNP network was merely a smaller subset of the standard MA plan network. All networks that were not classified as shared by this definition were classified as specialized. In supplementary exercises, we tested how robust our results were by using an 80 percent, 90 percent, and 99.5 percent threshold. For details of our definition of specialization, see online appendix section I.

We use the term “specialized” throughout to refer to this binary distinction of networks. However, this measure alone does not necessarily imply that the networks are tailored to the needs of D-SNP beneficiaries, only that they are distinct from standard MA plans offered by the same carrier.

To better understand whether specialized networks are designed to serve the needs of the dually eligible population, we analyze the breadth and provider specialty composition of specialized D-SNP networks in comparison to shared D-SNP networks and standard MA plans.

For each network, we characterize network breadth as the share of physicians operating in a county that are included in the network, where the county denominator was based on the number of eligible physicians in the OneKey database. In addition to overall breadth, we also considered network composition by investigating specialty-specific breadth, defined as the share of county physicians in each provider specialty category who participate in each network. Specialty categories included primary care, psychiatry, surgical specialties, dermatology, hospital-based specialties, internal medicine subspecialties, neurology, Ob/Gyn, and other specialists. For details, see online appendix section I.

Study variables:

Insurer characteristics included each insurer’s respective D-SNP enrollment in a county across all plans, their participation in the county’s MMC market, and for participating insurers, their county market-share among MMC plans. Plan characteristics included plan type (i.e., HMO or PPO), MA contract star rating (1 to 5), and Medicaid integration status. Provider market characteristics included physician density defined as the number of providers per 10,000 Medicare enrollees in a county (based on OneKey data), hospital market concentration, and indicators for whether a county was a primary care or mental health HPSA. Hospital market concentration was measured as a Herfindahl-Hirschman Index (HHI) of hospital shares of inpatient days in the CBSA based on the 2021 AHA annual survey. Finally, county level characteristics included metropolitan and rural status and total number of dually eligible beneficiaries per county.

Statistical analysis:

Our study included three analyses. First, we tabulated the proportion of D-SNP plans that offer specialized versus shared networks. We characterize the prevalence of specialized networks, both overall among all D-SNP plans as well as within the eight largest MA insurers (Anthem, BCBS, CVS, Centene, Cigna, Humana, Kaiser, United Healthcare), presenting both the share of plans with specialized networks and share of MA enrollees in such plans.

Our second analysis characterizes the network breadth and composition of D-SNP plans with specialized and shared provider networks. In appendix exercises, we present average enrollee-weighted total physician breadth and provider specialty-specific breadth, for standard MA plan networks, specialized D-SNP networks, and shared D-SNP networks. Our measures of physician breadth were county-specific, characterizing the share of physicians operating in the county that are included in a given network (i.e., overall network breadth) and the share of physicians with a given specialty operating in the county that are included in a given network (i.e., our measure of dermatology-specific network breadth characterizes the share of dermatologists operating in a given county that are included in the network). We then used regression analysis to evaluate if there were significant differences in overall and specialty-specific breadth between specialized and shared networks.

Restricting to our sample of D-SNP networks, we ran a separate regression where the primary outcome is a separate measure of network breadth (i.e., overall network breadth or one of the various specialty-specific measures of network breadth). The primary explanatory variable of interest was an indicator for whether a D-SNP network is specialized; regressions additionally included separate insurer and county fixed effects so that estimates characterize variation in breadth within a given insurer and a given county. Because some plans may have separate psychiatry networks not included in our dataset, we followed past work by only comparing psychiatry network breadth among plans that do not have unreasonably narrow psychiatry networks (Meiselbach et al., 2023; Zhu, Meiselbach, Drake, Polsky, 2023; Slade, Wu, Meiselbach, & Polsky, 2023). Kaiser plans were excluded from this analysis. See the online appendix for more details.

For our third analysis, we characterized factors associated with network specialization. We first compared the average characteristics of plans with specialized and shared networks and tested for differences using two-sided t-tests. We then used regression analysis to test how various factors related to network specialization. The main outcome of these regressions was an indicator for whether a plan had a specialized network. For all regressions, we included insurer indicators, so all coefficients characterize the marginal effects of each variable on specialization within a given insurer. We included several explanatory variables that capture theorized costs and benefits of specialization. We hypothesized that network specialization entails high fixed costs, suggesting that network specialization will be more prevalent in areas with a larger potential market for D-SNP plans, captured as the logged total number of dual eligible beneficiaries in the county. We also hypothesized that because specialization may reduce insurer negotiating power, specialization may be more common in areas with lower provider concentration. We present separate regression results using three different measures: hospital HHI, physician density, and county status as a primary care or mental health HPSA. Finally, we hypothesized that network specialization is more common among insurers with greater investment in the local Medicaid market, since such investment may correspond with greater knowledge of the needs of local Medicaid-eligible enrollees or existing networks for MMC enrollees. We included indicators for D-SNP being offered by an insurer that also offers an MMC plan in the same county and indicators for the D-SNP being fully or highly integrated with Medicaid. We additionally controlled for the state’s Medicaid expansion status. Kaiser plans were excluded from this analysis.

RESULTS

In 2022, under half (46 percent) of D-SNP plans offered specialized networks, comprising 55 percent of D-SNP enrollment (exhibit 1). There was considerable variation between insurers in the proportion of plans and enrollees with specialized networks. While 82 percent of United D-SNP enrollees were in a plan with a specialized network, access to specialized networks was uncommon among D-SNP enrollees in plans offered by Cigna (7 percent of enrollees), Humana (11 percent), and Kaiser (0 percent). There was considerable variation in the share of enrollees in specialized network plans offered by Anthem (39 percent of enrollees), BCBS (65 percent), CVS (25 percent), and Centene (41 percent).

Exhibit 1: Proportion of D-SNP Plans, Enrollees with Specialized Networks.

Exhibit 1:

Open in a new tab

Notes: This figure shows the share of 2022 D-SNP plans that offer specialized networks that are distinct from all non-D-SNP networks offered by the same insurer in the state. The dark bars show unweighted shares, corresponding to the share of D-SNP plans that offer specialized networks, and the light bars show enrollee-weighted shares, corresponding to the share of D-SNP enrollees in plans that offer specialized networks. The first two bars show overall proportion of overall D-SNP plans and enrollees, respectively, that offer specialized networks. The remaining bars restrict to plans offered by the eight largest insurers.

These patterns across insurers are robust to alternative thresholds of network specialization. While most shared D-SNP networks are identified through network-IDs, we also identify shared networks using measures of network overlap (appendix table 1). Among D-SNP networks offered in states where their carrier also offers a standard MA network, network overlap with the most similar MA plan network is high, with much of the mass is concentrated above 90 percent overlap (appendix figure 1). Reducing the threshold of overlap from 95 percent to 90 or 80 percent to characterize a network as shared reduces the proportion of plans identified as offering a specialized network from 46 percent to 36 and 28 percent, respectively (appendix table 2).

Ninety percent of D-SNP plans are offered in counties where the insurer also offers a standard MA plan, and the proportion of plans and enrollees with specialized networks among different insurers also do not change dramatically when we restrict to plans offered in geographies where insurers offer a standard MA plan or when we modify specialization measures to only take into account all physicians operating in the same counties, instead of state (appendix table 3).

On average, standard MA networks and D-SNP networks had similar overall breadth, covering an average 41 and 42 percent of county physicians, respectively (appendix table 4). However, among D-SNP networks, specialized networks tended to be broader than shared networks, both overall and in terms of provider specialty-specific breadth.

Adjusting for county and insurer fixed effects, specialized D-SNP networks were broader than shared D-SNP networks in terms of dermatologist, neurologist, and OBGYN access at the five percent level and psychiatrist access at the 10 percent level (exhibit 2). In unadjusted terms, specialized networks included 46 percent of the total county physicians in their networks compared to 38 percent among non-specialized networks, though this difference in overall breadth was no longer statistically significant after controlling for insurer and county fixed effects (appendix table 4). However, even after adjusting for insurer and county fixed effects, specialized networks had broader dermatology (3.5 percentage points, 8.5 percent broader compared to average breadth among shared D-SNP networks), neurology (3.7 percentage points, 8.6 percent), Ob/Gyn (2.2 percentage points, 4.9 percent broader), and psychiatry (2.2 percentage points, 8.8 percent broader) networks, though the difference in psychiatry networks was only significant at the ten percent level.

Exhibit 2: Breadth and Composition of Specialized D-SNP Networks Compared to Shared D-SNP Networks.

Exhibit 2:

Open in a new tab

Notes: This figure shows regression coefficients from separate enrollee-weighted regressions run on the set of all D-SNP network-county observations. The primary outcome of each regression is a measure of network breadth. We create a broad measure of network breadth defined as the share of all county physicians that are covered in-network as well as specialty-specific measures of network breadth (e.g., the share of county dermatologists who are covered in the network). Each regression regresses a measure of breadth on an indicator for whether the D-SNP network is a specialized network (as opposed to being shared with a non-D-SNP plan) as well as county and insurance indicators.

Each row shows the specialized network indicator coefficient from a separate regression with a different measure of network breadth as the outcome. The set of physicians from which the measure of network breadth is derived is indicated by the row label; the total number of network-county observations is listed underneath each row label. The first row shows the regression coefficient when the outcome is the network breadth including all provider specialties (including psychiatry). Because psychiatry networks are missing for 6% of D-SNP networks, psychiatry network breadth regressions are run on the 94% sub-sample for which we have reliable psychiatry. The second row shows the regression coefficient when the outcome is the network breadth for all providers excluding psychiatry; hence it is run on the full sample. Subsequent rows show regression coefficients for regressions where the primary outcomes restrict to specific specialties listed on the left (e.g., row 3’s regression restricts to dermatology networks).

Dots represent point estimates of specialized D-SNP network breadth relative to shared D-SNP networks and horizontal lines present 95% confidence intervals.

Compared to plans with shared networks, plans that offer specialized networks are more likely to be HMO plans, have greater integration with Medicaid, and are more likely to be offered by insurers that also offer an MMC plan in the same area (exhibit 3). 93 percent of plans that offer specialized networks are HMO’s, compared to 79 percent of shared networks. Compared to D-SNPs with shared networks, specialized D-SNP plans are over five times more likely to be fully integrated with Medicaid and 50 percent more likely to be offered by insurers that also offered an MMC plan in the county.

Exhibit 3:

Characteristics of Plans offering Specialized vs. Shared Networks

D-SNP Plans with Specialized Networks D-SNP Plans with Shared Networks Difference P-value
Plan- & Network- Level Characteristics
HMO (%) 0.933 0.792 0.141*** 0.000
Star Rating 4.011 4.075 −0.064 0.259
Total Network Enrollment 10704.273 7798.341 2905.932 0.092
Medicaid, D-SNP Involvement
Fully Integrated Dual Eligible Plan (FIDE) 0.171 0.031 0.140*** 0.000
Highly Integrated Dual Eligible Plan (HIDE) 0.268 0.198 0.070 0.115
Coordination Only Plan 0.561 0.760 −0.200*** 0.000
Insurer Offers Medicaid Managed Care Plan in Same County 0.561 0.374 0.186*** 0.000
Average MMC Market Share of D-SNP Insurer in Same County 0.132 0.106 0.026 0.183
County Level Characteristics
Metropolitan 0.715 0.709 0.006 0.830
Rural 0.042 0.039 0.003 0.702
Total Dual Eligibles Per County 15310.608 15112.888 197.720 0.957
Hospital HHI 0.261 0.289 −0.028 0.078
Physician Density (Number of MD/DO’s Per 10k Medicare Beneficiaries) 91.467 86.915 4.552 0.269
Primary Care HPSA 0.863 0.890 −0.027 0.074
Mental Health HPSA 0.902 0.920 −0.018 0.170
Open in a new tab

Notes: This first two columns of this table show the average characteristics of 2022 D-SNP plans that offer specialized and shared networks, respectively. The third column shows the coefficient from a two-sided t-test comparing the characteristics of D-SNP plans with specialized and shared networks.

Significance in differences is indicated with stars at the 95 percent (*), 99 percent (**), and 99.9 percent (***) level. The fourth column depicts the p-value of this difference.

While the proportion of plans and enrollees with specialized networks varies regionally (appendix figure 2), cross-sectionally, plans that offer specialized and shared networks tend to be offered in similar areas in terms of metropolitan or rural status, number of dual-eligibles, and measures of provider concentration or capacity (exhibit 3).

In adjusted regression analysis, network specialization is highest among plans that are offered in primary care and mental health HPSAs, are integrated with a MMC plan, and are offered by insurers also offering local MMC plans (exhibit 4). We find some evidence consistent with the hypothesis that provider negotiating power impacts specialization: while the likelihood of specialization is not related to hospital HHI nor physician density, the likelihood of offering a specialized network is 3.93 percentage points (9 percent) lower in mental health HPSAs.

Exhibit 4:

Determinants of Network Specialization

(1) (2) (3)
Network Specialization Network Specialization Network Specialization
Hospital HHI 0.0101
(0.66)
MD/DOs Per 10k Medicare Benes −0.0000733
(−1.38)
Primary Care HPSA −0.0137
(−1.15)
Mental Health HPSA −0.0393*
(−2.42)
Has MMC in County 0.203*** 0.215*** 0.215***
(12.95) (14.76) (14.73)
Fully Integrated Plan (FIDE) 0.0897*** 0.0741*** 0.0714***
(4.62) (4.34) (4.18)
Highly Integrated Plan (HIDE) 0.146*** 0.166*** 0.166***
(8.84) (10.89) (10.94)
Medicaid Expansion State 0.0999*** 0.0992*** 0.0968***
(7.99) (8.69) (8.49)
Ln(Total Dual Eligibles Per County) 0.00189 0.00839 0.00677
(0.36) (1.60) (1.40)
Observations 8309 9828 9828
Open in a new tab

Notes: This table shows the regression coefficients where the outcome variable is an indicator for a plan having a specialized network. Independent variables include multiple measures of provider market concentration (local hospital HHI, the number of MDs/DOs per 10,000 Medicare beneficiaries in the county, and indicators for the county being a primary care or mental health HPSA), indicators for whether the plan was offered by an insurer that also offered a MMC plan in the same county or whether the D-SNP was fully or highly integrated with Medicaid, logged total county dual eligibles, and an indicator for whether or not the county was in a Medicaid expansion state. All regressions included insurer fixed effects. All regressions exclude Kaiser plans. Below each regression coefficient is the t-statistic. Standard errors are clustered by county. Statistical significance is indicated by the 0.1, 1, and 5% level with three, two, or one star, respectively.

Further, consistent with the importance of investment in the local Medicaid market, across all specifications, we find that the proportion of plans offering specialized networks is roughly 21 percentage points (46 percent) higher among plans that are offered by insurers that also offer an MMC in the same county. Correspondingly, the three insurers with the lowest overall proportion of specialized networks (Cigna, Humana, and Kaiser) also have the lowest average MMC market share (appendix table 5). The likelihood of offering a specialized network is 7.1 to 16.6 percentage points (15 to 36 percent) higher among fully and highly integrated plans, consistent with these plans having greater established investment in the Medicaid market and bearing greater responsibility for certain types of care, including long term services and behavioral health. Finally, contrary to our theory of fixed costs, conditional on insurer, in all specifications, network specialization is unrelated to the size of the dually eligible population in the county. Though magnitudes vary, these patterns are robust to alternative control strategies (appendix tables 6 and 7) and a logistic, rather than OLS specification (appendix table 8).

DISCUSSION

We examined the extent to which insurers offer specialized D-SNP networks for dually eligible beneficiaries. Despite requirements to justify how their provider network has specialized expertise in the dually eligible population (Medicare Payment Advisory Commission, 2019; Quality improvement program, 2018; Requirements relating to basic benefits, 2018; Scoring Guidelines, 2024), only 46 percent of D-SNP plans feature distinct provider networks from standard MA plans offered by the same insurer and, on average, D-SNP networks and standard MA plan networks have similar overall breadth.

Much of the variation in network specialization is driven by differences across insurers, with some insurers such as Cigna, Humana, and Kaiser, rarely specializing their network, and others, like United, specializing their network for most plans.

While our measure of network specialization—being distinct from networks offered by a standard MA plan—alone is insufficient to conclude that specialized networks are tailored to meet the needs of the dually eligible population, comparisons of provider composition suggest that specialized networks do tend to include more providers that specialize in treating conditions more common among dual-eligible enrollees. Even after accounting for differences across counties and insurers, specialized networks tend to have broader inclusion of psychiatrists, neurologists, and Ob/Gyn’s. Broader psychiatric networks may be particularly valuable for D-SNP enrollees given that dual eligible beneficiaries are 95 percent more likely to report having a mental health condition, compared to Medicare beneficiaries that are not dually eligible (Peña et al., 2023). Though neurologists treat a number of conditions, access to neurologists to treat Alzheimer’s disease and related dementias may be particularly relevant for dual-eligible enrollees, given the relatively high incidence of these conditions in this population (Meyers, Rahman, Rivera-Hernandez, Trivedi, & Mor, 2021; Peña et al., 2023). Similarly, while there exists no work that documents the relative utilization rates of Ob-Gyn’s among dual eligible enrollees, relative to non-dually eligible Medicare beneficiaries, existing evidence suggest that nearly 8 in 10 women of reproductive age on Medicare are dually eligible for Medicaid and dual-eligible women enrollees are significantly more likely to use various types of contraceptives, compared to non-dually eligible Medicare enrollees (Freed et al., 2024; Ellison et al., 2024).

While specialized networks tended to include providers that may treat conditions more common among dually eligible enrollees, we find evidence of potential barriers to network specialization. First, network specialization was less common in HPSAs. Multiple mechanisms could drive this. Intuitively, in areas where there are fewer providers, there may be less opportunity to customize networks for dually eligible beneficiaries, specifically. Alternatively, in areas where there are more providers and any individual provider’s negotiating power is lower, insurers may be better able to customize their networks without losing significant negotiating power over providers from splitting their enrollee pool.

Second, we found evidence that network specialization was higher among plans and insurers that were integrated with Medicaid plans, and therefore had more extensive investment in the Medicaid market and were required to manage Medicaid spending and services, such as long term services and behavioral health services. Similarly, we also found network specialization was highest among plans that offered MMC plans in the same area. These insurers likely have greater knowledge of the specific needs of the Medicaid-eligible population in the area and could furthermore use some or all of their existing MMC network for their D-SNP plan.

Though D-SNPs were originally envisioned as a means of better coordinating care and tailoring plan benefits for dually eligible enrollees, prior studies have not found that enrollees in D-SNPs, especially non-integrated D-SNPs, have on average better healthcare experiences relative to enrollees in standard MA plans or TM (Haviland et al., 2021; Johnston et al., 2022; Meyers et al., 2023; Roberts & Mellor, 2022). While past work connects these findings to the lack of meaningful integration and care coordination among D-SNP, our results highlight a potential additional, complementary contributor: a lack of differentiation from standard MA plans in their benefit design. Though provider networks have been found to impact enrollee health outcomes (Atwood & Lo Sasso 2016; Fortney et al. 2001; Gruber & McKnight, 2016; Haeder et al. 2016; Mazurenko, Taylor & Menachemi, 2022; Rose et al. 2021; Wallace, 2023), we find that D-SNP networks are on average similar to standard MA plans. Further, only half of D-SNPs have provider networks that are at all distinct from standard MA plans. These results suggests that, at most, under half of D-SNPs have at all tailored a critical dimension of plan design—the provider network—specifically for the needs of dually eligible enrollees. Though D-SNPs could specially design other plan aspects, our findings, alongside the documented lack of Medicaid integration among most D-SNP plans, provide important context to previous findings that D-SNPs do not consistently perform better than standard MA plans.

As of 2019, while the dually eligible comprised only 17 percent of Medicare’s total population, including both TM and MA, the group accounted for 30 percent of the program’s total spending (Arnold Ventures, 2020). In MA, D-SNP plans yield higher average profit margins than standard MA plans (Xu et al., 2023). Given the significant needs of dually-eligible beneficiaries, the greater payments that are paid to insurers to cover this population, and the sizable share of Medicare expenditures these payments represent, greater attention should be paid to both the design and impact of D-SNP plans on outcomes for dually-eligible beneficiaries.

Limitations

Our analysis is subject to limitations. Our measure of network specialization and breadth are restricted to individual physicians, not facilities such as hospitals, and also excludes other types of providers, such as nurse practitioners and physician assistants, which are inconsistently incorporated into network adequacy requirements and have differing prescribing authority across areas. However, NPs and PAs are important to meeting the needs of D-SNP beneficiaries, especially in rural areas, and should be investigated in future work.

Second, our regressions on the determinants of network specialization are cross-sectional in nature and thus should be considered suggestive but not dispositive evidence of the presence of certain mechanisms.

Further, absent a causal analysis of clinical outcomes and measures of patient preferences, we cannot say for certain whether specialized networks are better in terms of enrollee health or satisfaction. However, we provide suggestive descriptive evidence on the composition of networks, testing whether there is greater coverage of provider specialties that specialize in treating conditions common among the dually eligible.

Finally, because our measures of specialization and breadth are unweighted by the capacity or care volume of different providers, both outcomes may be subject to measurement error. Our main definition of a shared network—having 95 percent provider overlap between two networks—may be overly or underly restrictive, depending on the care volume of these different providers. This concern may be particularly salient given the documented inaccuracies of provider network directories and presence of “ghost networks” (Haeder, Weimer, & Mukamel, 2016; Resneck, Quiggle, Liu, & Brewster, 2014; Zhu, Jane, Charlesworth, Polsky, & McConnell, 2022). While this potential limitation is present in most analyses of networks, we demonstrate that our results are robust to alternative thresholds of specialization in the appendix.

CONCLUSION

Though envisioned as a way to coordinate benefits between Medicare and Medicaid and serve the specific needs of the dually eligible population, only 46 percent of D-SNP plans feature distinct provider networks from standard MA plans offered by the same insurer. Network specialization varies considerably across insurers and is more common in plans offered by insurers with a greater presence in the local Medicaid market. Compared to shared D-SNP networks, specialized D-SNP networks have broader inclusion of specialties that specialize in treating conditions more common among dually eligible enrollees. Further work is needed to understand the nature and impact of D-SNPs plan design.

Contributor Information

Grace McCormack, University of Southern California, 645 W Exposition Blvd, Los Angeles, CA 90089.

Rachel Wu, Johns Hopkins, 615 N Wolfe St, Baltimore, MD 21205.

Mark Meiselbach, Johns Hopkins, 615 N Wolfe St, Baltimore, MD 21205.

REFERENCES

  1. Arnold Ventures. (2020). Who is the dual-eligible population and why is change needed? Retrieved from https://www.arnoldventures.org/stories/dual-eligible-beneficiary-fact-sheet
  2. Atwood A, & Lo Sasso ATL (2016). The effect of narrow provider networks on health care use. Journal of Health Economics, 50, 86–98. [DOI] [PubMed] [Google Scholar]
  3. Centers for Medicare & Medicaid Services. (2016). Announcement of calendar year (CY) 2017 medicare advantage capitation rates and medicare advantage and part D payment policies and final call letter. Retrieved from https://www.cms.gov/Medicare/Health-Plans/MedicareAdvtgSpecRateStats/Downloads/Announcement2017.pdf
  4. Clarivate. (2022). Healthcare research and data - decision resources group. Retrieved from https://decisionresourcesgroup.com/
  5. Ellison J, Pudasainy S, Bellerose M, Quinn D, Borrero S, Olson I, ... & Jarlenski MP (2024). Contraceptive Use Among Traditional Medicare And Medicare Advantage Enrollees: Study examines contraceptive use among Medicare and Medicare advantage enrollees. Health Affairs, 43(1), 98–107. [DOI] [PubMed] [Google Scholar]
  6. Freed M, Cubanski J, Long M, Ochieng N, & Salganicoff A (2024, April 30). Coverage of Sexual and Reproductive Health Services in Medicare [Review of Coverage of Sexual and Reproductive Health Services in Medicare]. KFF; Kaiser Family Foundation. https://www.kff.org/medicare/issue-brief/coverage-of-sexual-and-reproductive-health-services-in-medicare/ [Google Scholar]
  7. Fortney J, Thill JC, Zhang M, Duan N, & Rost K (2001). Provider choice and utility loss due to selective contracting in rural and urban areas. Medical Care Research and Review, 58(1), 60–75. [DOI] [PubMed] [Google Scholar]
  8. Graves JA, Nshuti L, Everson J, Richards M, Buntin M, Nikpay S, . . . Polsky D (2020). Breadth and exclusivity of hospital and physician networks in US insurance markets. JAMA Network Open, 3(12), e2029419. doi: 10.1001/jamanetworkopen.2020.29419 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Gruber J, & McKnight R (2016). Controlling health care costs through limited network insurance plans: Evidence from Massachusetts state employees. American Economic Journal: Economic Policy, 8(2), 219–250 [Google Scholar]
  10. Haeder SF, Weimer DL, & Mukamel DB (2016). Secret shoppers find access to providers and network accuracy lacking for those in marketplace and commercial plans. Health Affairs (Project Hope), 35(7), 1160–1166. doi: 10.1377/hlthaff.2015.1554 [DOI] [PubMed] [Google Scholar]
  11. Haviland AM, Elliott MN, Klein DJ, Orr N, Hambarsoomian K, & Zaslavsky AM (2021). Do dual eligible beneficiaries experience better health care in special needs plans? Health Services Research, 56(3), 517–527. doi: 10.1111/1475-6773.13620 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Ho K, & Lee RS (2017). Insurer competition in health care markets. Econometrica, 85(2), 379–417. doi: 10.3982/ECTA13570 [DOI] [Google Scholar]
  13. Johnston KJ, Wen H, & Pollack HA (2022). Comparison of ambulatory care access and quality for beneficiaries with disabilities covered by medicare advantage vs traditional medicare insurance. JAMA Health Forum, 3(1), e214562. doi: 10.1001/jamahealthforum.2021.4562 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Marr J, Polsky D, & Meiselbach MK (2024). Commercial insurer market power and medicaid managed care networks. Medical Care Research and Review, Retrieved from https://journals.sagepub.com/doi/10.1177/10775587241241975 [DOI] [PubMed] [Google Scholar]
  15. Mazurenko O, Taylor HL, & Menachemi N (2022). The impact of narrow and tiered networks on costs, access, quality, and patient steering: A systematic review. Medical Care Research and Review, 79(5), 607–617. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Medicaid and CHIP Payment and Access Commission. (2020). Integrating care for dually eligible beneficiaries: Policy issues and options. Retrieved from https://www.macpac.gov/publication/chapter-2-integrating-care-for-dually-eligible-beneficiaries-policy-issues-and-options/
  17. Medicare Payment Advisory Commission. (2019). Promoting integration in dual-eligible special needs plans. In report to the congress: Medicare and the health care delivery system. Washington, DC: [Google Scholar]
  18. Meiselbach MK, Drake C, Zhu JM, Manibusan B, Nagy D, Sorbero MJ, . . . Polsky D (2023). State policy and the breadth of buprenorphine-prescriber networks in medicaid managed care. Medical Care Research and Review: MCRR, 80(4), 423–432. doi: 10.1177/10775587231167514 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Meiselbach MK, Drake C, Saloner B, Zhu JM, Stein BD, & Polsky D (2022). Medicaid managed care: Access to primary care providers who prescribe buprenorphine. Health Affairs (Project Hope), 41(6), 901–910. doi: 10.1377/hlthaff.2021.01719 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Meyers DJ, Offiaeli K, Trivedi AN, & Roberts ET (2023). Medicare and medicaid dual-eligible special needs plan enrollment and beneficiary-reported experiences with care. JAMA Health Forum, 4(9), e232957. doi: 10.1001/jamahealthforum.2023.2957 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Meyers DJ, Rahman M, Rivera-Hernandez M, Trivedi AN, & Mor V (2021). Plan switching among medicare advantage beneficiaries with alzheimer’s disease and other dementias. Alzheimer’s & Dementia (New York, N. Y.), 7(1), e12150. doi: 10.1002/trc2.12150 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Meyers DJ, Rahman M, & Trivedi AN (2021). Narrow primary care networks in medicare advantage. Journal of General Internal Medicine, doi: 10.1007/s11606-020-06534-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Ochieng N, Biniek JF, Cubanski J, & Neuman T (2023). Disparities in health measures by race and ethnicity among beneficiaries in medicare advantage: A review of the literature. KFF. Retrieved from https://www.kff.org/medicare/report/disparities-in-health-measures-by-race-and-ethnicity-among-beneficiaries-in-medicare-advantage-a-review-of-the-literature/ [Google Scholar]
  24. Peña M, Mohamed M, & Burns A (2023, April 27). Medicaid Arrangements to Coordinate Medicare and Medicaid for Dual-Eligible Individuals [Review of Medicaid Arrangements to Coordinate Medicare and Medicaid for Dual-Eligible Individuals]. KFF; Kaiser Family Foundation. https://www.kff.org/medicaid/issue-brief/medicaid-arrangements-to-coordinate-medicare-and-medicaid-for-dual-eligible-individuals/ [Google Scholar]
  25. Peña MT, Mohamed M, Burns A, Biniek JF, Ochieng N, & Published PC (2023, Jan 31,). A profile of medicare-medicaid enrollees (dual eligibles). Retrieved from https://www.kff.org/medicare/issue-brief/a-profile-of-medicare-medicaid-enrollees-dual-eligibles/
  26. Quality improvement program, 42 C.F.R. § 422.152 (2018). https://www.ecfr.gov/current/title-42/chapter-IV/subchapter-B/part-422/subpart-D/section-422.152
  27. Resneck JS, Quiggle A, Liu M, & Brewster DW (2014). The accuracy of dermatology network physician directories posted by medicare advantage health plans in an era of narrow networks. JAMA Dermatology, 150(12), 1290–1297. doi: 10.1001/jamadermatol.2014.3902 [DOI] [PubMed] [Google Scholar]
  28. Requirements for dual eligible special needs plans, 42 C.F.R. § 422.107 (2019). https://www.ecfr.gov/current/title-42/chapter-IV/subchapter-B/part-422/subpart-C/section-422.107
  29. Requirements relating to basic benefits, 42 C.F.R. § 422.101 ( 2018). https://www.ecfr.gov/compare/current/to/2024-09-11/title-42/chapter-IV/subchapter-B/part-422/subpart-C/section-422.101
  30. Roberts ET, & Mellor JM (2022). Differences in care between special needs plans and other medicare coverage for dual eligibles. Health Affairs (Project Hope), 41(9), 1238–1247. doi: 10.1377/hlthaff.2022.00463 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Rose L, Aouad M, Graham L, Schoemaker L, & Wagner T (2021). Association of expanded health care networks with utilization among Veterans Affairs enrollees. JAMA Network Open, 4(10), e2131141–e2131141. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Scoring Guidelines. (2024). Ncqa.org. https://snpmoc.ncqa.org/scoring-guidelines-latest [Google Scholar]
  33. Sen AP, Meiselbach MK, Anderson KE, Miller BJ, & Polsky D (2021). Physician network breadth and plan quality ratings in medicare advantage. JAMA Health Forum, 2(7) [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Shepard M (2016). Hospital network competition and adverse selection: Evidence from the massachusetts health insurance exchange. doi:10.3386/w22600 Retrieved from https://www.nber.org/papers/w22600
  35. Slade EP, Wu RJ, Meiselbach MK, & Polsky D (2023). Psychiatrist and nonpsychiatrist physician network breadth in dual eligible special needs plans. Psychiatric Services, Retrieved from https://ps.psychiatryonline.org/doi/full/10.1176/appi.ps.20220239 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. US Government Accountability Office. (2015). Medicare advantage: Actions needed to enhance CMS oversight of provider network adequacy. Retrieved from https://www.gao.gov/products/gao-15-710
  37. USDA ERS. (2023). Rural-urban continuum codes. Retrieved from https://www.ers.usda.gov/data-products/rural-urban-continuum-codes.aspx
  38. Wallace J (2023). What does a provider network do? evidence from random assignment in medicaid managed care. American Economic Journal: Economic Policy, 15(1), 473–509. doi: 10.1257/pol.20210162 [DOI] [Google Scholar]
  39. Xu L, Welch WP, Ruhter J, Nguyen NX, Sheingold S, Lew ND, & Sommers BD (2023). Medicare advantage overview: A primer on enrollment and spending. Retrieved from https://aspe.hhs.gov/sites/default/files/documents/9b42ffbf2341726d5b63a9647b0aad15/medicare-advantage-overview.pdf [PubMed]
  40. Zhu JM, Meiselbach MK, Drake C, & Polsky D (2023). Network breadth for psychiatrists in medicare advantage is substantially narrower than in medicaid and ACA markets. Health Aff (Millwood), [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Zhu JM, Charlesworth CJ, Polsky D, & McConnell KJ (2022). Phantom networks: Discrepancies between reported and realized mental health care access in oregon medicaid. Health Affairs (Project Hope), 41(7), 1013–1022. doi: 10.1377/hlthaff.2022.00052 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Zhu JM, Zhang Y, & Polsky D (2017). Networks in ACA marketplaces are narrower for mental health care than for primary care. Health Affairs (Project Hope), 36(9), 1624–1631. doi: 10.1377/hlthaff.2017.0325 [DOI] [PubMed] [Google Scholar]

RESOURCES