Comparison of Office-Based Physician Participation in Medicaid Managed Care and Health Insurance Exchange Plans in the Same US Geographic Markets

Key Points Question How does the percentage of office-based physicians who participate in Medicaid compare with participation in health insurance exchange plans? Findings In this cross-sectional study of 67 057 office-based physicians in 5 states, Medicaid managed care plans included more physicians than health insurance exchange plans in the same geographic markets. Meaning These findings indicate that physicians are likelier to participate in Medicaid physician networks than previously believed, with important implications for the ongoing debate about the role of Medicaid in expanding health insurance and reforming the US health care system.


. The mean percentage of physicians in Medicaid vs HIX physician networks for insurers operating in both the Medicaid and HIX markets
The forty-five degree line is indicated by the thick, dashed gray line. a This point in the figure refers to a county in New York. The x-axis value is the mean percentage of physicians in the Medicaid physician networks of insurers that participate in both markets in that county. The y-axis value is the mean percentage of physicians in HIX physician networks of insurers that participate in both markets in that county. The measure of network overlap is the Jaccard similarity, which is the percentage of physicians participating in any pair of networks divided by the number of unique physicians participating in either network.

eAppendix 1. Sample construction
To construct our final sample we started with 104 networks, 30 MMC and 74 HIX, and removed two HIX networks in New York where we had concerns about quality of the network data. In our study sample, 24 non-metro counties (2 in Washington, 5 in Kansas and 17 in Nebraska) had zero office-based physicians whose primary address was located in that county. Any networkcounty pairs with these counties were excluded from the analyses.
To obtain a standard set of physicians we merged the MCO and HIX physician network directories to the SK&A Office-Based Physician Database. The database contained 76,311 NPIs of active office-based physicians in 2017 with their primary address and the answer to the question "Do you accept Medicaid (yes or no)?" The NPIs that did not merge to the database were removed. A physician's primary taxonomy was determined from the NPPES NPI Registry.
Of the physicians that merge to the SK&A database we exclude 0.8% (507/62,845) of Medicaid physicians and 0.9% (576/67100) of HIX physicians who are missing a primary taxonomy code.
We kept physicians with a primary taxonomy group of Allopathic & Osteopathic Physicians and excluded physicians who primarily specialize in Geriatric Medicine or Geriatric Psychiatry since they primarily serve the elderly. We make this restriction since the Medicaid program generally serves as a primary source of coverage for the non-elderly, ages 0-64. We assigned a single county to each physician based on their primary address. For ZIP codes that mapped to multiple counties we assigned the physician to the county containing the plurality of addresses for that ZIP code. For each ZIP code, we identified the county containing the plurality of its addresses by using the Housing and Urban Development zip-to-county crosswalk for 2017. We excluded physicians from directories whose primary address was located in a county that was not served by that network (eTable 2). (20)

eAppendix 2. Regression specifications
The analyses presented in Exhibit 3 for the "mean percentage of physicians per network" are based on specifications of the form: where p indexes plans, c indexes counties, is the percentage of office-based physicians in a county that are covered by plan p in county c, is a full set of county fixed effects, and is an indicator equal to one if a plan is offered in Medicaid managed care and zero if the plan is offered on the HIX, and is a noise term. Each observation is at the plan-county level.
The results for the "percentage of physicians in any network" are based on specifications of the form: where m indexes markets, c indexes counties, is the percentage of office-based physicians in a county that participate in any of the plans in a particular market (i.e. Medicaid managed care and the HIX), is a full set of county fixed effects, and is an indicator equal to one if the market is Medicaid managed care and zero if the market is the HIX, and is a noise term. Each observation is at the market-county level.
The results presented in eTable 4 that included insurer dummy variables are based on regressions of the form: where p indexes plans, c indexes counties, is the percentage of office-based physicians in a county that are covered by plan p in county c, is a full set of county fixed effects, is a full set of issuer fixed effects, and is an indicator equal to one if a plan is offered in Medicaid managed care, and is a noise term. Each observation is at the plan-county level. eAppendix 3. If Medicaid physician networks are larger on average why do slightly more physicians participate in the Exchanges?
We document the number of unique Medicaid managed care and HIX physician networks in each county. For Medicaid managed care, we used publicly-available documentation from each state to identify the number of plans participating in each county. For the HIX, physician networks are listed by rating areas, which were mapped to counties using the HIX crosswalk.
However, plans may not be offered throughout an entire rating area. We used the HIX Issuer County Report to include only the physician network-county pairs where the issuer had at least one on-market plan served by that physician network in that county.
We define the overlap between two physician networks at the county-level as the number of physicians practicing in that county that participate in both physician networks divided by the number of unique physicians participating in either physician network, (i.e. the Jaccard similarity). To obtain a measurement at the county-level by payer type, we take the average of all pairwise combinations of networks of the same payer type within the same county.
We explore two interacting hypotheses for why at the plan level Medicaid managed care physician networks cover a higher percentage of office-based physicians than HIX plans but at the county level, fewer physicians participate in Medicaid managed care than in the HIX. First, we document the fact that there tend to be more HIX plans per county operating in our sample states (eFigure 3). Pooling data from our sample states, we find that there are 1.10 (0.01 -2.18) more Exchange plans operating per county in adjusted analyses (eTable 5). Hence, even if at the plan level Medicaid managed care physician networks tend to be broader, at the county level more physicians may participate in the HIX simply by virtue of there being more plans.
However, this relationship is complicated by the degree of overlap within the Medicaid managed care and HIX markets. If different HIX plans tend to cover the same set of physicians then