Amidst recent and anticipated changes in health insurance markets, there has been a growing trend towards increased transparency of provider networks. An important part of this effort is the implementation of standardized, machine-readable provider directories. By enabling the aggregation of provider information and plan comparisons as never before, these directories permit researchers and regulators to advance what we know about health plan provider networks, including to what extent providers in those networks report that they are accepting new patients. This advance in network information and transparency comes at an important time—as policymakers consider new balances between regulatory requirements, product diversity, and consumer choice.

Background: Concerns with Provider Networks and Directories

With the establishment of health insurance exchanges in 2014, health insurers faced increasing pressure to narrow their commercial plan provider networks. Studies led by health economist Daniel Polsky and Janet Weiner and the consulting firms Avalere and McKinsey documented the “narrowness” of many of these networks. There are good reasons why health insurers chose to do this, including cost savings and the desire to drive business toward the providers that most effectively serve the unexpectedly sick and higher-cost populations that enrolled in these plans. Medical Loss Ratio-required rebates for high administrative costs forced insurers to look at the costs of maintaining providers that serve few plan members. Further, the exchange shopping experience, which, to date, allows consumers to compare plan cost sharing while providing scant information in most states on quality and number of providers, further incents the narrowing of networks. Studies suggest that narrow networks are not inherently bad. For example, research conducted by Simon Haeder et al. in Health Affairs and JAMA , in 2015 found no quality differential between narrow and broader network plans in the California market.

Nonetheless, consumer advocates and some regulators remain worried by the trend toward narrow network health plans. Heading into 2014, United HealthGroup and Anthem, to name just two, whittled-out certain hospitals and physicians from their Medicare Advantage and new Health Insurance Exchange networks. Although the insurers continued to meet regulatory network adequacy standards, these moves generated critical media coverage and queries from state insurance commissioners and members of Congress. Government watch-dogs were dispatched, and by the end of 2015, the Department of Health and Human Services (DHHS) Inspector General, the California State Auditor, and the Government Accountability Office had all issued unflattering reports on Medicaid and Medicare Advantage provider networks. Their studies suggested patterns of inaccuracies with reported provider networks, long appointment times in the Medicaid program, and gaps in oversight of provider networks in the Medicare Advantage program. Recent CMS provider directory oversight activities suggest continued widespread directory inaccuracies that prompted the agency to issue 21 warning letters. As of March 2016, six lawsuits were filed by enrollees against health plans claiming that they were misled into selecting plans based on incorrect information in the plan’s provider directory (one of these cases recently settled for $15 million).

A year ago, the National Association of Insurance Commissioners finalized a new model Network Adequacy Model Act. While the model does not have legal standing, it serves as a best practice for states to consider. Since that time, a handful of states — including Colorado, Georgia, Maryland, Ohio, and Connecticut — have passed laws or promulgated regulations establishing new or higher standards for health plan provider networks; a little more than half of the states were already conducting network reviews in keeping with much of the model act. Meanwhile, the Centers for Medicare & Medicaid Services (CMS) warned of additional oversight of Medicare Advantage provider networks in its 2017 Call Letter, and pushed state Medicaid agencies to do the same in 2018 via recent regulation.

Toward a Machine Readable Revolution

Most agree that consumers have a right to access accurate information regarding the physicians and facilities that are in network. Yet the error rate in provider directories is high; common problems include incorrect addresses and phone numbers and outdated listings of network providers and providers accepting new patients. Studies of selected Medicaid and Medicare Advantage network providers by the DHHS Office of the Inspector General and CMS have suggested error rates over 40 percent. Of course, many of the detected errors owe more to provider inattention than health plan negligence, and not all of these errors (i.e., incorrect fax numbers) impact the consumer experience. But the quantity of errors is, nonetheless, concerning.

While the implementation of machine readable directories has been, at times, painful and prone to glitches, it is producing some notable outcomes. First, CMS and some states use the directories as the data source for “doc finder” tools that enable a consumer to enter a physician or facility and easily find out plans in which that provider participates. Second, they permit the regulator to more effectively assess provider networks throughout the year. Third, they permit researchers to compare provider networks and launch other inquiries that were previously labor prohibitive. Regulators that become aware of directory errors and wish to monitor error rates will find it easier to check and recheck machine readable directories.

A handful of states that run their own health insurance exchanges—including New York, Washington, Maryland, and Colorado—required machine readable provider directories as early as 2014. CMS has required this data to be published in JSON (JavaScript Object Notation) format for federal exchanges since 2016, meaning that 35 additional states now have provider directories in a common machine readable format. Per last year’s regulation, machine readable directories will be implemented nationally for Medicaid in 2018. Absent a reversal of policy, a machine readable revolution is now underway.

In a related development, CMS is using the machine-readable data to pilot a measure that would allow consumers shopping on to compare plans by relative network size within a given county. This measure, called the Provider Participation Rate (PPR), reflects the proportion of participating providers in a specialty in a given county that are in a plan’s network. It is being piloted to categorize networks for primary care, pediatrics, or hospitals as Basic (Narrow), Standard, or Broad in four states (Texas, Ohio, Tennessee, and Maine) in 2017. Importantly, this approach stops far short of a “national provider database” that was foreshadowed in earlier CMS Medicare Advantage guidance and the agency backed away from a previous proposal to establish national network adequacy standards for Exchange plans. (A summary of the CMS network breadth pilot was published in the Health Affairs Blog last May). CMS’s goal appears to be ensuring “network transparency” rather than establishing national standards. States are the primary regulators of network adequacy, even in the federally run exchanges.

We are also beginning to see the results of research facilitated by machine readable provider directories. Daniel Polsky, Zuleyha Cidav, and Ashley Swanson, recently published research in Health Affairs which demonstrates a correlation between plan premium and network breadth. NORC at the University of Chicago has created customized analytics dashboards for health plans using this provider directory data. In an issue brief published in September, NORC released findings showing that what constitutes basic (i.e., narrow) and broad networks varies greatly across and within states, as does the overall network breadth of certain insurer types. Below, we also present first-time data on physicians’ new patient acceptance rates—analyses made possible by machine readable directories.

New Member Acceptance Rates

Some researchers, regulators, and advocates have raised concerns about health plans that may technically be “adequate”—in that they have enough providers to serve their existing members—but which are actually inadequate for new members who cannot find physicians willing to accept new patients. Critics have often portrayed this problem as particularly acute among patients seeking primary care physicians who are often in short supply. There has been little to no data analysis of new patient acceptance rates to show evidence of this problem, but that is beginning to change.

As part of their Provider Network IQ project, researchers at NORC gathered the machine-readable provider network JSON files made available by machine readable directories required by the Center for Consumer Information & Insurance Oversight (CCIIO) within CMS for plan year 2016. NORC then cleaned and aggregated the data files using the R statistical software package, focusing on de-duplication of providers and mapping networks to counties. In addition to specialty information included within the JSON files, NORC brought in ancillary data from the National Plan and Provider Enumeration System (NPPES) to ensure that each provider was mapped to the correct specialty. For 2016, our dataset comprises 1,172,626 unique providers and facilities participating in networks from 233 QHP issuers in the federally-facilitated marketplaces (FFMs).

Machine readable directories and the resulting data are far from perfect: health plans and researchers have noted the limitations of the current JSON specifications. This makes perfect compliance impossible, at least in year one. Beyond technical issues, the directories contain self-reported data with similar errors as the pdf and paper directories they replace.

Nevertheless, our findings represent the first data on this important topic, gathered by examining machine readable directories from exchange plans.

Acceptance Rates For New Patients By Geographic Type, 2016

Adult Primary Care93.4%92.2%92.5%97.1%

Acceptance Rates For New Patients By Type Of Insurer, 2016

 OverallBCBSIntegrated Health PlansNational For-ProfitCO-OPMedicaid
Adult Primary Care93.4%93.3%97.7%95.4%98.5%90.1%

BCBS: Refers to qualified health plans offered by the local BCBS affiliate plan.

Integrated Health Plans: In this study, we define an integrated health plan as a network of providers that care for a particular patient population. Our definition focuses on health systems that also offer their own health plan, as opposed to systems that are integrated but are not also functioning as payers.

Medicaid: Insurers offering qualified health plans in the exchange whose primary or original line of business was Medicaid managed care.

Nationally, 93.4 percent of adult primary care physicians participating in exchange plans in states where the federal government administers the exchange reported via the plan’s directory that they are accepting new patients in 2016. That rate is only 1.5 percentage points lower than the rate of pediatricians who reported that they were accepting new patients in 2016. The acceptance rate is notably higher in rural counties, which is important because the quantity of providers in rural counties is lower. Acceptance rates for different types of insurers—Blue Cross Blue Shield plans, integrated health systems, national for-profit plans, co-ops, and Medicaid plans—is fairly consistent. However, Medicaid plans have the lowest new patient acceptance rate, which may be a result of the lower reimbursement rates commonly attributed to plans that operate primarily in the Medicaid market.

Due to the limitations noted above, all data aggregated from the machine readable directories must be viewed with some skepticism. Our analysis required numerous rounds of data cleaning due to issues such as duplicate or missing provider identifiers, which necessitated bringing in ancillary data from the NPPES.

Products of the Revolution: Improved Accuracy and Regulatory ‘Harmonization’

To date, there has not been a study of the level of effort necessary to maintain machine readable directories — until this is considered, it is reasonable to wonder if the benefits of machine readable directories and other investments in increased network oversight merit the costs. At a time when the affordability of health insurance products is at the forefront of public discussion, it is important to be cognizant of anything that might further raise the costs of these products. But, as we discuss below, there are reasons to believe that the accuracy of these directories will improve over time and that there could be a corresponding reduction in costs to maintaining and using the data.

What does the future hold for provider directories? We believe, first, that we will see improved accuracy, and second, that machine readable directories will become the new normal across “harmonized” health insurance markets.

Improved Accuracy

Spurred by concerns with widespread inaccuracies, the health insurance industry and allied vendors are implementing solutions to improve the quality of provider directories. America’s Health Insurance Plans, the industry’s largest trade association, has implemented a pilot program featuring a single vendor providing directory checks as a shared service for subscribing plans in a given state. CAQH, a nonprofit alliance of health plans and trade associations, has implemented a program to harness provider credentialing information to improve directory accuracy. Second to None, a mystery shopping firm that supports many insurers has implemented a provider office outbound calling solution to confirm that provider offices offer the public the same information as listed in the provider directory. CAPG’s Sanator Provider Directory Initiative uses an application program interface (API) to track customer complaints about network data inaccuracy and then follows up with providers to address the issues. Quest Analytics offers health insurer clients a suite of provider information validation tools. And NORC’s custom analytic tool facilitates access to an unprecedented level of plan and county-level comparative information. While all of these solutions require time and investments, it is easy to see how these efforts will improve the accuracy and transparency of provider network information in the future.

The New Normal

Currently, machine readable directories will likely increase in use. They are already required in most of the exchanges. They will be required in Medicaid—the largest government managed care market—in 2018. CMS has previously spoken favorably about machine readable directories in Medicare Advantage guidance although it has not required machine directories in this program yet. In different guidance, as already noted, CMS has signaled a preference for “harmonizing” provider network requirements across Medicaid, Medicare Advantage, and the exchanges. Beyond the CMS-refereed markets, with most health insurers comfortably using machine readable files by 2018, more states may be tempted to bring it into the off-exchange commercial market, although states adapt to new technologies on their own timetables and according to unique market conditions. If policymakers seek alternatives to rigid network adequacy standards, the network transparency facilitated by machine readable directories could become attractive.

All of this suggests that in the next few years provider networks will be more easily and frequently measured, and directories will become increasingly accurate. The normalization of machine readable directories over time will also permit regulators and researchers to consider measures of network stability for the first time — by which provider network continuity can be considered alongside adequacy and accuracy. There could soon be a lens for understanding differences in health plans across insurance markets — commercial, Medicaid, and Medicare. All of this will, ultimately, revolutionize our understanding of how health plans and regulators ensure enrollees have access to providers and needed services.

Authors’ Note

We thank Aaron Wesolowski and Matthew Green of NORC and Samantha Strong of FaegreBD for providing research and analytic support in developing this essay.