Re: Request for Information: National Directory of Healthcare Providers & Services
Dear Administrator Brooks-LaSure:
On behalf of Trilliant Health, I am submitting the following comments in response to the request for information (RFI) on the National Directory of Healthcare Providers & Services (NDH). Trilliant Health is a privately held analytics company that equips providers, payers, and life sciences companies across the United States with market research and intelligence to support a range of strategic, operational, and financial decisions. One of our analytics products is a proprietary provider directory that equips healthcare stakeholders with a comprehensive view of more than two million providers.
Trilliant Health appreciates the opportunity to provide feedback on topics related to the current shortcomings of provider directories, data validation and verification, and the management of duplicate information. These comments are based on our experience designing and developing our proprietary provider directory, which combines numerous public and private datasets, including an all-payer database that offers a longitudinal view into the claims history of 300M+ Americans. What follows are our comments in response to specific questions asked by CMS in the RFI.
Question #1: What are some of the lessons learned or mistakes to avoid from current provider directories of which we should be aware?
Challenges with Timeliness of Self-Reported Data. As CMS evaluates policies to ensure the timeliness of data in the NDH, we would emphasize the challenges that existing directories face with collecting self- reported information. As CMS knows, provider information changes frequently, as providers move office locations, pursue specialization, enter contracts with payers, and change organizational affiliations. Based upon an analysis of our all-payer dataset and directory of more than two million providers, we find that 38% of providers changed their office location between 2019 and 2022. Additionally, 19% of providers changed their employer during that same period.
Existing provider directories that rely on self-reported information are unable to collect updates frequently enough to keep up with the pace of change. For example, the National Plan and Provider Enumeration System (NPPES) requires that providers update their information within 30 days of a change. However, this requirement is weakly enforced, and many providers only update their information once, when they first register for a National Provider Identifier (NPI). Our research found that primary practice address was inaccurate on 60% of records in NPPES when compared against the provider’s recent claims history, despite legal requirements to maintain this information.
With these limitations, we encourage the consideration of datasets that are updated in near real-time, such as an all-payer claims database, to maintain the most accurate and timely information, without burdening providers to make more frequent updates.
Limitations of Directories to Represent Complex Relationships. As CMS evaluates lessons learned, we would call attention to the limitations of existing directories to inform users about provider relationships. Existing directories often force a one-to-one relationship between providers and organizations. However, the reality is that many providers are affiliated with multiple organizations. Based on an analysis of a three-year period, our research found that 50% of providers billed claims under at least two organizations, and 24% of providers billed claims under three or more organizations. To represent the extent of a provider’s practice more accurately, we recommend CMS consider a data structure that allows one-to-many relationships between providers and organizations.
Limitations of Directories to Provider Geographic Footprint. Most existing provider directories assume that a provider practices at a single address, but our research finds that 72% of providers practice at multiple locations. We would encourage CMS to organize the NDH in a way that represents all the locations where a provider practices. Associating multiple physical addresses with providers enables users of the NDH to develop a more accurate understanding of access to healthcare across geographies. This information is helpful not only to inform consumer choice but also for payer and provider use cases related to network adequacy, physician staffing, and community health needs assessments.
Question #2: The Healthcare Directory initiative and FAST both identified validation and verification as important functions of a centralized directory. What data types or data sources are important to verify (for example, provider endpoint information, provider credentialing) versus relying on self- reported information? Are there specific recommendations for verifying specific data elements?
We recognize that verification of all data elements across all records in the NDH is not feasible, due to the massive cost and effort required to verify a dataset of that scale. To assess the quality of the dataset in a more cost-effective way, we suggest the agency consider periodic random sampling, which follows standard quality assurance protocols and can provide measures of the expected overall accuracy of the entire dataset. With expected accuracy, the agency can assess which data elements may fall below adequate levels and revise policies to improve accuracy.
In addition to random sampling, the agency should give special attention to data types that directly impact a patient’s ability to make care decisions. As patients seek information to meet their medical
needs, they will evaluate providers based on a multitude of factors (e.g., convenience, brand recognition, severity of illness, transportation, cost, network limitations). Inaccuracies across certain data types may raise significant barriers to timely and cost-effective care. Below are some data elements that we recommend be prioritized:
-
Provider Qualifications. A provider’s credentials, specialization, and taxonomy reflect their qualification to render services. As patients research care options, misinformation may cause patients to make decisions that result in delays in care. One method to verify self-reported specializations, taxonomies, and services is to check their consistency with claims datasets. There are strong financial incentives for providers to capture the services rendered in claims, with robust systems in place to verify the validity of claims.
-
Practice Location. Patients will use geographic proximity to assess if it is feasible to receive care from a provider. Without accurate location data, a patient may unexpectedly encounter unreasonable travel distances or other barriers to care. As with qualifications, the agency may consider a provider’s claims history to verify practice locations.
-
Participation in Insurance Networks. We recommend closely managing data elements where misinformation could create surprise out-of-network issues, which can place extreme financial burdens on patients who are underinsured or have high deductible health plans.
- Accepting New Patients. Inaccuracies about whether a provider accepts new patients may cause difficulties accessing care, especially for vulnerable populations. For example, the November 2021 version of the Georgia Department of Community Health indicated that nearly 12,000 providers were accepting new Medicaid patients. However, a comparison against our all-payer dataset in September 2022 shows that 25% of these providers have not seen a new Medicaid patient since 2016. These errors may cause insurance networks to over-estimate adequacy, compounding access issues over time. As with other data elements, we recommend that the agency consider an all-payer claims dataset to verify the accuracy of self-reported information.
Question #3: How should duplicate information or conflicting information reported from different sources be resolved to balance the reporting burden versus confidence in data accuracy?
As CMS creates a centralized data hub, the agency will inevitably encounter conflicting information. As noted in the RFI, existing provider databases have numerous flaws, with variance in data recency and quality. The nature of limitations will vary across systems, with each system's data management and collection policies influencing the ways in which the datasets are flawed.
The response above describes our recommendations on sampling to measure the quality of information from different sources. We would encourage the agency to consider a sampling and review process that enables the measurement of data quality by source and by field. With this information, it is possible to maximize confidence in data accuracy while minimizing the reporting burden. The NDH should prioritize sources by favoring the most reliable sources, then using less reliable sources as fallbacks. This method prioritizes accuracy but not at the cost of completeness, as fallback sources can be used to fill in missing information. For example, we have found NPPES to be a reliable source of slow-changing information such as provider name and gender even though addresses listed in NPPES are not reliable.
Equipped with a “ground truth” dataset it is possible to use probabilistic machine learning models to merge, join and select records from competing data sources.
Conclusion
We appreciate the interest that CMS has shown in building the NDH as a single source of truth. Such a solution would improve the flow of information across the health economy. We appreciate the opportunity to respond to this RFI, and we hope that the commentary we have provided is helpful as CMS evaluates next steps for the NDH.
Sincerely,
Hal Andrews
President, Chief Executive Officer Trilliant Health