3

State-Level Data and Collaborations

Lynn Blewett of the State Health Access Data Assistance Center (SHADAC), which is located at the University of Minnesota, noted that SHADAC works with states to leverage state and federal data in the interest of informing policy. Blewett said that large federal data projects using electronic health records (EHRs) and linked data are critical to patient outcomes research, but the time lags in the availability of these data make the results less actionable for state health policy. State analysts are dependent on federal agencies for data access, and congressional objectives may trump state needs. For these reasons and others, local data collaboratives informed by communities of patients, providers, and payers are key to informing state health policy. These data collaboratives make timely and targeted data projects possible, resulting in information that is focused on state needs and priorities and that state policy makers can use.

Blewett discussed several examples of state initiatives. The first example she discussed was the All-Payer Claims Data (APCD) databases. These databases collect and harmonize claims data from public and private payers and include patient demographics and provider codes as well as clinical, financial, and utilization information. The purposes and mechanisms enabling these databases vary across states. The primary objectives are to better understand the financing of health care at the state level, to inform state health reform activities, and to evaluate the outcomes of state reform strategies. The APCD Council provides a forum for states implementing APCDs to share information, expertise, and insight on their development

and use. Blewett said that 23 states have all-payer claims databases, and 6 more are being implemented.¹

Advantages of APCDs include the following:

• Cover the majority of residents in each state;
• Include geographic representation;
• Capture longitudinal information on a wide range of individual patients, providers, and payers;
• Offer comprehensive utilization and spending data at the state level;
• Are mandated by state legislation; and
• Receive federal funding through various initiatives.

Challenges among APCDs include the following:

• Data access for researchers varies by state.
• They provide no data on use of services by the uninsured.
• States cannot require a self-funded insurer and its third-party administrator to share claims data with a state APCD.
• They lack standardization of encounter-level claims from capitated health plans.

Another example discussed by Blewett was a voluntary local health system collaboration, the Minnesota EHR consortium COVID-19 project. There are 11 health systems that are voluntarily participating in this consortium to provide public health surveillance data in close to real time for decision makers. While discussions have been ongoing related to a variety of diseases, the collaboration quickly materialized at the beginning of the pandemic. Blewett said that no patient-level data are shared between systems. Vaccination information is reported by the state and then linked to participating EHR systems; summary data are aggregated at a central site. The project captures about 90 percent of the initial 1.5 million first and second vaccine doses administered in the state.

The following are what has worked in the data sharing consortium:

• Skilled and innovative researchers are embedded in the health systems.
• A distributed data network model is followed (avoids concerns about data privacy and simplifies data use agreements).

___________________

¹https://www.apcdcouncil.org/state/map.

Challenges in the data-sharing consortium include the following:

• There is interest in adding smaller independent clinics and Federally Qualified Health Centers, but that is more difficult and it is costly to build up infrastructure for data submissions.
• Race/ethnicity data need improvement.
• The consortium needs sustainable funding.
• There is a need for improving communications, engagement, and dissemination.

Blewett also discussed the Medicaid Outcomes Distributed Research Network (MODRN), a collaboration to analyze Medicaid data across multiple states to facilitate learning among Medicaid agencies. Participants include AcademyHealth’s State-University Partnership Learning Network and the Medicaid Medical Director Network. This distributed data network allows states to retain their own data and analytic capacity while being able to compare their outcomes data to those from other states. As part of this initiative, 11 university–state partnerships now participate to provide a comprehensive assessment of Medicaid treatment quality in addressing opioid use disorders.

The following are what has worked in MODRN:

• Distributed data network model (avoids concerns about data privacy and simplifies data use agreements);
• Engagement of local universities that have analytic expertise with state Medicaid analysts; and
• Collaboration around policy priorities and closer to real-time analysis.

Challenges with MODRN include the following:

• State participation is limited.
• Data sharing agreements and data use agreements are still required for university-based research access to data files unless all of the analysis is run by the state.
• Financing is needed to support sustainability of network/models.

Blewett offered four overall conclusions based on her experiences: (1) Locally based collaborations that are close to policy makers and decision makers are more feasible and more actionable for state health policy; (2) State regulatory requirements can be leveraged to facilitate data collection and then develop infrastructure for research capacity; (3) Collaborative distributed data networks with motivated and interested researchers

embedded within health systems and public agencies can lead efforts to support targeted data and analytic needs; and (4) Federal financing of local models can be used to inform other activities across the states.

She added that SHADAC collaborates with federal partners to obtain state-level data from the National Health Interview Survey, but it is a “heavy lift.” Among the challenges, she listed the need to have analysts who know how to interact with the National Center for Health Statistics and have special sworn status (from the Census Bureau), and the need for a new proposal every year.

Marsha Lillie-Blanton, George Washington University, focused her remarks related to state-level data and collaborations on broadening the concept of equity to include equity across states. She noted that the policies, practices, and characteristics of geopolitical areas, such as states, matter in the efforts to improve access and quality and to achieve person-centered care.

While the federal role has increased, states continue to be the main drivers of coverage and care for low-income population groups, with Medicaid being the major player in the landscape. Lillie-Blanton pointed out that the states that decided not to expand Medicaid during the early part of the expansion were disproportionately southern states, which have large Black or African American and Hispanic or Latinx populations. This illustrates how states can become drivers of inequities in access to quality care and person-centered care.

Lillie-Blanton discussed the Nationwide Adult Medicaid Consumer Assessment of Healthcare Providers and Systems Survey (NAM CAHPS), a survey that she worked on while she was at the Centers for Medicare & Medicaid Services.² A nationally representative survey of adult Medicaid recipients with state-specific samples, NAM CAHPS is a collaboration among several federal partners as well as 46 states and DC. The data produced include state-specific NAM CAHPS files, which states can get access to on the basis of a data-use agreement. Lillie-Blanton echoed Blewett’s comment about frequent lags in making these types of data available to the states.

Considering both challenges and opportunities associated with collaborations of this type, Lillie-Blanton highlighted the following as some of the areas that need attention:

• Aligning priorities: Both federal and state partners need to identify the data collection effort as a priority.
• Cost issues: Funding needs to be allocated for this type of data collection and analysis.

___________________

²https://www.medicaid.gov/medicaid/quality-of-care/quality-of-care-performance-measurement/nationwide-adult-medicaid-consumer-assessment-of-healthcare-providers-and-systems/index.html.

• Longitudinal data: While the baseline data have value, ongoing data collection (even if only every 3 to 5 years) is needed.
• Methodological issues: Comparative analysis across states requires adjustments for state variations in variables that may be unmeasured or not well-measured.
• Data linking: Future federal/state Medicaid surveys will need to include permission in the consumer consent form to link personally identifiable information.

Lillie-Blanton underscored the need for developing partnerships based on trust. Some of the potential partners for federal agencies include Medicaid Agencies, the Medicaid Medical Directors Network, and Public Health and Behavioral Health Agencies. Other stakeholders to consider are professional associations, clinicians and provider groups, advocacy groups, and consumer groups. Academic institutions, policy research organizations, and foundations could also serve as partners.

In terms of building data capacity, two areas emphasized by Lillie-Blanton are (1) supporting the development of state Medicaid infrastructure for data collection, analysis, and reporting; and (2) developing training opportunities and funding for researchers for collecting and analyzing Medicaid data. She noted that some state Medicaid agencies already have strong infrastructures for data collection (e.g., MA, MI, NY, AL) or partnerships with academic institutions (e.g., PA, AR).

Todd Gilmer, University of California, San Diego, made three key points about state-level data and collaborations. First, state-level data are useful for understanding health disparities, because lower-income individuals and families, including those with a significant disability, are underrepresented in many national and commercial datasets, while Medicaid data can provide comprehensive coverage on diverse populations. Second, state-level data can be challenging to work with and to acquire. The learning curves can be steep due to complex and bureaucratic systems, and a path for accessing protected data is not always clear. Furthermore, a high prevalence of Medicaid managed care may result in uneven data quality. Due to these challenges, long-term collaborations can facilitate the interpretation of and access to data. Third, promising opportunities also exist at the county level.

Among the types of data available at the state level, Gilmer highlighted the following:

• Medicaid data:
  • Provide information on health insurance coverage for low-income families and individuals (in expansion states), low-income elderly, and people with disabilities; and

• • Cover a racially/ethnically diverse population with complex health conditions.
• Inpatient and emergency department discharge data:
  • Are available in most states;
  • In California, are provided by the Office of Statewide Health Planning and Development; and
  • Are aggregated by the Agency for Healthcare Research and Quality (AHRQ), which maintains a national database Healthcare Cost and Utilization Project (HCUP) for these data, although with less detail than is available in the state data.
• State-level surveys and indices:
  • Can provide local-area information;
  • Have a greater focus on health disparities compared to national surveys and are potentially more customizable; and
  • In California, are derived from the California Health Interview Survey and the Healthy Places Index.

Gilmer highlighted Medicaid datasets as especially valuable. These datasets have fairly comprehensive coverage, including on medical care and pharmaceuticals. They also have good coverage of mental health and substance use care, particularly for those who are in high need of these services. The datasets also cover home- and community-based services and custodial long-term care. Due to these characteristics, the Medicaid datasets provide a unique platform for studying special populations and topics of interest.

Medicaid programs are uniquely innovative, in Gilmer’s view. Among the innovations, he highlighted the comprehensive, statewide multipayer delivery system and payment reforms that reward value as opposed to volume and support improvements in population health delivery system and payment reform. He also noted that Medicaid provides integrated services for people with complex needs, such as high-risk children and youth; adults eligible for Medicare and Medicaid, including those with long-term care needs; and people with complex physical health, behavioral health, and social service needs. Gilmer also highlighted experimentation with alternative delivery strategies, such as the use of community health workers to build health literacy and peer providers with lived experience to increase engagement in health care.

Gilmer described the Transformed Medicaid Statistical Information System (T-MSIS), which aims to provide Medicaid data on all U.S. states and territories in a more timely way than the previous information system. While there have been some delays in implementing T-MSIS, all states are now reporting data, so it is now starting to become possible to do the types of analyses discussed above. There are also efforts underway to improve data quality, targeting 21 indicators, such as reasonableness of eligible counts, beneficiary demographics, and completeness of key claims service data elements.

Gilmer also highlighted several challenges associated with using state-level data:

• Data access:
  • Each state will have a unique process and Institutional Review Board (IRB) requirements to access data.
  • Data are fragmented, and memoranda of agreement are required for multiple systems.
  • A significant investment is needed for a single study.
• Data quality:
  • Each state dataset will have some unique characteristics.
  • The high prevalence of managed care may affect data quality.
  • There are limitations associated with data on race/ethnicity, language, and sexual orientation.

State data access and quality would benefit from long-term collaborations and investment, Gilmer argued. This would mean maintenance of merged datasets, a standardized process for data access, and a shared understanding of data elements.

Gilmer also briefly discussed examples of data available at the county level. There are many large counties in California, and elsewhere, and they often manage some parts of the health care system. For example, in California mental health and substance use services are managed at the county level. Gilmer said that county-level data may have more detail than data at the state level. For example, San Diego county records include detailed data on race, ethnicity, language, and sexual orientation. When these records are combined at the state level, some of the details are lost due to missing data. It is also important to note that counties provide other social and public safety services, and linking to those datasets presents additional opportunities.

Claudia Steiner, Kaiser Permanente Colorado, discussed datasets produced by AHRQ based on inpatient, emergency department, and ambulatory surgery discharge data. Discharge data are available in all states except Alabama. AHRQ creates five nationwide databases, using a sampling technique that allows estimation to the nation. The nationwide databases are the National (Nationwide) Inpatient Sample; the Kids’ Inpatient Database; the Nationwide Ambulatory Surgery Database; the Nationwide Emergency Department Database; and the Nationwide Readmissions Database. AHRQ also creates three statewide databases, available for some states that allow the distribution of the data: the State Inpatient Databases; the State Ambulatory Surgery Databases; and the State Emergency Department Databases.

The state-level databases include some identifiers that make it possible to link to other databases. Identifiers could include hospital identifiers, encrypted physician identifiers, patient state/county Federal Information Processing Standards codes, and patient zip codes. In some cases, for some states, AHRQ can link additional data at the hospital, physician, or patient level, with permission from the state data organization. State data organizations often have more restricted and fully identified data, and they can perform linkages to additional restricted data available within the state. Examples include linking to birth or death certificates, state-level surveys, patient-reported outcomes, and social determinants of health data collections. Steiner noted that AHRQ is currently actively exploring links to social determinants of health data as well as physician practice variables for Medical Expenditure Panel survey and HCUP data.

Steiner echoed a point made by other speakers, namely that access to state-level data can be challenging because each state has its own process and IRB requirements for accessing the data. She added that the AHRQ HCUP supports a consistent approach to accessing state-level data, and costs are mitigated in many cases. Unique characteristics of the data in some of the states was another challenge discussed by Steiner. AHRQ standardizes the data across all states with a consistently defined set of variables, so state data access and quality have benefited from long-term collaborations with AHRQ. Steiner argued that additional funding and collaboration across federal agencies and within the state partnerships could yield additional value and versatility.

DISCUSSION

The brief discussion that followed the presentations further highlighted the inconsistencies in the availability and quality of the data produced by the states. Comments echoed prior observations that the data collected are not necessarily collected for research purposes and research considerations might not be a priority. For example, while zip-code information is useful, the purpose of zip codes is primarily administrative, which often does not represent the “on the ground” characteristics of an area. In addition, zip codes are sometimes changed. Speakers highlighted the need to develop partnerships built on mutual trust and benefit and to support state data collection systems and analytic efforts.

Participants discussed the additional time needed to aggregate state data at the national level. The delays affect some types of data more than others, and the extent to which having up-to-date data is necessary also varies by the type of data or question, but increased consistency among the states and automation could reduce the time necessary to produce national datasets.

CONCLUSIONS

Many states have robust data collection systems and can produce information that is useful to state and local policy makers. State-generated data are also valuable at the national level, including for answering broader questions about issues that may be influenced by local policy, such as health care access and disparities.

CONCLUSION 3-1: There are opportunities to learn from what states have accomplished in building data capacity.

The data collected, their quality, and ease of access all vary by state. Challenges associated with access, ranging from how the data are stored to the processes involved in accessing them, make the use of state-generated data for research at the national level particularly difficult. The lack of standardization and lag times in data availability present additional challenges.

CONCLUSION 3-2: The usefulness of data available for patient-centered outcomes research could be increased by the sharing and adoption of best practices among the states for the data collected, their quality, and ease of access.

This page intentionally left blank.