Establishing a health information exchange research network (HIERN) to support population health studies

Abstract

We propose the Health Information Exchange Research Network (HIERN) to advance population health research by leveraging state and regional HIEs and their representative real-world data. HIERN builds on existing governance and infrastructure to create centralized or distributed data repositories for longitudinal, multisource data integration. It helps study important clinical and epidemiological topics, such as Alzheimer’s Disease and Related Dementias, one of the fastest-growing chronic conditions in the United States.

Background

The COVID-19 pandemic exposed critical limitations of national health data infrastructure¹. Siloed data systems between healthcare and public health prevented cooperation at local, state and federal levels. The situation was exacerbated by the complex web of conflicting data privacy laws at the local, state, and federal levels, which undermined efforts to aggregate health data for population health.

The Health Information Technology for Economic and Clinical Health (HITECH) Act and the 21st Century Cures Act were pivotal to incenting the adoption of and interoperability among electronic health records (EHRs), which included endorsing existing Health Information Exchange organizations (HIEs) and encouraging new HIEs where needed². Community, regional, and state HIEs aggregate representative, population-level data for individuals living in defined geographies. While HIEs are primarily used for patient treatment and care coordination (i.e., “primary” uses of health data), they are increasingly being considered for “secondary” uses (i.e., reuses), including population health research. HIEs are local, usually non-profit organizations that facilitate the secure exchange of health data (including those from EHRs, as well as from laboratory and imaging facilities) among providers and other authorized healthcare stakeholders. Many states made appropriations to leverage federal matching funds to further enhance their exchange infrastructure, including for secondary reuses, to support their population health objectives^3,4. Today’s HIEs are more likely than ever to participate in national exchange networks and provide value-added services (i.e., analytics, quality measurement), and programs (e.g., Qualified Clinical Data Registry for CMS)^3,5. Notwithstanding, developing functional capabilities that support public health reuses of HIE data is required to support multisource, heterogeneous data integration for population health research, including addressing barriers to harmonizing longitudinal multi-modal data (clinical, social, environmental) from multiple sources and with varying state-specific data laws. This includes requirements for harmonizing and adopting enhanced security and privacy guardrails associated with integrated data or to incentivize the curation of high-quality data for reuse in research.

Technical challenges aside, it is fittingly stated that data moves at the speed of trust⁶. Effective multistakeholder governance overseeing data use agreements (DUAs) is essential for managing the complexities of policy regarding integration and utilization of multisource HIE data for reuse. While the Office of the National Coordinator for Health IT (ONC) created the Trusted Exchange Framework and Common Agreement (TEFCA) to be that national “interoperability floor” for data exchange, several characteristics of TEFCA limit its relevance for population health research. Specifically, the DUAs under TEFCA may be supportive of public health reporting—not population health research—and are focused on enabling data exchange as opposed to data reuse. Also, critical data, such as on behavioral health and social determinants, is inconsistently collected in EHRs and not mandated by TEFCA⁷. Finally, TEFCA participation is voluntary and superseded by state-level consent and privacy requirements, enabling at best a patchwork of coverage. Last fall, TEFCA had exchanged its 1 millionth message—an important achievement. By contrast, most of the nation’s 75 HIEs had exchanged 1 million messages between midnight and 10AM that same day. Hence, to facilitate meaningful and timely population health research, HIEs play a key role beyond, and in tandem with, national interoperability frameworks.

HIE-based repositories

Key national population health data initiatives include PCORI’s Patient-Centered Outcomes Research Network (PCORnet), the CDC’s BioSense, the National Institute of Health’s (NIH) National Center for Advancing Sciences (NCATS) and the National COVID Cohort Collaborative (N3C), and the FDA’s Sentinel Initiative. While critical, many national population health data initiatives do not systematically harness the considerable data aggregation capabilities of HIEs or Regional Health Information Organizations (RHIOs) and typically rely instead on available data from payers, academic medical centers' catchment areas, or voluntary contribution of data from disparate sources. This convenience sampling limits the completeness and therefore representativeness of the data, severely limiting the scope of unbiased population health research that can be conducted. For representative and robust population health research, data should ideally be captured as a byproduct of all healthcare encounters in a certain geography for a whole population, which optimizes data collection and costs while also minimizing bias in the collected data. This is exactly how data is captured in most HIEs and thus, there is a real opportunity to build on the existing national HIE technical infrastructure and governance structures to develop specialized data repositories for population health research. While some HIEs may not be collecting data from all health care providers in their area, with the appropriate resource and governance support, HIEs are quite capable of aggregating and normalizing comprehensive and representative population-level, real-world data for research and organizations like the National Institutes of Health (NIH), Centers for Disease Control and Prevention (CDC), and Food and Drug Administration (FDA) benefit from this data resource for conducting their national public health functions⁸.

This is particularly relevant given the shift towards the “Health Data Utility” (HDU) model by the HIE industry. HDUs are regional or statewide entities that combine, enhance, and exchange electronic health data across care and service settings for treatment, care coordination, quality improvement, health equity, and public and community health purposes^7,9. Key characteristics of HDUs include their neutrality (i.e., neither provider nor payer nor regulator) and flexibility in meeting stakeholder needs; designated authority by region or state; advanced technical capabilities, multistakeholder, cross-sector participation; and most importantly, transparent, inclusive local governance in a non-profit model. These entities earn community trust and utilize consensus-based policy levers to advance data sharing and infrastructure for the aggregation and integration of multiple data sets from their expansive networks to payers, providers, community information exchanges, and social services, thereby including a broader range of data types (Table 1)¹⁰. Such capabilities enable HIEs, as they evolve into statewide or regional HDUs, to assemble representative health information on individuals from payers, providers, community support services, and government agencies, all bounded by local privacy laws and constrained by participatory community-level governance¹⁰.

Table 1.

Data types and sources HDUs may leverage

Data type	Summary	Example data sources
Demographic	Information about an individual’s age, gender, race, ethnicity, language, sexual orientation, address, contact information, medical record number, guarantor number, and other person-specific details.	Multiple sources with individual patient-level data, such as electronic health records (EHRs), registries, claims, and/or administrative data.
Claims	Information collected from health insurers related to clinical care provided and paid for by the health plan.	Commercial health plans, Medicaid, Medicare, Claims data systems.
Clinical	Information collected about a patient’s health-related information history, progress notes, symptoms, and diagnoses associated with treatment and care coordination. Data may include a patient’s demographics, medical, medications, treatment plans, immunization dates, allergies, test results, and radiology images.	Multiple sources, such as EHRs, clinical data registries, remote monitoring devices, image systems.
Administrative	Health care administrative data collected by health care organizations for administrative purposes, such as billing and payment, compliance with regulations, and managing health care services. Administrative data is typically used to support the administrative and operational functions of health care organizations.	EHRs, claims clearing houses, membership files, eligibility, and enrollment data systems.
Community and Service Resources	Information on community resources, locations, and services available for individuals, specific populations, or the community (e.g., food banks, homeless shelters, crisis intervention support, behavioral health services and availability).	Local government websites, licensing systems, Medicaid provider credentialing systems, CMS, National Plan and Provider Enumeration System (NPPES), payer contracted provider directories.
All-Payer Claims Database	Claims data registry data base across all Medicaid, Medicare and commercial payers in a state or region.	Commercial health plans, Medicaid, Medicare.
Prescription Drug Monitoring Programs	Dispensed controlled substance and non-controlled substance prescriptions data are collected and tracked in a state electronic database, known as a PDMP, and is used by health care providers and pharmacists for informed decision-making about prescribing and dispensing controlled substances. Data may include information on dispensed medications (e.g., drug name, dosage, quantity, date, refills), patients receiving medications, providers prescribing medications, and dispensers dispensing medications.	PBMs, EHRs and Prescription Information Management Systems.
Social Risk and Needs Data	Information collected, managed, and used by organizations that provide services to individuals, families, and communities. This type of data may include demographic, social needs and risk factors, economic, and health-related information about the individuals and families who receive services, and information about the services provided and their outcomes.	HUD Homeless Management Information Systems, State and county human service systems, community-based organizations, and state licensed human services organizations.
Public Health	Information collected and used to support public health services at the state and local levels in disease prevention, infectious disease surveillance, and interventions protecting and promoting health. Data sets may be available from national data sources, such as survey systems (e.g., Behavior Risk Factor Surveillance System (BRFSS), Pregnancy Risk Assessment Monitoring System (PRAMS), Children’s Health Assessment Survey (CHAS)).	Immunization Information Systems (IIS), disease registries, surveillance systems from clinical data sources (e.g., syndromic surveillance, case investigation, notifiable conditions (National Electronic Disease Surveillance System), Electronic Lab Reporting from commercial labs, state labs, or hospital labs). Specialized registries (e.g., birth defect registry, trauma registry, environmental health hazard surveillance, and electronic vital statistics systems).

(Adapted)¹⁰

State and regional HIEs may offer a holistic view of population health thanks to their unique ability to link data across healthcare settings, where data from hospitals, clinics, laboratories, pharmacies, post-acute care, and long-term care facilities can be integrated. The importance of this linking is emphasized by previous studies that have shown that Medicare patients see more than seven different providers each year and that primary care providers are expected to coordinate care with 229 other providers across 117 other organizations each year on average¹¹. Existing evidence of enhanced research capabilities made possible through HIE infrastructure and governance frameworks supports our hypothesis. For example, the Indiana Network for Patient Care created a specialized registry with clinical and claims data on patients with traumatic brain injuries, spinal cord injuries and stroke. The registry’s data enabled research that strengthened the evidence linking traumatic brain injury and ischemic stroke risk^12,13. HIEs also played a key role in providing data for epidemiological studies during the COVID-19 pandemic^14–17.

We, therefore, propose developing a national population health research-oriented HIE network, i.e., HIE Research Network (HIERN), that builds on the existing HIE infrastructure. Unlike EHR vendor-centric offerings, the rich longitudinal data available in the clinical data repositories (CDRs) of HIEs pools data from multiple distinct sources. This feature, coupled with their local governance authority over that data, suggests the possibility of combining the efforts of the HIEs to establish a population health research network across the country. Specifically, data from multiple, mature, well-established HIEs could help create a resource (data governance + data + analytics) for research on disease diagnoses, treatments, social determinants, and other information from real-world data sources on diverse communities. Unlike current approaches that rely on limited datasets from single provider or EHR sources or administrative claims data, this approach would draw on more robust, representative data from all providers and all EHR vendors used in each community with the aim of supporting public health research with more accurate and generalizable findings. International efforts such as the European Health Data Space and other hybrid federated–central research infrastructures provide useful parallels, and HIERN is conceptually aligned with these initiatives while leveraging uniquely US-based HIE governance and exchange infrastructure.

One of the features of this proposed approach is that HIEs receive their data in real time from most sources and thus can serve not only as the best source of historical information but also as the continuously refreshed source of the current state of the nation’s health and well-being. Thus, while many research repositories contain static data that quickly grows stale, a network of HIEs would, by their very nature, provide up-to-date information, ensuring that the repository reflects current clinical practices, patient populations, as well as the impact of changes in regulations or patient authorizations. A network built from HIEs would enable researchers to explore how different factors (e.g., socio-economic status, rurality, access to care) influence disease progression and treatment outcomes. Moreover, it would enable critical capacity to conduct population studies that are not confined to or biased by the type of care setting, coverage, political or policy environment, or socio-economic factors, avoiding many biases and constraints introduced by data collected in hypothesis-driven research. We emphasize that this manuscript presents a conceptual architectural and governance framework intended to guide implementation pilots rather than to report empirical results. Future HIERN work will explicitly evaluate and benchmark centralized and distributed approaches through real-world pilot deployments.

HIERN architecture

The limitations of HIE data, such as data mapping discrepancies (e.g., many HIEs do not map lab codes to Logical Observation Identifiers Names and Codes (LOINC) or map to older versions of LOINC), the inconsistent use of other controlled vocabularies (e.g., SNOMED CT), or non-synchronized update intervals often lead to non-uniform schema, variable HIE data granularity, and create temporal misalignments that distort longitudinal analyzes. For these reasons, we present both a centralized architecture for the HIERN repository, which would curate and store limited data sets (LDSs) from multiple HIEs, as well as decentralized architecture featuring data that remains within each participating HIE’s secure environment but is mapped to a common data model across the distributed network of HIEs.

While the eventual goal is to establish a decentralized HIERN virtual repository, we recognize that certain research questions may require a centralized repository to iron out data, technology, and governance aspects of data or to add specific data not normally collected by HIEs, such as genomic or imaging data. Not all projects can be completed efficiently in a decentralized system, particularly when major transformations of data, including for data linkages, are required to answer a research question of interest. For these reasons, we describe the architecture for a centralized repository as one possible approach. This centralized approach is likely to assure fit-for-purpose data for research questions of interest, as well as replicability of results for regulatory use by federal agencies. Furthermore, a centralized architecture may be justified for certain AI and surveillance use cases that may better leverage such a model, requiring more extensive clinical and unstructured data than usually curated in HIEs, such as use cases requiring raw, high-dimensional data (e.g., clinical notes, imaging data).

Data sharing with a centralized node may not be feasible in many instances, such as due to legal or policy constraints. In these cases, federated learning approaches may be employed, which allow multiple parties to collaboratively train a model without physically moving the data, wherein the model is trained at the data site, and only the model parameters derived from the data are shared with a federation server, which aggregates them across users and sends back the updated model to the users. At the same time, however, it is to be recognized that centralized data approaches would need to address more legal and regulatory issues, may be more costly, and there may be less willingness from HIEs to export data out of their own secure systems. Nonetheless, both architecture options would require mapping HIE data to a common model like the Sentinel Data Model or Observational Medical Outcomes Partnerships (OMOP) Data Model, and we feature a coordinating center across both architectures to be tasked with supporting these processes, akin to the Sentinel program.

Design A: centralized architecture

Figure 1 presents a centralized architecture for HIERN and its data flows using an example set of three HIEs that contribute their LDS to a HIERN Coordinating Center (HCC) cloud environment. The LDS will contain dates (e.g., encounter dates, procedures dates) and zip codes so that queries can address temporal and geographical aspects. Once LDSs are created by the HIEs, they are also replicated to the HCC cloud environment under appropriate HIPAA and governance policies for the use case. Note that while covered under HIPAA and minimizing re-identification risks with Expert Determination or Safe Harbor de-identification of data¹⁸, some HIEs may consider even stricter rules about sharing data with outside entities.

Fig. 1. Centralized HIERN architecture.

All HIEs receive real-time data from multiple data sources (hospitals, health centers, labs, etc.) which are first analyzed by an Enterprise Master Patient Index (eMPI) technology to match individuals from different data sources and link records longitudinally in the CDR. The CDRs normalize data to national data standards and code sets following ONC requirements. HIEs would then likely create a replica from their CDR which mirrors the databases in real-time or extracts a subset of the data required for epidemiological and population health studies. This allows the HIE teams to perform any manipulation of data in the CDR without affecting the real-time production environment of the HIEs which is intended to efficiently serve clinical providers with data. Using the extracted data from CDRs, the HIEs can create a limited data set (LDS) as defined by HIPAA by removing 16 of the 18 elements considered as protected health information. In a limited data set, the following identifiers are removed: names; street addresses (other than town, city, state, and zip code); telephone numbers; fax numbers; e-mail addresses; Social Security numbers; medical records numbers; health plan beneficiary numbers; account numbers; certificate license numbers; vehicle identifiers and serial numbers; device identifiers and serial numbers; URLs; IP address numbers; biometric identifiers; and full face photos. Health information that may remain include dates (e.g., admission, discharge, service, date of birth/death), city, state, five-digit or more zip codes, and ages in years, months or days or hours.

Using cloud data replication technologies can simplify by reducing the need for complex Extract, Transform, and Load (ETL) procedures, where each of the participating HIEs has a mirrored copy of its database on the cloud platform used for the HIERN, and relies on a single, reusable transformation pipeline that standardizes the data to a common data model, thereby cutting duplicate ETL processes. Once in the central HIERN environment, research teams can then expose the combined data to query tools such as i2b2 without direct record-level access. The i2b2 is a self-service, open-source “cohort discovery tool” that allows you to explore and query clinical data that has been deidentified and aggregated. The tool has additional features that standardize and make it easy to run exploratory reports as well as set data standardization^19,20. Of course, the centralized repository will require enhanced security and cyberattack protection protocols and features, including for monitoring all traffic and exchange of information, access authentication tools for access control, permissions, updates, and other functions. Feasibility queries (or Prepare to Research—PTR queries) from authorized researchers can be run to assist them in determining cohort availability for their specific studies. Depending on the needs of the study, privacy-preserving record linkage (PPRL) solutions may be applied to link data across multiple HIEs to further combine records of patients from different regions and achieve longitudinal records for specific patients²¹. Once a study is deemed feasible, a researcher may choose to request a deidentified or LDS extracted from the HIERN for the actual execution of the research activities using defined protocols.

Design B: distributed architecture

Figure 2 presents a distributed HIERN architecture—Examples of regional HIEs organizing within a distributed architecture already exist, albeit in the context of public health data reporting and not research. Specifically, CRISP shared services (CSS) provides shared technical services across multiple HIE organizations serving Maryland, Connecticut, Virginia, West Virginia, Washington D.C., and Arkansas, and achieves this within a distributed architecture. This architecture is similar to the one used for Sentinel Initiative, which is the Food and Drug Administration’s (FDA) postmarket surveillance system that accesses post-adjudication claims data for over 300 million individuals²². The quality assurance and pharmacoepidemiologic query tools are designed and developed by a coordinating center in collaboration with FDA experts and run on a Sentinel Distributed Database²². Another exemplar for this network is the OHDSI Project, which has been adopted by researchers all over the world to conduct studies utilizing high-volume clinical and administrative data on a global scale from hundreds of domestic and international data sources. The OHDSI project provides a robust tool set to support the organization of the data and the distribution of analyzes. Each node in a decentralized network will use a standardized common data model. For example, the Sentinel Initiative uses a Sentinel Common Data Model, and other data models have also been used for other decentralized systems like OHDSI, PCORnet and others^23,24.

Fig. 2. Distributed HIERN architecture.

In a distributed architecture, the LDS is created within the firewalls of each HIE and queries and analysis packages are sent to each node from a coordinating center. Each node in the network relies on a standardized common data model and shares results to a central coordinating center for collation and interpretation. Orange lines represent LDS contributions. Dotted purple lines represent data query and response flows. Dotted green lines represent data sharing flows.

The decentralized HIERN will follow principles already adopted by Sentinel and PCORnet regarding receiving research queries centrally at the HCC and distributing these to each node to run the queries^25,26. Rather than duplicating these programs, HIERN complements them by filling gaps in timeliness, coverage, and activation speed, all of which frequently constrain population-health and public health analytics. While landmark population health studies were enabled by these programs, they do not enable streamlined population-wide analytics for critical needs, such as requirements for actionable data with near-real-time clinical signals, demands for regionally representative coverage across fragmented ambulatory and community settings, or rapid activation under a common legal and governance framework. However, multi-HIE research networks would be built atop operational exchange infrastructure that connects unaffiliated providers, hospitals and laboratories within and across regions, with cross-organization person linkage and low-latency feeds and participation agreements for pre-positioned governance that enable faster cycle times for studies.

State and regional HIEs rely absolutely on the trust earned by their local governance involving multiple stakeholders. HIERN, whether centralized or decentralized, will build on these trusted local relationships. The HCC will provide administrative support to the organizational governance and will include representation from each node of the HIERN. Transparency, access control, security, and auditability will be built into the governance architecture as described in the technical description above. The technical, governance and security implications of data aggregation in healthcare are never easy. Given the data quality, security, and governance issues of programs such as Sentinel, PCORnet, and individual HIEs underscore the advantages of a banding of HIEs within a network and a coordinating center, which would provide data science, security, and privacy protection expertise that exceeds the capacity for most individual HIEs to enable population health research. Of course, aggregating and normalizing HIE data is half the battle. Additionally, HIERN must develop infrastructure to support federated analytics, which may leverage existing technologies like DataSHIELD for non-disclosive analysis or PopMedNet for distributed queries. This topic merits a separate in-depth analysis, and we confined our focus in this paper to the data aggregation dimension, which is a key barrier at present.

Operational approach: one disease area at a time

A pilot HIERN focused on a single disease area can be the starting point for a group of HIEs to codevelop the interoperability enhancements that would be required for streamlined data aggregation from multiple clinical and non-clinical sources. Piloting on a single disease domain may also be necessary to narrow the legal and ethical compliance considerations around aggregating this data for population health surveillance and research²⁷. Since HIEs are community collaborations among various health systems and data providers, it is important to establish clear use cases and benefits of any data sharing beyond the original intent of interoperability for patient care. Thus, identifying the benefits to local populations for participating in research studies conducted through HIERN will go a long way in generating further local support and participation in the network, thereby reinforcing the learning health system approach.

The pilot project allows to implement and learn the practical aspects of operationalizing the network, assessing the requirements for activating multi-HIE data for research queries, as well as testing governance frameworks aimed at ensuring data privacy, confidentiality and security. Furthermore, while multi-state-HIE data would enable research across many disease areas, it will likely do so differentially across disease areas based on the availability of high-quality data from participating HIEs in respective states. Testing the network for the type of questions that could be answered will be an ongoing task. For instance, certain queries may require linkages with All Payers Claims Data or with dispensed drugs information from local pharmacies, and establishing these linkages would enhance the capabilities of the network but would certainly be more permittable when requested on a use-case or disease area basis rather than blanket linkage for all patient data in the HIERN.

Moreover, by earmarking investments to specialized registries—focused on a single disease domain—as well as to a proscribed use case (population health research), the technical complexities for standardizing data models may be reduced and it facilitates the finding and managing of data by individual researchers. A diversified set of funding options could be relied on to launch and sustain the network, such as existing federal discretionary grants and cooperative agreements through designated federal agencies; state-based funding and appropriations, such as Section 1903(a)(3) of the Social Security Act’s Medicaid Enterprise Systems and Federal Finance Participation (MES FFP) program, which allows funding to Medicaid agencies for modernizing data infrastructure and interoperability, or Section 1115 Waivers and Demonstration Projects are also applicable, which allows states to apply for waivers to federal requirements that add flexibility in using federal funds. Furthermore, HIERN could charge participation fees to payers, accountable care organizations, managed care organizations, research institutes, and other data users through different funding structures, including membership fees, tiered access fees, transaction fees, implementation and onboarding fees, or a hybrid fee model²⁸.

Reidentification risk is non-trivial with either a centralized or federated HIERN architecture^29–31. A multi-state HIERN must maintain compliance to relevant privacy laws across multiple jurisdictions. At the federal level, all HIEs would comply with the Health Insurance Portability and Accountability Act (HIPAA) and the mandates from the 21st Century Cures Act around information blocking and standardized application programming interfaces (APIs) for data access. Moreover, the Office of the National Coordinator for Health IT’s TEFCA provides the national framework for network-to-network exchange. HIE governance frameworks have adequately protected patient privacy and data security over the years. However, a HIERN would pair these with an additional multi-party trust framework that accounts for varying state laws, tracks provenance and consent at the record level within each state, layers data segmentation with consent metadata, assesses reidentification risks based on data type, data breadth, and ability to link to external data sets, as well as implement technical controls such as role-based access and multiparty computation. This enables governance based on the state of origin of data, data type, consent, user type, and the intended use. Overall, HIERN data governance will reflect the data use and governance stipulations of the respective state/regional communities, providers, and patients represented by the participating HIEs, with consensus-based decision-making, i.e., through a governance board, advisory and technical committees, and independent auditors.

A key objective of the initial disease-focused pilot will be to conduct a structured technical maturity survey, data quality, and governance readiness assessment across participating HIEs, which will serve as an empirical feasibility study for subsequent HIERN expansion. Critically, while we recommend that the HIERN start by pursuing the development of a pilot repository focused on a single disease domain, developing yet another siloed repository should be avoided. Thus, HIERN’s pilot must be designed and built on reusable components that can be leveraged to scale beyond the beachhead disease domain chosen for the pilot and scaled up as a decentralized data system for future sustainability. The single-disease pilot would serve to, in parallel, codevelop a disease-agnostic infrastructure required for a HIERN, including PPRL, provenance, computable consent metadata, data segmentation, data quality assessments (e.g., on conformance, completeness, timeliness), and semantic normalization. The pilot would also serve to more accurately assess the costs and appropriate financing mechanisms for startup and sustained operations, and support cost-benefit analyzes, including with regards to projected or achieved public health and health equity outcomes supported with HIERN data.

Use case example: Alzheimer’s disease & related dementias (ADRD)

This ADRD example is intentionally presented as a quasi-feasibility scenario to illustrate how HIERN infrastructure would be operationalized for a real disease domain, including cross-site linkage, data harmonization, and governance processes required for population-level analyzes. The growing prevalence of ADRD underscores the urgent need for robust, representative datasets that can support research into prevention, diagnosis, and treatment. ADRD diagnosis varies by geography and the likelihood of receiving ADRD diagnosis varies based on the place of residence³². The role of geography and other social determinants of health (SDOH) factors cannot be ignored in the diagnosis and access to proper treatment for ADRD. Moreover, the emergence of novel FDA-approved treatments that target beta-amyloid, a protein that is implicated in AD, has prompted apprehensions regarding the equitable distribution of these therapies, given the historical under-diagnosis of ADRD in minority populations and younger adults³³. People living with dementia experience barriers to obtaining a formal dementia diagnosis, and person-level barriers to diagnoses are poorly understood, in part due to limitations in the available data commonly used across studies, missing the inclusion of factors in such analysis which are likely to be important risk factors for undiagnosed dementia, such as marital status, primary language, and nonmedical determinants (e.g., living alone, transportation challenges, home safety).

Current ADRD data repositories, such as the Alzheimer’s Disease Neuroimaging Initiative and the National Alzheimer’s Coordinating Center, while quite valuable for ADRD research, may be somewhat limited by a lack of diversity and limited geographical coverage^34,35. HIERN could fill these gaps with longitudinal real-world data from diverse patient populations with pre- and post-diagnosis data on ADRD, including diagnostic codes, imaging results, laboratory tests, demographics, and SDOH, as well as other potentially confounding or contributing conditions and factors. This would support research in the prevention, treatment, and prognosis of ADRD through research inquiries that are otherwise not feasible with existing data sources, such as analyzing the impacts of early interventions on ADRD incidence and outcomes, identifying factors impacting access to new treatments, or conducting stratified analyzes of ADRD morbidity and mortality to strengthen hypothesized correlations, or reveal new ones, between disease and sociodemographic factors (e.g., estimating the independent association of SDOH factors with the probability of diagnosed ADRD, adjusted for dementia risk within race and ethnicity subgroups).

To operationalize such analyzes, HIERN would rely on established techniques for cross-domain record linkage, data harmonization, and governance-controlled data access. Within each participating HIE, longitudinal patient records are first assembled through enterprise master patient index (eMPI) systems that probabilistically or deterministically link encounters across providers using demographic attributes and encounter metadata. For multi-state studies, PPRL approaches could be employed to link records across HIE nodes without exposing direct identifiers, for example, through cryptographic hashing or tokenization techniques applied to shared linkage variables. Once records are linked, participating sites would transform their data to a common data model, such as the Sentinel or OMOP data model, to enable harmonized analysis across heterogeneous source systems. Secure networking and analysis would follow established distributed data network approaches in which standardized analytic queries are distributed from a coordinating center and executed locally behind each HIE’s firewall, with only aggregated results returned.

Because HIERN relies primarily on HIPAA-defined LDSs, research data sharing would occur under DUAs between participating HIEs and the coordinating center. These agreements would specify permitted uses, reidentification prohibitions, and security safeguards. For observational studies, waiver of patient consent will be sought from institutional review boards (IRB). Additional governance mechanisms, including role-based access control, audit logging, and IRB oversight where applicable, would ensure that data use remains consistent with federal and state regulatory requirements while enabling population-level research.

Discussion

This paper makes the case for establishing a HIERN repository using a network of HIEs and presents some practical architecture details of such a network. The HIERN presents several potential advantages. In addition to providing comprehensive data on diverse populations across state and regional HIEs, it builds on established technical infrastructure and governance structures that are in compliance with federal and state laws, which are continually enhanced and expanded by states. Furthermore, HIEs collect and curate data in the routine delivery of care. Since this approach relies on consolidating clinical data from existing processes without additional effort, it optimizes the operational costs for large-scale data collection and curation while also reducing the burden on individual data providers to export data for specific research purpose. Hence, this approach promises to be a sustainable and efficient methodology for aggregating multi-source EHR data without placing an additional demand on health systems. Of course, participating HIEs will need to conduct follow-on harmonization of their data and data use policies (e.g., semantic standards, metadata standards, privacy and security safeguards, permitted reuses), but could minimize the administrative burden for establishing such standardization by using common tools and strategies, as well as leveraging recent “interoperability AI” tools³⁶.

Another advantage is that the quality of HIE data will continue to improve through federal regulations seeking to further standardize EHR data across health IT vendors, including through enhancements in data elements to the United States Core Data for Interoperability (USCDI), i.e., the standardized data set of health data classes and constituent data elements and use of Fast Healthcare Interoperability Resources (FHIR) APIs for data exchange. Moreover, HIERN can also spread the financial burden of technical enhancements amongst participating HIEs, thereby reducing the individual costs while gaining efficiencies through shared expertise and infrastructure. A comprehensive and almost real-time data resource like HIERN can also reduce costs for conducting clinical trials and postmarket surveillance of pharmaceutical products by the private pharma industry, as well as the FDA. Finally, while only a few states have designated an entity or group of entities to serve as the states’ HDU, we anticipate the number will increase given the consolidation trend in the industry, which reflects the increasing importance of economies of scale to compete with various alternative options for supporting interoperable exchange⁵. The included stakeholders and the governance mechanisms for a state’s HDU would naturally vary, but the overarching function of collecting, aggregating, enhancing, and sharing more comprehensive health records from clinical and non-clinical sources remains the same. This presents an opportunity for step-change enhancements in the national capability for population health research that go beyond the aggregation of claims and EHR data to using other patient-related nonmedical data sources, which are so critical for population health research.

From an analytic perspective, the value of HIERN lies not only in data aggregation but in enabling reproducible, multi-site population health analyzes that are currently infeasible due to the fragmentation of real-world data across care settings, vendors, and jurisdictions.

Conclusion

Fragmented real-world health data impedes research. Building on years of investments made into HIE infrastructure and governance for creating population health repositories through a national HIERN would enhance research capacity without incurring the typically prohibitive costs of establishing new national data assets. This approach fosters a more integrated health data ecosystem and, as states continue moving towards the HDU model, the approach establishes dynamic, up-to-date repositories for population health surveillance and research.

Author contributions

A.K., E.O., T.W.C., D.C., D.K., T.P., D.P., N.S. conceived of the idea. A.K., A.A., E.O., X.L., J.B. wrote the main manuscript. T.W.C., D.C., D.K., T.P., D.P., N.S., G.S. reviewed and gave useful input. All authors approved the final version.

Data availability

No datasets were generated or analysed during the current study.

Competing interests

Anjum Khurshid and Eliel Oliveira are co-founders of a private company, PulsarHealth. All other authors declare no competing interests.