Cautious Optimism Building: What HIE Managers Think About Adding Artificial Intelligence to Improve Patient Matching

Abstract

Each year an estimated 440,000 medical errors occur in the U.S., of which 38% are a direct result of patient matching errors. As patients seek care in medical facilities, their records are often dispersed. Health Information Exchanges (HIEs) strive to retrieve and consolidate these records and as such, accurate matching of patient data becomes a critical prerequisite. Artificial intelligence (AI) is increasingly being seen as a potential solution to this vexing challenge. We present findings from an exploratory field study involving interviews with 27 HIE executives across the U.S. on tensions they are sensing and balancing in incorporating AI in patient matching processes. Our analysis of data from the interviews reveals, on the one hand, significant optimism regarding AI’s capacity to improve matching processes, and on the other, concerns due to the risks associated with algorithmic biases, uncertainties regarding AI-based decision-making, and implementation hurdles such as costs, the need for specialized talent, and insufficient datasets for training AI models. We conceptualize this dialectical tension in the form of a grounded theory framework on Cautious AI Optimism.

1. Introduction

Health Information Exchanges (HIEs) are secure information systems that allow medical providers to share and access medical information immediately across providers, unifying a patient’s information across those providers in a standardized manner, even when the information is stored and presented differently by each provider. HIEs serve as critically important interorganizational platforms for sharing patient information in an accessible manner across providers. They enable medical services and care providers to obtain a potentially comprehensive and unified picture of a patient’s medical history, across health systems boundaries (Furukawa et al. 2013), regardless of where the patient receives medical care throughout the course of her/his treatment (Gefen et al. 2019). As the use of HIEs as a strategic tool is gaining popularity and acceptance (Wu and LaRue 2017), they are increasingly becoming a part of the healthcare agenda across the United States (Holmgren and Adler-Milstein 2017). HIEs have demonstrated the capacity to serve as a critically important source of medical information at the point of care (Rosenthal et al. 2024). HIEs can lead to better and timely diagnoses, more informed treatment decisions, and fewer medication errors (Grannis et al. 2019). Importantly, using HIEs reduces mortality (Pai et al. 2022), as well as shortens length of stay and 30-day readmission rates (Janakiraman et al. 2023). HIEs also considerably reduce pathology and laboratory orders and the need for radiology imaging (Al Braiki et al. 2024). They also enable the sharing of patient data even when the electronic health records (EHRs), IT standards, and data formats are different across the providers from whom those data are polled. Such a provision of a transparent continuum of care across diverse medical providers, despite their differing IT standards and medical practices, is essential for providing safe and robust patient care and treatment (IEEE 2024).

However, realizing the aforementioned benefits depends on a critical prerequisite step called patient matching (Godlove and Ball 2015). Patient matching is the process of identifying and linking patient records to the correct person and ensuring that the right patient identity is returned when querying across HIEs before proceeding with treatment decisions. Essentially, obtaining the right information, about the right patient, and at the right time, adds significant value to the healthcare delivery process (Kaelber and Bates 2007). When patient matching is inaccurate, healthcare practitioners are unable to access a patient’s complete medical history (Riplinger et al. 2020), which can lead to serious adverse events, including medication errors, misdiagnoses, and even patient mortality (Kohn et al. 2000). Adding to the difficulties of patient matching is the problem of confusing patient consent models of opt-in/opt-out across the HIE ecosystems, which further underscores the complexities and challenges that affect HIE data flow and interoperability. Discrepancies in consent requirements across states also create confusion for both patients and providers, hindering effective data sharing. Providers often struggle to explain these models to patients, adding to the uncertainty and complicating informed decision-making. Patient matching is therefore a pressing need facing the U.S. Healthcare System. A 2019 study by the eHealth Initiative found that 38% of healthcare providers witnessed an adverse event within the two years prior because of patient matching issues. Additionally, it is also estimated that up to 168,000 medical errors occur in the U.S. each year as a direct result of patient matching errors. In 2019, The Emergency Care Research Institute (ECRI) listed errors in the patient identification and matching process as one of the top 10 threats to patient safety. As an example, inaccurate patient matching can result in major invasive procedures being performed on the wrong patient (Chassin and Becher 2002). One informant we interviewed shared the following about this issue:

Could you imagine the horror, you are going under anesthesia for a surgery 5, 4, 3, 2 and as you are going under you hear them say the wrong patient’s name. Thankfully they caught it, but could you imagine waking up in that panic!

Given the critical imperative to identify the factors that are standing in the way of fostering effective and robust patient matching in and across HIEs, our study endeavored to discover through a series of in-depth interviews with senior HIE executives the prospects and pathways to using artificial intelligence (AI) to improve patient matching.

HIE Patient Matching—The Challenges

HIEs are becoming very popular. As an Office of the National Coordinator (ONC) briefing reported in 2021, more than 6 in 10 hospitals engaged in key HIE transactions (send, receive, query), a 51% increase since 2017. Additionally, a 2021 study published in the Annals of Internal Medicine reported a 34% uptick in the number of specialists visited, and a 12% increase in visiting five or more providers per year (Barnett et al. 2021), thus adding to HIE repositories. The number of hospitals able to access data through an HIE from outside sources to the point of care significantly increased over the last four years, to 71%, showing more than ever that patients are mobile and that data repositories are abundant. This increased usage is partly due to the increase in the systems that generate the data that drive the proliferation of HIEs and electronic health records (EHRs). EHR usage has witnessed a staggering growth since 2008, and as of 2021, nearly four in five office-based physicians (78%) and almost all non-federal acute care hospitals (96%) had adopted a certified EHR system. This marks a substantial 10-year progress since 2011, when 28% of hospitals and 34% of physicians had adopted an EHR system (HealthIT.gov 2022).

Despite their immense promise, as our interviews with HIE professionals reveal, HIE patient matching efforts continue to be plagued with challenges due to data entry errors, poorly tuned algorithms, and a lack of systems integration (Vest and Gamm 2010). It is a matter of balancing type-I and type-II errors, which are false positives in that you matched records incorrectly, and false negatives in that you had the data but did not make a patient match, respectively. Moreover, our interviews suggested that drawing this balance is bound to be context- and patient-specific and must also consider the multitudinous pressures being put upon healthcare providers at large. The abundance of silos and the proliferation of data for one, combined with the recent popularity of EHRs and HIE usage, has increased patient data repositories and resulted in an increased awareness of data quality concerns in the context of patient matching.

Additional contributing factors to contend with across the HIE ecosystem are technical disruptions prompted by government policy. Furthermore, due to privacy concerns, the U.S. Congress, in 1998, prevented the government from promulgating a universal patient identifier (UPI) a limitation that persists to this day (Sragow et al. 2020). This outright prohibition on a UPI scheme precludes HIEs from benefitting from an agreed upon matching ID standard across healthcare boundaries. As one informant puts it, HIE managers are then left to build their own matching technologies, creating different starting points to the matching process by each HIE.

By not having a universal patient identifier, we are all building our own proprietary matching programs. When you think of it that’s not efficient, when you think about interoperability, we’re all starting from a different point based on maturity levels.

This misalignment, in and of itself, causes dispersion in the industry as to how patient matching is implemented and maintained. HIEs use different algorithms and a multitude of matching techniques to meet their needs, but, arguably, in a system that demands perfection and emphasizes prior standardization and interoperability, the stubborn persistence of hurdles and an excessive number of integration points in an already fragmented health system creates a strange conundrum. As stated earlier, value is generated when hospitals and other providers can accurately match patients as they traverse health systems, and this realization has HIE constituents and vendors vying for solutions that improve the process. Nonetheless, one may ask why this value is not being realized.

The core of HIE systems relies heavily on technology, and as such, this study engages senior HIE managers and experts on the different stages of the patient matching process. It examines the techniques used to optimize master patient management index (MPI) databases, and how queries for patient information are handled in real-time, as well as in backend processes such as ADTs (adds, deletes, and transfers).

This continuing dilemma also opens the door for exploring solutions through the use burgeoning innovative technologies such as AI. AI models leveraging health information exchange data show promising predictive capabilities, despite the challenges posed by them, according to a recent study (Borna et al. 2023). Additionally, public benefit capabilities through HIEs, such as patient education, geocoding health data, and epidemic and syndromic surveillance (Shaban-Nejad et al. 2018), have extensive implications. There is a sense of cautious optimism that AI holds the promise to improve patient matching processes. Nonetheless, there is also the realization of the need to balance these benefits with the potential cost of diminished human oversight and judgment, which still plays a critical role, as well as the risks, as we shall discuss further below, of changing the values of those involved in the process.

2. Literature Review

2.1. The History of HIEs

There is no denying that technical advancements have changed the way business is conducted, including in healthcare through electronic health records (EHRs). Arguably the first EHR was developed in 1972 by The Regenstreif Institute, spearheaded by Dr. Charles Clark, who wanted a way to keep track of his diabetes patients electronically through manual hard coding (McDonald et al. 1999). Then, in 1974, by adding database systems, Purdue University computer scientists Abramson and Nanamaker created a streamlined and efficient way to record and share clinical data (McDonald 1981). Further adoption ensued in the 1980s and 90s, and successful reports appeared, indicating the quality and efficiency of EHRs. Unfortunately, there were limitations, as the benefit only served the specific hospital and did not allow access to those medical data as the patient transitioned to other providers (Kuperman 2011). As pressure mounted in the United States to expand the use cases, President George W. Bush launched what would eventually be the Office of The National Coordinator (ONC), a government agency that would oversee building out interoperability and broader adoption and reform for effective EHR use (Allen 2004). This essentially sparked what we now know as Health Information Exchange.

As with other new industries, HIE systems also took time to develop their potential value, and problems arose in their 30-year evolution. On the value side, it has been established that when patient records are matched and data is accurately shared across health systems, positive outcomes arise. These outcomes include enhanced treatment efficiency (Everson et al. 2016), better informed clinical decisions (Halevy 2011), reduced patient admissions (Kash et al. 2017), reduced duplicative imagery (Lammers et al. 2014), and overall cost savings (Hersh et al. 2015). Conversely, the problems associated with patient matching are equally well-documented. When EHRs are not matched effectively or when barriers obstruct the matching process, numerous detrimental effects arise, including treatment delays (Sherifi 2019), medical errors during transitions of care (Slager et al. 2017), and incomplete patient information (Eden et al. 2016). These negative outcomes compromise not only patient safety issues (Greer 2020) but also early use, cost, and quality of care (Rahurkar et al. 2015). During the 2003 to 2014 period, stakeholders were bullish and valued the HIE systems’ promise despite multiple challenges (Rudin et al. 2014). An updated systematic review (2014 to 2018) showed that strides had been made, and, overall, HIE systems were heading in the right direction, but largely delivering on anticipated improvements incrementally in terms of quality of care and cost efficiencies (Menachemi et al. 2018). The growing allure of HIEs bears testimony to the convergence of technology and regulatory policy, thereby creating value for the entire healthcare ecosystem, and this has kept interest and investment high among the public and private stakeholders in the healthcare sector (Swain et al. 2015).

Nevertheless, HIEs consistently encounter obstacles and difficulties with patient matching. To develop a comprehensive understanding of these difficulties, it is necessary to uncover the current perspectives of HIE experts regarding both the potential benefits and inherent challenges faced by HIEs, as well as their views on AI as a prospective solution. A qualitative research approach, specifically employing a grounded theory framework, is especially advantageous. This methodology facilitates the discernment of the benefits, problems, and future promise of AI in addressing these issues, and aids in capturing the subtle nuances inherent in expert perspectives. The subsequent sections will further delve into the theoretical viewpoints that guided this study.

2.2. Patient Matching—Sources of Value

The literature on the value of HIEs shows that first, purpose and mission are paramount for success, with patient matching emerging as the foundational “brainwork” of HIEs. Efficient patient matching promises to reduce errors and improve data quality, thereby facilitating the seamless sharing of health information across different systems (Zhuang et al. 2020). Research emphasizes that accurate patient matching is not merely a procedural step but a critical prerequisite for achieving interoperability and, ultimately, realizing the core value proposition of data sharing across healthcare boundaries (Riplinger et al. 2020). Second, value realization is achieved through programmatic and structural elements that are essential for effective HIE operation. These include cultivating strategic partnerships and effectively collaborating with stakeholders, which proves to be an important aspect towards success (Heath et al. 2017), investing in robust infrastructure to facilitate efficient data query and exchange (Torres et al. 2014), prioritizing data normalization to ensure data quality (Albarak and Bahsoon 2018), and continuously refining patient matching algorithms to improve match rates proves to be a worthwhile best practice (Bouhaddou et al. 2011). While many organizations aim for sustainability and value-adding opportunities, an ecosystem approach, particularly one that fosters positive public benefit use cases for improved health outcomes (Shapiro et al. 2011), can unlock significant returns on investments. The U.S. government has spent $30 billion over a 10-year period to promote the use and sharing of electronic health records through HIEs (Liao and Chu 2012). So, gaining additional benefits by leveraging early government funding into HIEs is a welcomed byproduct. As one of our interviewees put it,

We are finally getting those benefits of early investment, our grants, it is paying off with what we are now doing, able to do, and deliver to our stakeholders. We are seeing, they are seeing it.

2.3. Patient Matching—The Challenges

The challenges associated with patient matching within HIEs are multifaceted and deeply rooted in many complexities (Rudin et al. 2014), but one area worth delving into in modern healthcare is data management practices. Research consistently highlights the detrimental impact of the patient data intake processes, where omissions, inconsistencies, and errors at the point of data capture significantly undermine the accuracy of patient identification (Rahurkar et al. 2015). This is further compounded by the unstandardized collection of demographic attributes data across diverse healthcare providers (Deng et al. 2023) and EHRs, creating significant barriers to interoperability practices and cross-boundary data exchange. Another area to consider in terms of problems would be data growth and fragmentation. The sheer volume of patient information generated today is unprecedented (Hulsen 2020), and this coupled with the persistent issue of data silos due to legal, privacy, or organizational constraints exacerbates these challenges, and leads to compromised data quality and potential patient safety risks (Bloomrosen and Berner 2022; Holmgren and Adler-Milstein 2017). Beyond the inherent challenges of data integrity, there is additional literature that focuses on the critical role of robust programmatic infrastructure and governance in ensuring effective patient matching. These studies point to the adverse effects of legacy technologies (Gopal et al. 2019) and poor architectures, key contributors to limiting the ability of HIEs to leverage advanced interoperability standards and data exchange capabilities. Furthermore, a rigorous data stewardship program is important to patient matching (Hripcsak et al. 2013), and often an unsung hero of matching strategies. Data stewardship is a never-ending, tedious, and at times, a thankless job, but it is imperative to the matching program to have the ability to review records that are in question when past data collections and the intake of attributes result in close matches. This has implications for integrity, patient safety and privacy. As one of our interviewees put it,

It’s a Sisyphean effort. You know, once you open that queue you’ve got 92,000 matches to review. It just feels like all of this effort. It’s difficult to see in any sort of tangible way how it is positively, you know, affecting you or positively influencing the data. The data, it never stops.

2.4. Patient Matching Techniques—The Current State

Algorithm usage in patient matching is an imperative in handling the large volumes and variations in data caused by the explosive growth of patient data. The literature shows a multitude of techniques and algorithms used to match patients. Deterministic, probabilistic, and referential are all algorithms that have been studied and are in use today by HIEs. It is also important to note that no one algorithm will achieve a 100% match rate (Riplinger et al. 2020) and we posit that there is a “right tool for the right job” analogy to be used in selecting algorithms, which we will uncover in our findings section. Probabilistic matching algorithms have demonstrated accuracy, with one study reporting 99.98% overall accuracy (Cox et al. 2012). The allure of implementing algorithms is to reduce duplicate patient record creation, in some instances up to 30% (Thornton and Hood 2005). However, deterministic algorithms can produce missed matches, leading to underestimated readmission rates and biased analyses (Hagger-Johnson et al. 2017). But other research points to deterministic wins in cases using specific identifiers like name and date of birth; this has shown high accuracy in linking viral hepatitis data (Bosh et al. 2018). Referential algorithms have their place as well, as in a study from Indiana, referential patient matching exhibited greater accuracy than probabilistic matching for health information exchange (Grannis et al. 2022).

The upshot of this literature offers strong testimony for the prevalence of inefficiencies in a multitude of areas, but, more critically, it highlights the pressures it can put on patient matching programs, and a major goal for all HIEs is to minimize detrimental matching scenarios, most notably of both false positives and false negative varieties (Parker and Adler-Milstein 2016). False positives occur when patient records are incorrectly merged; this poses a catastrophic threat, potentially resulting in serious patient injury or even death. Conversely, false negatives occur when valid patient data fails to match and leads to the creation of duplicate records (Just et al. 2016). This scenario contributes to treatment delays and medical errors and further exacerbates operational inefficiencies and cost increases through an HIE ecosystem. These outcomes underscore the critical need for comprehensive strategies that address the direct pressures contributing to patient matching errors.

3. This Study

The ongoing tension between the value and problems of HIE patient matching justifies the need for further exploration, positioning AI as a plausible, yet guarded solution. We employed a qualitative, constructivist, grounded theory methodology (Charmaz 2014) to develop a deeper understanding of how Health Information Exchange (HIE) experts perceive the value and challenges associated with their patient matching programs. The complexities of patient matching, including the lack of standardized metrics and techniques within the multifaceted HIE ecosystem, necessitate a research approach that can explore these nuances without preconceived theories. This inductive strategy prioritizes the lived experiences of participants, allowing for the development of theory directly from their professional insights.

We interviewed 27 HIE experts over a four-month period from May 2024 to August 2024. An initial pool of 35 potential participants was identified and approached. The informants were sourced using a combination of purposive and snowball sampling strategies. The informants were all senior HIE executives. Starting with New York State, we identified the informants through the professional connections of the lead author, himself a senior HIE executive in New York, and using a snowball sampling approach, requested those informants to put the lead author in contact with other HIE executives, including those also in other U.S. states. This was deemed necessary because we did not expect HIE executives to be easily accessible. Rather, as turned out to be the case, a referral from another HIE executive they knew became a necessary prerequisite to ensure participation in our study. Specifically, leveraging the first author’s professional network as a Chief Information Officer (CIO) of a New York State-based HIE, initial connections were made through personal contacts and the LinkedIn platform. Participants were also recruited after the research proposal was presented at several healthcare coalition committee meetings focused on patient matching. Subsequently, the participants made introductions to additional experts, enhancing the diversity of the sample.

This snowball sampling method facilitated access to numerous HIE industries leads—necessary, as there is no existing listing of those people. Once an informant completed the interview, we asked them to make introductions to additional participants in their professional circle. Those recommended were then pre-screened to ensure they had the applicable HIE ecosystem experience. The selected interviewees included HIE leaders, government policymakers, health technology experts, data scientists, payors, and healthcare providers. This diverse representation ensured a well-rounded view of the patient-matching landscape. All participants held positions ranging from Director of Operations to CEO, including some MDs who have transitioned into informatics roles, as well as vendors and consultants. This pool covered experiences across 30 of the 48 contiguous U.S. states that utilize HIE.

The final informant pool represented a diverse group of stakeholders, including HIE leaders, government policymakers, health technology experts, data scientists, payors, consultants, vendors, and healthcare providers (including MDs who have transitioned into informatics roles). Participants held positions ranging from Director of Operations to CEO and possessed a combined 581 years of relevant experience. These experts represented HIE operations and perspectives across 18 of the 48 states that utilize a Health Information Exchange system.

3.1. Interview Protocol

A semi-structured interview protocol was developed to guide the interviews. The protocol was created through an iterative refinement process, incorporating feedback from the co-authors to ensure the questions were designed to elicit meaningful and reflective responses. The protocol included nine open-ended question categories, each with relevant prompts for follow-up dialog. Consistent with a grounded theory approach, the protocol was treated as a flexible guide. The semi-structured interview protocol received Drexel University Institutional Review Board (IRB) approval #2402010358 in February of 2024. After the initial five interviews, the protocol was reviewed and amended to incorporate additional prompts that could further explore emerging themes and interesting directions identified in the early data. The objective of the protocol was not to test a pre-existing hypothesis but to inductively build a novel understanding of patient matching challenges and opportunities from the ground up.

We conducted interviews with our informants over a four-month period. Each interview averaged 50–60 min in length, with some extending to 90 min based on the depth of the discussion. All interviews were conducted via Zoom to leverage its audio, video, and transcription capabilities. In addition to the formal interviews, field notes were meticulously recorded and reviewed after each session. Further data were collected through follow-up communications to clarify specific points and explore topics in greater detail. These follow-ups occurred via video conference, email exchanges, and, in three instances, face-to-face meetings.

All the interviews were conducted by the lead author. The interview transcripts were analyzed and coded by the lead author, an HIE executive and subject matter expert, advised and overseen by the co-authors. There were but a few disagreements and those were clarified by the lead author, explaining in more detail what he did and why. Up to 50 interviews were initially approved by the IRB, 35 HIE experts were prescreened, and 27 accepted and completed the interview process to form the primary data source. Data saturation was indeed reached at 27 informants, as no new information or themes emerged from the data. Based on Charmaz (2014), we stopped collecting new interviews at that stage. In the paper we present representative quotes. Quotes that repeat themes brought up by other interviewees were excluded in the interest of being succinct and avoiding duplication. To provide context, the HIE community is not that large; only 48 states run HIEs. And in total in the USA, there are only about 120 HIE systems, with some states having multiple HIEs.

3.2. Data Analysis

A grounded theory approach was used for the analysis, focusing on the iterative process of data collection and coding to allow theory to emerge directly from the data (Charmaz 2014). The analysis was a multi-stage process that began after the interviews were complete. First, over 1600 pages of interview transcripts were thoroughly cleansed and de-identified to ensure informant confidentiality. The prepared transcripts, along with field notes and supplemental data, were uploaded to Delve, a qualitative analysis research software. The analysis began with an initial, open coding process to label concepts and phenomena within the data. This was followed by a secondary stage of grouping these initial codes into more defined first-order categories, and then, through axial coding, we generated conceptual focused codes. This iterative refinement allowed for the identification of patterns, the sentiment of the industry, and nuanced themes, which were then organized to build a comprehensive theoretical framework explaining the dynamics of patient matching in HIEs.

The process started with open coding, an essential part of the analysis. This is where we actively engaged with the informant’s perspectives, teasing out the meaning in their experiences (Charmaz 2006) and using gerunds to identify key actions, such as “constantly cleaning our database” and “inconsistent intake hurting data quality”. This provides essential context for the value and problems HIEs must respond to in their programs. Additionally, opening codes such as “feeling AI has industry promise”, “needing AI explainability”, and “lacking AI talent”, started to unearth AI optimism, caution, and hurdles expressed by our HIE experts. The total output for the initial coding process yielded 998 open codes. Further refinement and analysis of the data led to the identification of 108 first-order categories. The first-order categories are vital to grounded theory as they represent the initial conceptualizations derived from the data, allowing for the organization of thoughts and themes that reflect the informants’ lived experiences. For example, first-order categories such as “using normalization practices” and “staffing issues hurting intake” and building off the above open codes, illustrate, respectively, the positive and negative impacts of patient matching activities. Further development of the data yielded focused codes, which are crucial in advancing analysis as they capture the central themes and patterns that emerge. They highlight the relationships between concepts, enabling a deeper understanding of the phenomena under study. For instance, the focused code “exploring AI proactively” is synthesized and reflects the journey from open coding and first-order categories. As an example, “evaluating AI capabilities”, leads to, “testing AI use-cases”, which leads to, “exploring AI proactively”. The interplay between first-order categories and focused codes was essential in the continuation of our grounded theory, but further breakdown was needed to systematically break down the complexities and nuances of the topic. This presented an opportunity to further categorize and create a bridge that developed abstract concepts. We proceeded to analyze the 23 focused codes developed, using an intersectional matrix map, to track towards a grounded theory. This deeper analysis revealed eight aggregate dimensions representing synergies in the coding process. Furthermore, these aggregate dimensions are broken down into three aggregate groupings that tell the story of value, problems, and AI pathways. Figure 1 summarizes the coding breakdown numbers and Figure 2 the emerging data structure.

4. Findings

4.1. Grounded Theory Overview

In this section we will present an overview of our emergent theoretical framework, explaining how our experts perceive the core value propositions, as well as the pressures HIEs face within their patient matching programs. Our analysis reveals two aggregate groupings: Value Striving and Patient Matching Pressures. Value Striving contains the aggregate dimensions Purpose Serving and Programmatic Scaffolding. The first dimension, Purpose Serving, highlights the essential role patient matching plays in HIEs and the work directed towards the core mission of faster treatments, better decision-making, and improved patient outcomes. The second dimension, Programmatic Scaffolding, conveys the solid infrastructure and processes that are built to run an effective patient matching program. The Patient Matching Pressures aggregate grouping encapsulates the myriad internal and external obstacles hindering these efforts; it contains the aggregate dimensions Matching Miscues, Data Besieging, and Programmatic Hindering. Together they convey how matching errors are a result of poor data lifecycle management and program issues.

The third aggregate grouping, and a key takeaway surrounding the story, is AI Forging Pathways, which conceptualizes the perceived role of AI as a potential, albeit carefully considered solution for improving interoperability and addressing long-standing coordination issues. Together, the two aggregate groupings illustrate the continuous tension within the HIE ecosystem, from the foundational need for accurate patient identification to the complex environment in which it operates, culminating in the hesitant embrace of advanced technologies like AI to chart a path forward. Our next section will detail the AI sentiment with definitions, along with supporting interviewee quotes. These aggregate groupings are shown in Figure 3 and Figure 4.

4.2. AI Forging Pathways—A Preview

AI Forging Pathways represents the aggregate grouping that captures the hopeful yet cautious outlook HIE leaders have towards AI’s use for patient matching. Refer to the right-hand side of Figure 3. It conveys an industry acknowledgment of AI and the hype surrounding the innovation. There is a recognition of AI’s potential to transcend current patient matching limitations, accentuating value and alleviating persistent problems. Yet this grouping also underscores the significant hurdles and ethical considerations that temper optimism, framing AI not as a panacea but as a cautiously integrated evolutionary pathway towards effective patient matching solutions. This section describes our findings on how HIE managers are navigating the complex landscape of AI adoption, from recognizing critical deterrents such as lack of explainability and algorithm bias to the realities of resource constraints that can challenge AI integration.

4.2.1. Perceiving AI Imperative

There is growing recognition among HIE experts that AI is an undeniable force that will impact information data exchange. This focused code reflects their understanding of AI’s burgeoning presence and the urgent need to engage with its implications for patient matching, rather than resisting its inevitable integration. This imperative stems from a profound understanding that existing patient matching challenges are too complex and deeply entrenched for traditional methods alone to fully resolve. HIE leaders acknowledge that AI represents a foundational shift, with the potential to fundamentally alter how data is managed, analyzed, and linked across disparate systems. The industry recognizes the profound weight of AI and the strategic necessity of preparing for its widespread impact. This perception is not just about technological advancement; it is also about addressing critical patient safety and operational efficiency gaps that current methods struggle to overcome, positioning AI as a crucial next step in achieving true interoperability. As one HIE executive mentioned,

We can’t deny AI has really a kind of bright future I can see in the HIEs and healthcare sector for sure. It’s coming and we have to prepare, there is a buzz, and conversations are happening more and more.

Perceiving AI Imperative

This focused code reveals that while HIE leaders perceive AI’s imperative, there is a significant knowledge gap regarding its diverse capabilities and specific applications for patient matching. Due to AI’s relative novelty in healthcare, many organizations do not yet fully grasp the distinct nuances of subsets like machine learning (ML), deep learning (DL), or natural language processing (NLP). Consequently, there is an industry-wide recognition that a comprehensive understanding of which AI approaches are best suited for particular patient matching challenges is still nascent. This lack of granular knowledge can hinder effective strategic planning and adoption, as HIE managers struggle to identify the most impactful use cases for their specific operational needs. Experts highlight the critical need for targeted education and collaborative learning to demystify AI’s various methodologies and clarify which capabilities can most effectively address complex patient matching scenarios. One of our interviewees noted,

Not every customer understands what AI can produce for them right now, its early, training, talent, they are not really set up. So, it will require some time for the customers to understand what they require certain prototypes and AI use cases they even have, what AI to use.

Exploring AI Proactively

This focused code captures the awareness within the HIE community that AI technology is no longer theoretical but is actively being tested and evaluated as a pathway forward. HIE leaders recognize AI’s burgeoning influence and are taking deliberate steps to understand its practical applications in patient matching, aiming to get ahead of its inevitable integration. This involves piloting solutions, assessing vendor offerings, and engaging in strategic discussions about implementation feasibility. The current focus is on controlled experimentation to understand AI’s capabilities and limitations within their specific ecosystems, rather than immediate widespread deployment. This active engagement signifies a shift from mere observation to strategic preparation, as HIE managers aim to leverage AI to address long-standing challenges before the technology becomes fully ubiquitous. One HIE executive expressed,

I think there’s tremendous opportunity in that field of AI. We haven’t defined a specific application that we’re pursuing, but we do think about using it for natural language processing and code normalization. It would make sense to utilize some machine learning in the algorithmic matching process. Especially where we have humans looking at records and making decisions. Conversations are started and we are testing.

Improving Precision

This focused code points to the expectation among HIE leaders that AI can significantly enhance the accuracy and reliability of patient matching beyond what current algorithmic methods can achieve. Experts anticipate that AI’s advanced pattern recognition and predictive capabilities will lead to a substantial reduction in both false positives and false negatives, which have plagued patient matching for decades. By sifting through vast, complex datasets with greater sophistication, AI is expected to identify subtle correlations and anomalies that human review or traditional algorithms might miss, thereby minimizing errors. This anticipated improvement in precision is seen as critical for directly enhancing patient safety by ensuring the right data is consistently linked to the right individual. Ultimately, the hope is that AI will drive match rates to unprecedented levels, creating a more trustworthy and efficient foundation for HIEs. As one interviewee noted,

We have been dealing with the last mile problem in patient matching for years, it’s that final percentage of records that just won’t resolve. Our hope with AI is that its predictive power, its ability to learn from millions of data points and spot patterns, will finally help in this regard, more testing to do and trust is a big part of it, but this is how we are thinking to utilize AI.

Hopeful AI/Human Stewarding

This focused code suggests that AI is anticipated to significantly streamline operational workflows related to patient matching, leading to greater efficiency and a substantial reduction in manual burden. HIE systems currently grapple with immense data volumes and the labor-intensive process of resolving near-matches and duplicates. Experts envision AI as a powerful tool to automate repetitive tasks, accelerate data processing, and free up valuable human resources from mundane reconciliation efforts. By intelligently handling the “heavy lifting” of data analysis and preliminary matching, AI could enable human data stewards to focus on more complex cases requiring nuanced judgment. This potential was brought forth by an HIE executive as follows:

We need a collaborative effort of AI and human. The idea of getting all this technology is to increase efficiency and reduce the cost, and have the human do more strategy, enable more cognition as a tool. Let the computers handle the mundane and volume-based tasks, this is evident for manual review, humans cannot keep up, let AI handle the bulk load.

4.2.2. Confronting AI Challenges

This aggregate dimension represents the inherent risks and ethical dilemmas associated with AI itself, which generate caution and potential distrust among HIE leaders. Despite the promising potential, HIE leaders recognize that AI is not without its pitfalls and requires careful navigation to ensure responsible and equitable implementation. These challenges are not merely technical, they extend into profound ethical and societal considerations that demand rigorous attention and mitigation strategies to maintain trust and safeguard patient well-being.

Demanding Explainability

This focused code addresses the critical need for transparency and interpretability in AI decision-making, particularly concerning the “black box” nature of some algorithms. HIE leaders are acutely aware that for AI to be trusted in sensitive applications like patient matching, its conclusions cannot be opaque. They require the ability to understand how an AI system has arrived at a specific match or non-match to ensure accountability, validate outcomes, and troubleshoot errors. This demand for transparency is fundamental to integrating AI responsibly into a system where patient safety and data integrity are paramount, ensuring that AI-driven decisions can be justified and understood by human oversight. An interviewee observed,

When an AI system makes a decision about patient matching, we need to know why, it boils down to a few things, it’s about patient safety and data integrity. If we cannot understand why the AI made a specific match we compromise our ability to build trust with clinicians using the system.

Addressing Algorithmic Bias

This focused code highlights the profound ethical concerns among experts regarding the potential for AI models to perpetuate or amplify health inequities if trained on biased or incomplete data. HIE leaders understand that AI systems learn from the data they are fed, and if that data reflects historical disparities in healthcare access, treatment, or documentation, the AI could inadvertently disadvantage certain patient populations. There is a critical awareness of the need to actively identify and mitigate existing biases within training datasets to ensure fair and equitable matching outcomes for all individuals, regardless of their demographic background. This concern underscores a commitment to ethical AI development, emphasizing that technological advancement must not come at the expense of widening health disparities or reinforcing systemic discrimination. One HIE executive brought forth this concern vividly as follows:

Algorithmic bias is a critical ethical challenge that AI introduces into an HIE network, you learn from your data, and data from one part of the HIE might be demographically different from another part of the state, or better yet nationally. AI is a huge help to the industry, but we need guardrails and trust, we don’t want to inadvertently create a different set of divides within our communities.

Protecting Data Privacy

This focused code reflects pervasive concerns about the security and privacy implications of using AI with sensitive health information. Given the immense volume and granularity of data required to train and operate effective AI patient matching systems, HIE leaders are highly focused on ensuring robust data integrity and patient confidentiality are maintained. This involves grappling with questions around data anonymization, secure data access protocols, and compliance with stringent regulations like HIPAA. The worry extends to potential vulnerabilities that AI systems might introduce, such as new avenues for data breaches or misuse of highly sensitive patient information. Protecting data privacy remains a top priority, driving the need for rigorous security frameworks and clear governance policies around AI’s handling of health records. As an interviewee noted,

AI does introduce a whole new layer of privacy concerns for us in the HIE space to worry about. We are taking precautions, but how do we know others are, it is a scary time, we need regulation around this, we need business agreements to have AI protections.

4.2.3. Realizing Resource Realities

This aggregate dimension captures the practical implementation hurdles and organizational considerations that HIE managers face when integrating AI solutions. It addresses the real-world constraints related to financial investment, workforce adaptation, and the necessary infrastructure required for successful AI deployment. Beyond the technical and ethical challenges, HIEs must contend with the tangible resources and organizational shifts required to transition to an AI-augmented patient matching environment.

Resource Doubting

This focused code pertains to the significant challenges HIE managers anticipate in acquiring adequate financial investment, necessary technological infrastructure, and sufficient datasets for training and deploying AI models. Implementing AI is not inexpensive: it demands substantial upfront capital for software, hardware, and specialized personnel. HIEs often operate under tight budgets, leading to skepticism about their ability to fund these large-scale technology shifts. Furthermore, accessing and preparing the massive, high-quality datasets required to effectively train robust AI algorithms presents a considerable hurdle. This “resource doubting” reflects a pragmatic assessment of the economic and logistical barriers that could impede the seamless adoption of AI, forcing HIE managers to carefully weigh the costs against the promised benefits. One HIE executive brought out this challenge,

So, I think there’s a place for AI, absolutely. But it’s not something that we’ve embraced yet, and I think ultimately our implementation of AI is probably going to depend a lot on budget. There’re limited resources in my HIE. AI would be great, but I have to balance operations, there’s tech costs, talent costs, training to think about.

Lacking AI Competency

This focused code highlights the critical need for developing specialized AI skillsets within the existing workforce and addressing staff concerns related to training, new roles, and adapting to AI-driven processes. AI implementation requires expertise in data science, machine learning engineering, and AI ethics—skills not typically prevalent in traditional HIE operations. There is a clear recognition that simply acquiring AI tools is not enough; the workforce must be upskilled and re-trained to effectively manage, monitor, and leverage these new technologies. This involves developing comprehensive training programs and fostering a culture of continuous learning to bridge the knowledge gap and ensure staff feel equipped, rather than threatened, by AI’s introduction. One HIE executive noted,

The reality is our HIE runs pretty lean and exploring and/or introducing AI into our environment adds an entirely new dimension of talent needs. This is a significant challenge when your operational model is built around constraints or talent deficiencies in the industry, most of us are non-profits. We simply cannot effectively integrate AI or truly leverage its potential within our HIE at this time.

Unclear Role Shifting

Workforce shifts may occur, as broader organizational adaptations might be required, including potential changes to roles, responsibilities, and overall workforce dynamics, especially if AI gathers widespread adoption within patient matching. As AI integrates into aspects of the matching process, particularly matching software, techniques, and manual data stewardship practices, HIE leaders foresee a need to redefine job functions and potentially reallocate personnel to higher-value tasks. This is not just about training; it is about strategically planning for how human-AI collaboration will reshape the daily operations of patient matching teams. Managing these transitions effectively, ensuring staff buy-in, and mitigating fears of job displacement are crucial for a smooth and successful integration of AI into the core functions of an HIE. An executive described this requirement as follows:

What I think if AI comes to fruition isn’t just the AI itself, it’s the change management. How it will change what our people actually do, and will staff be ready for that, it’s a new paradigm for all levels, boards, leaders, staff. What do we let AI handle, what do our people handle, tasks aren’t going to be the same. We’ll need new roles we haven’t even defined yet.

4.3. Value Striving

Value Striving is an aggregate grouping label that contains the aggregate dimensions Purpose Serving and Programmatic Scaffolding. The first dimension, Purpose Serving, highlights the essential role patient matching plays in HIE and the work directed towards its mission. The second dimension, Programmatic Scaffolding, conveys the solid infrastructure and processes built to run an effective program.

4.3.1. Purpose Serving

The aggregate dimension of Purpose Serving underscores the essential role of patient matching in the effectiveness and viability of HIE. It reveals that patient matching is not merely an operational task but a foundational element vital to the very existence of HIE. By facilitating efficient data exchange and enhancing the quality of care, patient matching significantly contributes to the overall value proposition for HIEs. The insights gathered from HIE experts illustrate a collective understanding of how this impacts healthcare delivery, highlighting its importance in achieving successful matching outcomes throughout the ecosystem. HIE Leaders acknowledge their continuous responsibility to ensure accurate patient identification, recognizing it as a non-negotiable component of their mission. This foundational commitment to precise matching positions HIEs as crucial infrastructure, often “swimming upstream” against historical tendencies to silo information yet acting as a “beacon of light” in fostering seamless digital information flow. This was brought out powerfully by an HIE executive,

Some HIEs have always been ahead of their time. They were the first organizations to really think about how to move data at scale and how to build infrastructure to do so. And they’re all oftentimes just swimming upstream against policy and or you know, I’ll call it supply chain alignments in a world where most of the care continuum was to not make it easy for digital information to flow and to share. Some of these HIEs have been a beacon of light.

Essentiality of Matching

The focused code Essentiality of Matching emphasizes that the very existence and core mission of HIEs are rooted in effective patient matching. HIE experts consistently acknowledge that patient matching is not merely a supportive function; it is the cornerstone of HIE operations. They articulate that HIEs fundamentally rely on this matching process to enhance their core services and deliver substantial value to stakeholders. Without effective patient matching, the potential benefits of HIEs cannot be realized, rendering data collection efforts largely meaningless. Experts widely recognize the profound impact patient matching has on overall healthcare delivery and the positive outcomes that arise from successful matching efforts, making it absolutely essential for organizational success and the overarching mission of data interoperability. As one expert put it, “If you could not match, what’s the sense of collecting these records?” Additionally, another noted,

When I came to HIE I realized, because it’s a critical component of how the records are queried, stored, or used to match patients. And it was a huge problem for HIE in general if you couldn’t match. We had a tremendous number of patient records and if you could not match, what’s the sense of collecting these records.

Perpetuating Value

The focused code Perpetuating Value highlights the tangible benefits that consistently arise from effective patient matching within HIE operations. When HIEs invest diligently in patient matching processes, they create crucial efficiencies that provide healthcare practitioners with a comprehensive view of their patients. This “clear pane of glass” enables clinicians to make critical medical diagnoses and informed treatment decisions right at the point of care, ensuring a holistic understanding of a patient’s journey across various providers and settings. Beyond this comprehensive view, accurate matching directly enhances patient safety, as it is paramount to ensuring the right data is consistently used to treat the right patient, thereby preventing potentially catastrophic medical errors. Furthermore, effective patient matching translates into significant cost efficiencies across the entire healthcare ecosystem, including payers, providers, and ultimately patients, by reducing duplicative tests, labs, and imaging. This consistent generation of value underscores patient matching as a continuous driver of improved care and reduced waste. This value was articulated by an HIE executive as,

The value of HIE is this comprehensive view into the member as they’ve moved to multiple different provider types over the years and you know the acutely sick have many visits, right? They’re seeing specialists, they are seeing primary care providers, they are getting admitted into various hospitals and hospital systems. So, the MPI is the is like the one thing that brings all of that together for us. Otherwise, you know, you don’t have the complete picture of a patient.

4.3.2. Programmatic Scaffolding

The second aggregate dimension in this cycle is Programmatic Scaffolding, and it conveys the solid foundation built to run an effective program with careful attention to programmatic actions to improve patient match rates as well as critical thought towards other HIE constituents.

Cultivating Strategic Partnerships

The focused code of Cultivating Strategic Partnerships underscores the essential role of collaboration among stakeholders in enhancing the effectiveness of HIEs. By fostering relationships that lead to improved patient outcomes and shared buy-in, HIEs establish a collective commitment to their goals. An example shared by an informant highlights the leveraging of community expertise and alignment with public health initiatives,

You guys are the people that hold the strings here when it comes to providing healthcare services to people, I need your help to fix this. So, we basically got around the table, enlisting community experts to solve issues and there was a coalition of healthcare stakeholders assembled to do just that.

System Enhancing

The focused code System Enhancing conveys the essential work of establishing a solid foundation for effective patient matching, ensuring reliable technology platforms and high-quality data. This includes meticulous efforts in ensuring system uptime and performance metrics, alongside an obsessive commitment to data management principles such as actively addressing bad data feeds, implementing stringent normalization practices, and establishing robust data governance. This dual focus on technology and data integrity is crucial for underpinning successful patient matching operations.

We’ve seen over the last years more and more focus on governance. So, on data governance and being aware of not just setting standards from a national level, but setting standards from an organizational level of saying, you know, these are minimum data standards we’re adhering to, and we are making the technology work groups actually adhere to them.

Optimized Matching Methodologies

Optimized Matching Methodologies highlights an HIE’s proactive and strategic approach to maximizing patient matching accuracy. This involves a continuous effort to aggressively tune algorithms, recognizing that data quality, not just the algorithm itself, is paramount. Furthermore, HIEs employ diverse “winning formulas” by strategically deploying a combination of deterministic, probabilistic, and referential algorithms based on data quality and exchange goals. This multi-pronged strategy is further augmented by the indispensable practice of human data stewardship, particularly for near-matches, to minimize duplicates and ensure data integrity. One informant noted,

But the approach what we have taken most is we have taken more like the hybrid approach, and we can combine algorithms, and we can do reviews, we want to push up scores, you know the most.

4.4. Patient Matching Pressures

4.4.1. Matching Miscues

The aggregate dimension Matching Miscues contains the focused codes Matching Misses and Merging Dangerously. These represent the false negative and false positive outcomes, respectively, that occur within the patient matching landscape. The perspectives from our HIE experts affirm these have detrimental effects on all players of the HIE ecosystem. Treatment delays, medical errors, and cost inefficiencies are the result of missed opportunities, and in the case of false positives, pose a significant patient safety concern and can cause serious injury. The expert commented,

We haven’t done it right in most cases, and when we can’t match a patient to their records or we make errors, we haven’t helped the patient, we haven’t equipped the doctor with the tools to get at that data in a way that gives them clarity.

Matching Misses

The focused code Matching Misses represents the missed opportunities in HIE patient matching or the false negatives. This means that you had the data to match the patient in your system, but you did not make the match. This can happen due to a number of programmatic, technical, process, or legal issues, but the same outcome is obtained: a suboptimal record that forces medical practitioners to reconstruct a patient’s history from scratch in order to provide care. As one interviewee observed,

The expense of repeated medical care due to duplicate records costs about a thousand $950 per patient per inpatient stay and then over $1700 per emergency department visit. We know that 35% of all rejected claims result from inaccurate patient identification, which costs the average hospital 2.5 million, and the US [healthcare] system over 6.7 billion annually.

Merging Dangerously

Merging Dangerously is the focused code in this aggregate dimension. It represents the false positive nature of matching. It occurs when you have two patient records that look similar but are not, and the system or a person performing a manual review merges the records incorrectly. This can be catastrophic and is considered a significant patient safety issue, and it is a major concern towards patient privacy issues. An HIE executive shared that,

If it’s not connected or it’s improperly overlaid or merged, it’s just, I think, a patient safety issue and with the exchange just growing and growing. You can’t control it after it’s gone on. So, if you’re using someone’s information and they’re not the right person or it’s not the right person’s information.

4.4.2. Data Besieging

Data Besieging is an aggregate dimension that conveys the very real issue and sheer raucous nature of HIE data. Data is the raw material, a resource of HIEs. There is negative sentiment from industry experts on how it is handled and accounted for in an ecosystem. The downstream impacts an HIE’s data can cause through the industry are a point of frustration, and due to the growth, volume, and fragmented nature, also exacerbate quality concerns. Together these focused codes spell out the trials and tribulations HIEs must contend with and how this can upset match rates within their program. One informant mentioned that,

I’m challenged by, you know, examples where organizations were, you know, told or informed about bad matches, false positives right and chose not to do anything about it. I’ve always struggled in that kind of a dynamic because of the impacts across, it hurts everyone. I would rather the outcome be correct than you know, sort of a head in the sand approach.

Chaotic Patient Intaking

When a patient presents at the facility and registers, this is patient intake. It is the “quasi” starting point of the patient matching process. The data received and entered into a patient record will either follow them or not, depending on how the patients’ information matches up, system to system. The lack of standardization at this critical juncture of the process poses major concerns for industry experts. To add to the chaos, HIE experts have weighed in on the further complexities to the patient intake equation with what they call “edge cases.” These are situations that even with the best standards in place are susceptible to error. One informant mentioned that,

Everybody has their own set of intake protocols, right? So, registration for, you know, for me at Hospital A has 15 different fields that I have to answer, you know, patient questionnaire, but if I go to Hospital B, maybe it’s 25, so there is no consistency from the start. This complicates things.

Proliferating Data

The HIE industry is currently experiencing substantial data growth, driven by the increasing popularity and adoption of HIEs among hospitals and healthcare providers. As more facilities participate in HIEs, the volume of shared data continues to rise, reflecting a growing demand for seamless information exchange. While this growth is beneficial for improving healthcare delivery, it also presents challenges in managing and integrating the data effectively, often exacerbating issues related to data fragmentation. The continuous flow of information often occurs without oversight or normalization, complicating efforts to maintain data quality and accuracy in patient matching. Specific events and new technologies further support this growth, such as the COVID-19 pandemic and the rise of telehealth services, adding complexity due to varied rules and rapid implementations. As patient mobility increases and the landscape of healthcare evolves, HIEs must develop effective strategies to manage and utilize this expanding data environment. An example of this was provided by an HIE executive,

You know you could have person X, being a record, B, record, C, record, E, record. But you have to be careful about building a bridge across these sources, like you could have accidentally created a new person. That is an agglomeration now, it is way bigger now. It’s not just matching two records from A and B source. You could be matching against 20 records, 30 sources. So, the false positive avoidance is pretty important.

Undermining Data Quality

The HIE industry faces significant challenges related to data quality, driven by the reliance on data from others throughout the ecosystem. When adding more nodes to a system, the overall complexity increases. The reality of “garbage in, garbage out” underscores the impact of inadequate data on the Master Patient Index (MPI). The expectation of bad data is common across the industry, as no organization maintains flawless records. Smaller practices, in particular, can exacerbate these issues, undermining the collective health of HIEs. As patient demographics change and data is incomplete, organizations must navigate the complexities of incoming data, which can introduce inaccuracies and complicate patient identification efforts. Widespread inconsistencies in MPI statistics raise concerns about the accuracy of patient matching processes. Organizations often face challenges in reconciling unmatched records and aligning their MPIs with actual population sizes. Cleaning up MPIs is both time-consuming and costly, yet it is essential for maintaining accurate patient identification. This quote by an interviewee offers a vivid instance of these challenges,

Some HIEs just absorb as much data as they can and some HIEs will actually put a little bit of governance upfront and be like hey I’m not going to take a sin record Right? If it’s not, if it’s a name and date of birth. That’s not enough. If you have a policy where you can enforce data quality standards at the source, that’s helpful for an HIE. But the fact remains we don’t see that and when someone lets bad data through, and we’ve seen it, it’s out there, and it’s not getting changed.

4.4.3. Programmatic Hindering

The Programmatically Laboring aggregate dimension refers to the challenges faced by HIE managers when their technology and operational processes are inadequate or outdated. This encompasses a range of issues that hinder effective patient matching. Additionally, there is Underemphasized Data Stewardship, which involves the crucial manual review of data. When HIEs rely on outdated or ineffective systems, it directly impacts their ability to match patients accurately. Inefficient operations and a lack of enforcement of processes further exacerbate these challenges, leading to increased errors and inconsistencies in patient matching. As a result, the shortcomings in the HIE’s programmatic approach can undermine the overall effectiveness of data management and patient care, highlighting the critical need for modernization and robust operational practices within these systems. An informant noted,

Just the amount of infrastructure and people you need to be able to support it. It is a lot, right? So that’s been kind of like. And it’s an activity that really needs to be enterprise wide and it’s not, it misses in many ways, from tech to process, and we’ve seen this in the HIEs we support in the state, there is varying emphasis on technology programs.

Faltering Systems

The Faltering Systems focused code captures another critical set of challenges within Health Information Exchanges (HIEs) that significantly impact patient matching and interoperability, stemming from both aging infrastructure and fundamental architectural design flaws. Technical dispersion across different systems can lead to inconsistencies and complications in cross-querying, ultimately increasing complexity and hindering effective data integration. When HIEs are queried for patient information, the performance of their systems becomes paramount; if these systems are not up and available, or if their underlying architecture was not built for modern interoperability needs, accurate data exchange is severely compromised. Daily operational issues stall technology performance, creating a cascade of limitations that affect matching applications. Many HIEs have agreements that stipulate the timely return of data, governed by Service Level Agreements (SLAs). Failure to meet these agreed-upon timelines is considered a technical failure and can contribute to false negatives in patient matching. This situation poses a significant risk: having the data available for exchange but being unable to utilize it due to technological limitations. This pain was evident in a quote from an HIE executive interview.

But if you can’t get good connectivity, if you can’t connect and others are down, you can’t get the data you need, you can’t communicate reliably and can’t find the person or if it is the same person from place to place.

Lagging Program Operations

Lagging Program Operations highlights significant shortcomings in the operational practices of HIEs that hinder their effectiveness. Many HIEs face operational misses, where there is a lack of emphasis on continually improving processes and algorithms. This includes failing to tune algorithms adequately, which contributes to ongoing quality issues in data management and patient matching. Additionally, the absence of proper onboarding of data feeds into the program impacts operations. Many HIEs do not prioritize the care and maintenance of their own data, which undermines their ability to operationalize data improvements effectively. Feedback loops are also an important aspect of program operations, especially when part of a network that shares data. The slow adoption of recent innovations in patient matching techniques also leaves HIEs at a disadvantage, as they often rely on outdated methods. Budget constraints are frequently cited as a reason for not investing in technological improvements, stifling innovation and perpetuating existing problems. One HIE executive said,

I had been given the task to maintain the MPI when I came on 4 years ago, it had been started, I want to say 2017 … and never looked at. So, it hadn’t been tuned, it hadn’t been, you know, nothing was identified like nothing was working so what purpose was it serving?

Underemphasizing Data Stewardship

Underemphasized Data Stewardship highlights how a critical component of winning matching strategies becomes a problematic area when neglected, leading to missed opportunities and pollution of an MPI. No algorithms are 100% accurate, and due to the myriad of issues discussed, human review at some point in the matching process is an operational essential in order to remove duplicate records from the system. When close possible matches lack a process for flagging and reviewing, and when factoring in hundreds of thousands of transactions a day, opportunities to shrink the MPI are missed. Reasons for this neglect range from overlooking its importance, over-relying on technology, an inability to keep up with volume, or lacking funding for data stewardship. As noted by an informant,

We get 15 million messages a month on average. So that’s a whole bunch of maybes that we’re still we can’t handle it like. There’s we cannot. There’s “no I could hire full time person to be their dedicated job just to do the human intervention, and that person wouldn’t be able to complete them all in a day, month, year.”

4.5. Grounded Theory Summary

A diagrammatic summary of the above, emphasizing the “AI Forging Pathways” section on the right side, is shown in Figure 3. This figure articulates what we conceive as a processual framework to depict “Cautious AI Optimism,” the central emergent theory describing HIE experts’ perceptions on AI solutions to possibly aid in patient matching programs. The theory’s foundation rests on two affirming aggregate groupings. First, Value Striving depicts HIE managers’ fundamental commitment to patient matching, driven by their core mission and sustained by positive programmatic actions. Second, Patient Matching Pressures shows the significant obstacles that stem from poor data handling practices and operational shortcomings, which collectively lead to detrimental matching outcomes. These inherent tensions and the striving for value necessitate “Critical Thought Into AI Pathways.” Thus, the diagram’s centerpiece, AI Forging Pathways, illustrates how HIE experts cautiously explore the potential of AI to both accentuate value and directly respond to matching pressures. This pathway details a progression from acknowledging the importance of taking this new innovation seriously with the aggregate dimension of Perceiving AI Imperative, grasping its capabilities through proactive exploration, with the ultimate hope of improving precision and enabling effective AI and human stewarding.

However, the optimism surrounding this imperative is consistently tempered by two critical sets of challenges. Confronting AI challenges highlights concerns about the demand for explainability, with many experts insisting that “we need to understand how AI makes its decisions.” Addressing algorithmic bias to ensure healthcare equity and thinking through how patient data will be protected through AI channels is a top priority for HIE experts. Furthermore, Realizing Resource Realities underscores practical implementation hurdles, including resource doubting (financial, technological, data constraints), lacking AI competency (talent gaps), and the complexities of addressing workforce shifts (redefining roles). These challenges serve as crucial intervening conditions, shaping and ultimately moderating the pace and nature of AI adoption, leading to the observed cautious optimism. The diagram thus conveys a strategic, yet circumspect, journey toward integrating AI into the core functions of HIE systems.

5. Discussion

A key, rather surprising, discovery of this study is that while the interviewed HIE executives thought about incorporating AI into their HIE systems, and some even started testing it, none of them had actually started using AI in their HIEs. The interviews reveal some of the reasons leading to that caution. The grounded theory section combines those insights into a model, shown in Figure 3. The model presents the AI Forging Pathways, comprising a perceived AI imperative but also challenges, and how those relate to patient matching pressures and values. What is also interesting in that model is that there was no mention of trust or distrust in the AI or the companies implementing it. That is surprising, because in other contexts trust is an important antecedent of AI adoption (Benk et al. 2024), including in medical settings (Prakash and Das 2021), as well as how it is used (Gefen et al. 2025). It is possible that because of the strong medical regulatory context of HIE and the oversight it entails that trust, as in the case of other IT where regulations are strong and enforced (e.g., Gefen and Pavlou (2012)), is of secondary importance considering the power of regulations and oversight to guarantee outcomes. More research on that is needed.

Our interviews reveal a prevailing sentiment of cautious optimism among the HIE experts we interviewed. There is an acknowledgment that AI, particularly machine learning, could transcend the limitations of current algorithms by identifying complex data patterns to improve matching accuracy (Borna et al. 2023). However, this optimism is tempered by significant and valid concerns that prevent widespread and acceptable adoption through the HIE ecosystem.

One barrier to embracing AI is the risk associated with its current state, perhaps indirectly implicating the matter of trust. Experts rightly pause due to the “black box” nature of some algorithms in general, further exacerbated by sentiments that there is also a lack of transparency HIE to HIE on how the other matches. As one of our interviewees put it,

I’d say, it might be a little bit of a black box, so we don’t know exactly what they’re doing. So, I think that’s kind of always a big open question for me of what and how the algorithms are doing it.

This challenges the trust and explainability required in a clinical setting where errors have severe consequences (Holzinger et al. 2019). As another informant stated,

If I’m going down an AI path, I need the AI to be transparent enough, that when we do quality control, we understand why it did what it did. Where I know what the calculation is, judge, the outcomes so that when I’m looking at an identity later I can systematically explain why that match was made.

This need for transparency is compounded by the profound ethical risk of amplifying health inequities through biased training data (Wiens et al. 2019). This could lead to certain patient populations being disproportionately affected by matching errors. The responsible development and implementation of AI in healthcare also raises the need for a robust framework for fairness, accountability, and transparency to mitigate these risks (Grote and Berens 2020). As one of our interviewees put it,

We worry about this whole algorithmic bias issue of just because your model performs really well on the data you’ve trained it on does not mean it will perform really well on everybody’s data across every healthcare system across everything, right? That’s where they tend to make their mistakes, in the extreme cases on data you haven’t seen.

Finally, implementation hurdles also contribute to the cautious stance of HIE leaders. As our research indicates, HIEs often operate with tight budgets and limited resources. The potential cost of developing or acquiring sophisticated AI solutions, the need for specialized data science talent, and the challenge of integrating these new technologies with legacy EHR and HIE systems are significant practical barriers that also cannot be overlooked.

The above analysis lands us on the horns of a dilemma. On the one hand, the criticality of effective patient matching in the effective function of HIE systems is self-evident. On the other, the continuing prevalence of barriers and hindrances in ensuring robust and error-free patient matching processes beckons a need to explore the potentialities of technological innovation, especially AI, in overcoming the challenges. AI, however, in itself is a not a foolproof technology and is still emerging.

Several other insights emerge from our research, providing a comprehensive understanding of the current HIE patient matching landscape and the prospects of AI. Our findings reveal that Value Striving, inherent in HIE operations, is not merely an aspiration but a foundational aspect that drives program actions. HIE leaders consistently emphasized that accurate patient matching is essential to HIE existence, a prerequisite (Godlove and Ball 2015) and mission fulfiller, that directly contributes to timely diagnoses, more informed treatment decisions, and fewer medication errors (Grannis et al. 2019). This foundational commitment drives HIE managers in their efforts to perpetuate value through comprehensive patient views, enhanced safety, and cost efficiencies (Menachemi et al. 2018). Furthermore, the dedication to building great programs, Programmatic Scaffolding, through cultivating strategic partnerships, actively enhancing systems and data, and optimizing matching methodologies, highlights a proactive stance. This aligns with existing literature emphasizing the importance of robust infrastructure and collaborative efforts for effective information exchange (Heath et al. 2017; Torres et al. 2014).

HIEs are also subject to considerable technical and programmatic pressures, which impede their efforts and contribute to the continuous tension described in our theory. Matching Miscues encompasses both false negative and false positive outcomes and possess severe patient safety (Greer 2020) and privacy risks (Shen et al. 2019) and are direct outcomes of these pressures. Our findings also show that data quality is a pervasive issue (Just et al. 2016), stemming from the journey that data takes from provider intake and the multiple repositories and data management practices it is subject to. This aligns with research highlighting the detrimental impact of data capture errors and fragmentation on patient identification (Hulsen 2020; Rahurkar et al. 2015). Moreover, from a programmatic perspective, downed IT systems, faulty architectures, and lagging operations further exacerbate these challenges. The sheer volume and Sisyphean nature of data stewardship, as expressed by one of the informants, resonate with established concerns regarding the labor-intensive demands of maintaining data integrity (Hripcsak et al. 2013).

Our data also indicates a strong optimism towards AI initiatives, and elevates the understanding that AI is an undeniable technology we must look at and that its use in HIEs could be a transformative force to aid patient matching programs. This recognition drives proactive engagement, as HIE managers and leaders are learning and exploring AI through testing and piloting solutions, signifying a shift from observation to possible preparation, and from gap to deployment (Fichman and Kemerer 1999). The anticipated benefits are significant, and the hope is that AI can improve precision of match rates, particularly addressing the “last mile problem” in matching, where edge cases are almost always put into a backlog and forever exist as orphaned records. AI also has experts postulating that it can streamline mundane tasks, enabling humans to focus on higher-value activities. This aligns with broader industry hopes for AI to enhance efficiency and decision-making in healthcare (Borna et al. 2023).

However, the optimism within AI Forging Pathways is tempered by two key sets of challenges. First, we need to confront AI challenges that highlight profound ethical and technical considerations (Sonko et al. 2024). The demand for explainability for “black box” algorithms is paramount for trust and accountability in clinical settings (Holzinger et al. 2019). Equally important is addressing algorithmic bias to prevent health inequities (Wiens et al. 2019). Furthermore, protection of patient records amidst the vast data requirements of AI systems (Grote and Berens 2020) is of paramount importance. These concerns suggest a need for responsible engagement with inherent AI risks.

Another challenge unearthed is the realization by HIE experts that resource acquisition and availability are important if AI solutions are to play a role in patient matching. Resource doubts reflect significant concerns over the substantial financial investment and technological infrastructure (Sandeep et al. 2025) and adequate datasets required for AI deployment, especially for non-profit HIEs operating with tight budgets. This economic constraint often necessitates reliance on vendor solutions, as one informant stated. Furthermore, the lack of AI talent within the existing workforce and the complexities of what roles will be needed and how work will shift all weigh heavily on experts. The need for specialized AI skillsets and the redefinition of job functions is a critical barrier to widespread adoption, indicating that technological advancement must be coupled with human capital development. Compared to earlier technology implementations of AI, the unique demands for transparency, bias mitigation, and significant workforce restructuring present a complex adoption landscape for the HIE industry.

6. Limitations

The study analyzed the interview transcripts of 27 senior HIE executives, combined with the notes the lead author took during those interviews. The number of interviewees might seem small in comparison with survey research; however, it is sufficient in qualitative interview research according to Charmaz (2014) whose methodology we applied, especially when the themes being raised by the interviewees converge and stop revealing new themes. There are also only so many senior HIE executives that could be interviewed, and, knowing their busy schedules, that we managed to interview 27 of those is rather outstanding. Nonetheless, we acknowledge the value of interviewing more senior HIE executives, especially outside the U.S., where regulations allow for a better integration of HIE systems across providers. We initially considered running a comparative analysis to learn what led some HIE managers to adopt AI and others to wait. Being able to make such comparisons could have shed light on the risks and values of adding AI to HIE. However, as it turned out, all the interviewees told us that their organization had not yet implemented AI. They were all thinking, talking, and testing, but had not yet transitioned to deployment of AI-based tools. Future studies can track the deployment processes that will inevitably occur as AI-based technologies mature and use cases become more prevalent.

7. Conclusions

Our research brought to light the critical dynamics influencing patient matching and AI adoption in HIE systems: the enduring commitment to Value Striving, the pervasive Patient Matching Pressures, and the nuanced AI Forging Pathways. The significance of acknowledging the continuous tension between their essential mission and existing operational hurdles is paramount, as this tension precisely fuels a willingness to test the untested—the hesitant embrace of AI. Although AI holds a tentatively promising future for enhancing patient matching by improving precision and optimizing workflows, its integration is critically moderated by significant ethical, technical, and resource-related challenges.

Our study highlights that AI is not perceived as a simple “fix-all” but as a powerful yet demanding tool requiring careful consideration of its implications for explainability, bias, privacy, budget, talent, and workforce adaptation. Understanding “Cautious AI Optimism” is crucial for the HIE ecosystem, and it can be used to strategically guide responsible and effective integration of AI into vital patient matching processes, ultimately contributing to improved patient safety and healthcare interoperability.