Fragmented by Design: The Electronic Health Record Problem That Limits What AI Can Know About You

Imagine presenting to a cardiologist with a new complaint of intermittent chest discomfort. The cardiologist's institution uses Epic. Your previous cardiac workup, conducted three years ago at a different health system, lives in a Cerner database. The echocardiogram you had at an independent imaging center is stored in a proprietary PACS system. Your primary care physician, who retired last year, kept records in a now-discontinued EHR platform. And the AI-powered diagnostic tool your cardiologist's hospital recently deployed? It can see only what lives within its own system.

This is not a hypothetical. It is the daily reality of healthcare in the United States—and it represents one of the most significant structural barriers to the realization of AI-assisted medicine's potential.

At StethyAI, we have a particular stake in this conversation. The accuracy of any AI diagnostic or health insight platform is directly proportional to the completeness of the data it can access. A tool operating on a partial patient record is not merely less effective—it can be actively misleading, generating recommendations calibrated to an incomplete clinical picture.

How America Ended Up with a Health Data Archipelago

The fragmentation of American health records is not accidental. It is the product of decades of decentralized technology adoption, competitive market dynamics, and regulatory frameworks that prioritized individual privacy before the concept of interoperability was fully understood.

The HITECH Act of 2009 provided financial incentives for hospitals and physician practices to adopt certified electronic health record systems, accelerating a transition away from paper records. By the early 2010s, EHR adoption had increased dramatically. What the legislation did not adequately address was how these newly digitized records would communicate with one another.

Vendors—Epic, Cerner (now Oracle Health), Meditech, Allscripts, and dozens of smaller players—built systems optimized for their own institutional clients. Data sharing with competing platforms was not a design priority; in some cases, it was actively disincentivized. Health systems that had invested heavily in a particular vendor's ecosystem had limited commercial motivation to make patient data easily portable to competitors.

The result is what health informaticists describe as a data archipelago: islands of rich clinical information separated by technical and institutional barriers that are difficult and expensive to bridge.

What Your AI Health Tool Cannot See

For patients using AI-powered health platforms—whether through their hospital's patient portal, a standalone wellness application, or a telemedicine service—the practical consequences of this fragmentation are significant.

Consider a patient managing a complex cardiac condition who has received care at multiple institutions over fifteen years. Her current health system's AI platform may have access to her most recent labs, her current medication list, and her last three cardiology notes. It almost certainly does not have access to the stress test she had at a different hospital in 2014, the medication trial that was discontinued due to side effects in 2017, or the family history documented by a physician who has since retired.

An AI tool operating on this truncated record will generate insights calibrated to an incomplete version of this patient. It may fail to flag a drug interaction because it cannot see a medication prescribed by an out-of-network specialist. It may recommend a diagnostic test she has already had. It may miss a pattern in her longitudinal data that would be apparent if the full timeline were visible.

These are not edge cases. A 2021 study published in Health Affairs estimated that the average Medicare beneficiary sees seven different physicians annually. For patients managing chronic conditions, the number of institutions holding relevant records is frequently higher still.

The Interoperability Standards Landscape

Regulatory efforts to address data fragmentation have accelerated in recent years, with mixed results.

The 21st Century Cures Act, signed into law in 2016 and implemented through rules finalized in 2020 and 2021, introduced significant new requirements around data sharing. The legislation explicitly prohibited information blocking—practices by health systems or vendors that impede the access, exchange, or use of electronic health information—and mandated support for standardized application programming interfaces (APIs) based on the FHIR (Fast Healthcare Interoperability Resources) standard.

FHIR, developed under the auspices of HL7 International, represents the most promising technical foundation for interoperability currently available. It defines a standardized way of representing health data that allows different systems to exchange information in a common format. Major EHR vendors have implemented FHIR-compliant APIs, and a growing ecosystem of third-party applications—including several AI-powered health tools—is built on FHIR-based data access.

Progress, however, has been uneven. Implementation quality varies substantially across institutions and vendors. Smaller hospitals and independent physician practices often lack the technical resources to implement interoperability standards fully. And FHIR, while a significant advance, does not resolve the underlying question of whether institutions will share data in practice, even when they are technically capable of doing so.

Privacy Regulations: Protection and Unintended Consequence

HIPAA—the Health Insurance Portability and Accountability Act—was landmark legislation when it was enacted in 1996. Its privacy protections for individually identifiable health information have genuine and important value. They also, in certain configurations, create barriers to the data sharing that AI-powered diagnostics require.

The complexity of HIPAA compliance has made many health systems risk-averse about data sharing, even in contexts where sharing would be legally permissible. Institutional legal teams, wary of enforcement actions, frequently interpret the regulations conservatively. The result is a culture of data restriction that exceeds what the law actually requires.

More recent regulatory developments have attempted to clarify that patient-directed data sharing—instances where patients themselves request that their records be shared with a third-party application—is generally permissible under HIPAA. The practical implementation of this principle, however, remains inconsistent across health systems.

Emerging Solutions Worth Watching

Several technical and policy developments offer genuine grounds for optimism.

The CommonWell Health Alliance and Carequality network represent industry-led efforts to create national interoperability frameworks connecting thousands of healthcare organizations. These networks have facilitated billions of record exchanges and continue to expand their reach.

The TEFCA (Trusted Exchange Framework and Common Agreement), established under the 21st Century Cures Act, provides a governance framework for nationwide health information exchange. The Recognized Coordinating Entity overseeing TEFCA implementation is actively working to onboard qualified health information networks, and the initiative has the potential to significantly expand the scope of accessible patient data.

Patient-held health records, enabled by platforms like Apple Health Records and Android's equivalent, allow individuals to aggregate data from multiple institutions onto their personal devices. While adoption remains limited, these tools represent an important patient empowerment mechanism that could complement institutional interoperability efforts.

AI-assisted data normalization is itself emerging as a partial technical solution. Machine learning models trained to reconcile inconsistent terminology, coding standards, and data formats across disparate systems can improve the quality of aggregated records even when full standardization has not been achieved at the source.

What Patients Can Do Now

While systemic solutions develop, patients can take practical steps to improve the completeness of their own health data.

Request copies of your records from every institution where you have received care. Many health systems now provide access through patient portals compliant with FHIR APIs, and you have a legal right to these records under HIPAA. Consider using a personal health record platform to consolidate this information. When you engage with a new provider or AI health tool, proactively share relevant records from other institutions rather than assuming the system can access them.

Perhaps most importantly, understand the limitations of any AI health platform you use. A tool that has access only to data from a single health system is not providing a comprehensive picture—and its recommendations should be interpreted accordingly.

The promise of AI-powered medicine—truly personalized, longitudinally informed, comprehensive health insights—depends on solving the data fragmentation problem. At StethyAI, we are committed to transparency about where that promise currently falls short, and to advocating for the technical and policy progress needed to close the gap.