Healthcare has spent the better part of a decade building data lakes. The original problem was straightforward and worth solving: clinical data lived in silos, providers could not see the full picture of a patient’s care, and aggregation across the ecosystem was a prerequisite for nearly everything that came next.
That foundational work has now largely succeeded. Health information exchanges, private interoperability platforms, and the expanding Qualified Health Information Network framework have made it possible to move enormous volumes of clinical data between organizations. In roughly a year, the national framework supporting this exchange has grown from 10 million records to nearly 500 million health records exchanged.
Industry attention is shifting from collection to activation. Much of the most valuable clinical detail lives in formats a data lake struggles to reach: physician notes, imaging reports, and discharge summaries. A 2025 study in the Journal of Medical Internet Research that analyzed 1.8 million primary care records found that only 13% of the clinical concepts captured in free-text notes had matching counterparts in the structured data of those same records. Even structured records have meaningful gaps. For example, a 2022 study in the Journal of the American Medical Informatics Association found that only 59.4% of chronic conditions were consistently captured across encounter diagnoses and problem lists within a network of more than 500 community health centers. Activation of data, rather than aggregation alone, is the new differentiator.
Yet many of these data repositories have quietly turned into what could be accurately described as “data swamps.” They are vast, deep, and nearly impossible to navigate without the right clinical lens applied on top. The real challenge facing health data platform companies and health system data teams is what to do with all of it.
A fragmentation problem
Industry conversations often default to framing this as a data quality problem. That framing understates the issue. Quality certainly matters. Source records routinely contain text that does not match accompanying codes, mappings that point to the wrong concepts, retired ICD-9 codes that were never updated, and placeholder values such as 9999 that masquerade as real clinical data. Many physician practices and smaller health systems lack the informaticists necessary to maintain a clean data master at the source.
The deeper problem is fragmentation. Even when individual data points are accurate, they often arrive disconnected from the clinical context that gives them meaning. A laboratory result without its associated problem, a medication without its indication, or a diagnosis without its supporting evidence cannot answer the questions clinicians actually ask. Data quality work alone produces tidier data, which falls short of the ultimate goal of creating clinical understanding.
Three capabilities missing from most data lakes
Three capabilities consistently separate platforms that activate clinical data from those that merely store it.
The first is data extraction. A significant share of clinically relevant information lives in PDFs, scanned documents, free-text notes, and discharge summaries. Without natural language processing (NLP) tools that can convert that narrative content into reliable, coded, structured data, the most clinically rich material in any patient record remains inaccessible to downstream analytics, reporting, and AI systems.
The second is a clinical lens. Clinicians do not think in data tables. Rather, they think by problem, such as the status of any given patient’s heart failure, diabetes, or recent surgical recovery. Activating a data lake requires the ability to filter, organize, and present information by problem, by specialty, and by relevance to the decision at hand. That capability depends on a curated clinical knowledge graph that understands the relationships between diagnoses, symptoms, tests, treatments, and outcomes. Without such a layer, a longitudinal patient summary remains an undifferentiated wall of records.
The third is scrubbing. Tools that resolve duplicates, reconcile conflicting records, normalize terminology, replace invalid or retired codes, and validate diagnoses against supporting evidence allow the underlying data foundation to be trusted. Without that layer, every downstream model, dashboard, or AI agent is operating on shifting ground.
An activated data lake
Picture a clinician at the point of care opening a portal from inside the electronic health record. The portal displays everything known about the patient from external sources, organized longitudinally and filtered by the problem the clinician is treating that day. The clinician can drill down on a single specialty, screen for findings the chart may have missed, and selectively import only the data relevant to today’s encounter. The patient’s diabetes management is in clear view. The unrelated ankle sprain being treated by another provider stays in the background.
While simple on the surface to the clinician user, that experience requires several layers underneath working together: a normalized clinical data foundation, NLP that converts narratives into structured concepts, mappings to standard terminologies and interoperability formats including SNOMED CT, LOINC, RxNorm, FHIR, and C-CDA, and a clinical knowledge graph that organizes concepts by clinical meaning. Each layer reinforces the others. Likewise, pulling any one of them out collapses the workflow.
From aggregation to activation
Building toward that kind of architecture is what separates a data lake from a data swamp. Health data platforms and health systems that invested heavily in aggregation are right to be proud of the foundation they laid. The next phase of competitive differentiation will favor those that turn accumulated data into clinical intelligence. That requires treating clinical context as a core architectural concern from the foundation up. Storage was the easy part. Meaning is the hard part, and meaning is what clinicians, patients, and AI systems all depend on.
Health systems and platform vendors that take that step will quickly discover what their data was meant to do.
Photo: Weiquan Lin, Getty Images
David Lareau is Chief Executive Officer of Medicomp. Lareau joined Medicomp in 1995 and has responsibility for operations and product management, including customer relations and marketing. Prior to joining Medicomp, Lareau founded a company that installed management communication networks in large enterprises such as The World Bank, DuPont and Sinai Hospital in Baltimore. The Sinai Hospital project, one of the first PC-based LAN systems using email and groupware, was widely acknowledged as one of the largest and most successful implementations of this technology.
Lareau’s work at Sinai led to the founding of a medical billing company that led, in turn, to his partnership with Medicomp. Realizing that the healthcare industry made less use of information technology than almost any other industry, particularly in the area of clinical care, Lareau immediately saw the potential for Medicomp’s powerful technologies and joined the company to help fulfill Peter Goltra’s vision.
This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.
