The Hospital That Couldn't Answer a Simple Question

A hospitalist at a large academic medical center is treating a patient admitted through the emergency department. She needs to know three things: the patient's current medication list, their allergy history, and any prior hospitalizations in the past two years. The patient is unconscious and cannot provide this information.

The answers exist. They are distributed across an Epic EHR at one health system, a legacy Cerner installation at a second system the patient visited for a specialist, a pharmacy benefit management database, a state immunization registry, and paper records that were faxed from a primary care practice that has not yet adopted electronic records. None of these systems talk to each other. The hospitalist makes her medication decision with incomplete information. It is a decision made thousands of times per day at hospitals across the country.

Why Healthcare Data Is the Hardest Master Data Management Problem in Any Industry

Healthcare data combines every complexity dimension that makes enterprise data difficult — at higher stakes than any other domain. The data is fragmented across thousands of independent systems operated by organizations with no contractual obligation to share. The entities involved — patients, providers, facilities, medications, diagnoses — are represented by different identifiers in every system, with no universal ID. The data contains both structured fields and unstructured clinical notes that carry critical information in formats that resist automated processing. And the regulatory environment governing data sharing is a patchwork of federal law, state law, contractual restrictions, and institutional policy that makes even technically feasible integrations legally complex.

This is not a technology failure. Every individual system in the healthcare data landscape is technically capable of producing interoperable data. The failure is architectural: there is no agreed-upon Bronze layer for the healthcare industry — no universal raw data standard that every system writes to before any transformation is applied. And without a Bronze layer, there is no foundation for a Silver layer, and therefore no foundation for AI.

The Five Healthcare Data Problems That Mirror Every Enterprise's Legacy Challenge

• Patient identity fragmentation — the same person exists as different records in different systems, linked only by probabilistic matching algorithms that are accurate enough for billing and dangerous enough for clinical care
• Medication reconciliation gaps — medication lists that are updated in one system are not propagated to others, creating divergent records of what a patient is actually taking
• Diagnosis code inconsistency — the same clinical condition is coded differently depending on which system, which coder, and which year, making longitudinal analysis unreliable
• Unstructured clinical note dependency — critical clinical information lives in free-text physician notes that require natural language processing to extract and that vary dramatically in structure and terminology across providers
• Temporal data quality degradation — data that was accurately recorded at a point of care becomes progressively less accurate as time passes and the patient's condition changes, with no mechanism to flag stale records

What the Silver Layer Looks Like in Healthcare

The healthcare systems that have made the most progress on AI — the Mayo Clinics, the Kaiser Permanentes, the Intermountain Health systems — have all built, in different forms, a Silver layer that addresses these problems within their own enterprise boundaries. They cannot yet solve the cross-organizational fragmentation problem at scale, but they have solved it within their own network.

The core Silver layer transformation in healthcare is Master Patient Index construction: the process of resolving all representations of the same patient across all internal systems into a single, canonical patient record with a unique enterprise identifier. This transformation requires probabilistic matching algorithms, human review workflows for low-confidence matches, and continuous reconciliation as new records are created. It is expensive and time-consuming. It is also the prerequisite for every meaningful clinical AI application.

The Enterprise Lesson: Every Industry Has a Healthcare Problem

The healthcare data challenge is extreme, but it is not unique. Every organization that has grown through acquisition, operated multiple business units with independent technology stacks, or accumulated data across decades of system changes has a version of this problem.

The entities are different — customers instead of patients, products instead of medications, transactions instead of diagnoses — but the structural challenge is identical: the same real-world entity is represented differently in different systems, and without a Master Data Management layer that resolves these representations into a canonical form, AI models will learn fragmented, contradictory truths.

Follow OBT on LinkedIn