Medallion Architecture for PE-Backed Provider Groups

‍

PE-backed specialty clinic groups face a data problem that generic medallion architecture guidance wasn’t built to solve: a data estate that fragments with every acquisition. Each new clinic brings a different EMR, a different billing system, and a different data model. Standard approaches break under that kind of structural instability. In this article, we’ll go over how Bronze, Silver, and Gold layers solve multi-site data fragmentation for outpatient specialty groups.

‍

Why Generic Medallion Guidance Fails the Multi-Site CIO

‍

Most medallion architecture content was written for data engineers inside stable source systems, not for teams inheriting a new Athena instance every time the PE sponsor closes a deal. CAQH research puts the annual cost of routine administrative workflows in U.S. healthcare at $90 billion, with another $20 billion in savings available from wider electronic adoption. That is the operating environment for teams in this position.

‍

The Real Problem: A Data Estate That Never Stops Changing

‍

The core challenge is the data is structurally designed to keep fragmenting. A PubMed-indexed study found that PE-acquired physician practice sites grew from 816 sites in 2012 to 5,779 by 2021. A separate 2024 study found that 51.6% of PE-acquired practices exited within three years, meaning the infrastructure built in year one is what a buyer evaluates in year three. This is not a one-time problem. The infrastructure must absorb new sites without full platform re-engineering every six to nine months.

‍

Why Raw EMR Exports Create Brittle Gold Layer Pipelines

‍

Some teams skip the structured architecture and build dashboards directly on raw EMR exports. It works for about 90 days. Then a source system upgrade reshapes an Athena schema, a newly acquired cardiology group documents CPT codes differently than the allergy practices already in the portfolio, or a corrected 835 remittance breaks the reconciliation logic. MGMA data shows 60% of medical group leaders reported higher claim denials in early 2024 than the year prior. Brittle pipelines make that problem invisible until it is expensive.

‍

The Architectural Argument: Why Bronze-Silver Separation Scales

‍

The medallion framework solves this by separating raw data preservation from normalized data. As defined in both the official Databricks medallion architecture guide and Microsoft Fabric's medallion guidance, the goal is progressively more reliable, analytics-ready data without sacrificing source fidelity. Bronze captures everything as-received. Silver normalizes it. Gold delivers domain-specific datasets to operations, finance, and clinical leadership. That layering lets a multi-site outpatient organization absorb new clinics without rebuilding reporting every time.

‍

Bronze Layer: Raw Ingestion Across a Heterogeneous Tech Estate

‍

Bronze is a vault, not a transformer. Its job is to preserve everything exactly as it arrived, with a timestamp and a source label. Normalization and interpretation happen in Silver.

‍

The Outpatient Specialty Source Inventory

‍

For a multi-site specialty clinic group, Bronze ingests HL7 v2 and FHIR R4 feeds from:

‍

• Ambulatory EMRs
• CCDA clinical encounter summaries
• 837P professional claim files
• 835 remittance advice from payers
• 270 and 271 eligibility transaction files 
• Scheduling data from practice management systems
• General ledger exports from financial systems like NetSuite
• ERP data from platforms such as Workday
• HR and timesheet data from HRIS systems like ServiceNow

‍

Each carries different payload structures and timing expectations. Financial and HR sources matter here too, since timesheets and labor data feed directly into cost and margin calculations. Getting all of it in without loss or transformation is the foundation everything else depends on.

‍

The EMRs That Actually Define Outpatient Specialty Environments

‍

The outpatient specialty world runs on EMRs like Athena, eClinicalWorks, Modernizing Medicine, and NextGen. Each exposes data differently. Athena publishes FHIR APIs with one set of conventions. eClinicalWorks approaches interoperability differently. Modernizing Medicine's certified FHIR API is built around specialty-specific data models. NextGen's FHIR API has its own structure and authentication patterns entirely. Two practices on the same EMR can send structurally different data depending on how their workflows were locally configured.

‍

Source System Heterogeneity as the Defining Bronze Layer Challenge

‍

Heterogeneity, not volume, is the defining Bronze challenge in a PE-backed portfolio. The answer is schema-on-read, consistent timestamping, and a source tagging convention that tells the Silver layer exactly where every record came from. Forcing a Modernizing Medicine encounter export to look like an Athena export at ingestion destroys the downstream context needed for both analytics and audits.

‍

Why Early Normalization at Bronze Undermines Claim Disputes and Due Diligence

‍

When a payer disputes a claim, the denial must be traceable back to the original 837P file. When a PE acquirer conducts technical due diligence, they want metrics traceable to raw source data. Early normalization at Bronze collapses that audit trail. The result is a Silver-like layer masquerading as raw storage and losing the forensic capability that separates a credible platform from a fragile one.

‍

Silver Layer: Where Most Outpatient EDWs Break

‍

If Bronze is a vault, Silver is a translation bureau. It turns raw source data from multiple EMRs into something consistent enough to support analytics. It is also where most outpatient enterprise data warehouse projects quietly fail.

‍

Patient Identity Unification Across Clinic Locations Through EMPI Logic

‍

The same patient appears in the Athena instance at the allergy practice and the eClinicalWorks instance at the cardiology clinic under two different medical record numbers. Without enterprise master patient index logic in Silver, the analytics treat one person as two people, care gap reports are wrong, and population health metrics are broken. EMPI uses probabilistic matching across name, date of birth, address, and insurance identifiers to collapse duplicate identities into one canonical record. Every new acquisition reintroduces this problem.

‍

Claim Denial Structuring for RCM Root Cause Analysis

‍

A raw 835 remittance identifies a denied claim but does not indicate whether the denial was clinical, eligibility-related, or prior-auth-related without parsing adjustment reason codes. Silver separates those categories so Gold analytics can pinpoint which payer, location, and denial type drives avoidable write-offs. Without that structure, RCM gets managed on gut feel.

‍

Prior Authorization Data as a Silver Layer First-Class Concern

‍

Prior auth workflows generate approvals, denials, and overturn records. The AMA's 2024 prior authorization survey found practices complete an average of 39 prior authorizations per physician per week, consuming roughly 13 hours of physician and staff time, with only 31% of transactions fully electronic. A 2024 federal interoperability rule requires payers to report auth turnaround and denial metrics publicly beginning in 2027. Silver must treat prior-auth records as a first-class entity structured for operational and contract performance reporting, not buried in a raw transaction log.

‍

Provider Scheduling and Productivity Normalization Across Locations

‍

Provider productivity is a standard board ask in PE-backed portfolios. The challenge is that visit volume and documentation practices vary across every acquired clinic. Silver normalization for scheduling data requires explicit logic for how visit types map across sites, how incomplete encounters are handled, and how provider-level output is aggregated into genuinely comparable numbers. Getting this wrong means comparing locations on metrics that measure different things.

‍

Why Silver Layer Design Requires Clinical and RCM Input

‍

Decisions in Silver about how to classify a denial, attribute a patient encounter, or define a completed visit carry direct revenue consequences. Engineering teams designing normalization logic without RCM and clinical input produce a structurally clean but operationally misleading Silver layer. With medical practice operating costs rising roughly 11.1% year-to-date in 2025, there is no margin for Silver layer rework caused by misaligned definitions.

‍

Gold Layer: Domain-Specific Datasets for Operations and Finance

‍

Gold is where the medallion stack delivers value. It is not a single dataset but a collection of domain-specific views built for specific audiences, from specific Silver inputs, with metrics defined in advance.

‍

Revenue Cycle Gold Domain

‍

The RCM Gold domain is where denial rate trending, days in accounts receivable, net collection rate by payer and location, and clean-claim rate by CPT family live. These are the metrics the VP of Revenue Cycle and COO review weekly. A denial rate that cannot be reconciled to the underlying 835 data creates false confidence, and a clean-claim rate not segmented by payer and location cannot drive targeted improvement.

‍

Clinical Operations Gold Domain

‍

Provider productivity, visit volume by service line, referral conversion rates, and care gap closure rates belong here. A gastroenterology group needs Gold centered on colonoscopy scheduling lag and prep adherence. An orthopedic group needs imaging utilization, surgery conversion, and post-op therapy leakage. A retina group needs injection cadence, authorization status, and visit adherence. These are specialty-specific signals, not generic BI metrics.

‍

Population Health Gold Domain

‍

For practices carrying VBC risk, the population health Gold domain covers chronic disease management, preventive screening completion, and avoidable utilization. These are the metrics that determine whether a specialty group is managing its attributed population or simply reacting to it. A cardiovascular group tracking preventive screening completion across its panel, or a gastroenterology group monitoring avoidable ED utilization among chronic GI patients, needs this domain built and current before the payer sends its first performance report.

‍

Financial Performance Gold Domain

‍

Cost per visit, administrative cost as a percentage of revenue, and payer mix shifts aligned to the EBITDA metrics the PE sponsor tracks live in the financial performance Gold domain. When built correctly, the CFO can explain a payer mix deterioration at a specific location to the board without a data analyst pulling a custom query. That self-service capability is the operational payoff.

‍

Why Gold Layer Design Is a Business Alignment Exercise

‍

The most expensive Gold layer mistake a CIO can make is designing it without first aligning with the CFO and VP of Revenue Cycle. When stakeholders reach different conclusions from the same underlying data because metric definitions were never documented, the credibility of the entire platform collapses. Gold layer design starts with those structured business conversations. Engineering follows.

‍

Post-Acquisition Onboarding: The Use Case That Justifies the Investment

‍

The clearest use case for the architectural investment is post-acquisition onboarding. Every new clinic triggers the same sequence: 

‍

• Raw EMR extraction 
• Bronze schema mapping
• Silver entity resolution
• Gold metric promotion 

‍

The question is whether that sequence takes four weeks or four months.

‍

The Predictable Sequence Every Acquisition Triggers

‍

Bronze receives the new source system's data with proper tagging. Silver mapping runs against the existing patient population to resolve overlaps and normalize new data. Gold metric promotion adds the new location to existing dashboards. The architecture either handles this as a repeatable playbook or as a bespoke engineering project every time.

‍

What a Mature Onboarding Playbook Looks Like

‍

A mature playbook includes a source system intake checklist covering API credentials and field mapping specs, automated Bronze schema validation, Silver mapping templates for the most commonly acquired EMR types, and Gold promotion logic that adds locations without rebuilding datasets. When a new Athena instance arrives, it is a configuration exercise, not a net-new engineering engagement.

‍

The Time and Cost Argument

‍

For a specialty group where the average PE exit occurs within three years, a four-month onboarding cycle per acquisition is a structural limitation on how fast the portfolio can produce unified reporting. The difference between a four-week onboarding and a four-month one, across six acquisitions over two years, is a significant cost differential and a major reduction in the time-to-visibility the PE sponsor needs to make decisions.

‍

Build vs. Buy: Databricks or Microsoft Fabric

‍

The platform choices for a PE-backed outpatient specialty group have narrowed to Databricks and Microsoft Fabric. Both support a medallion-like architecture natively. Neither delivers healthcare-specific Silver layer normalizations out of the box.

‍

What Platforms Like Databricks or Microsoft Fabric Actually Provide

‍

Microsoft Fabric combines OneLake unified storage, native Lakehouse architecture, and built-in Power BI into a single governed platform. No separate governance tool is required. Databricks organizes a similar stack through Delta Lake tables and Unity Catalog for governance. Fabric's F2 starter capacity begins at roughly $262.80 per month pay-as-you-go. On the Databricks side, Lakeflow Jobs compute runs at $0.15 per DBU and Databricks SQL Pro starts at $0.22 per DBU. Platform subscription cost, though, is only part of the total build equation.

‍

The Decision Variables

‍

The variables that actually drive the decision are portfolio size and EMR complexity, internal engineering team capacity, payer contracting sophistication, and time horizon to exit. A ten-location dermatology group with a two-year sale timeline has a fundamentally different calculus than a forty-location cardiovascular group with complex VBC contracts and a five-year hold.

‍

The 2-Year Exit Runway Case

‍

For a specialty clinic group with a short exit runway and limited engineering resources, Microsoft Fabric is the stronger choice, particularly for organizations already running Azure and Microsoft 365. The native Power BI integration reduces tooling complexity and the managed infrastructure reduces DevOps burden. Speed to a credible, auditable platform matters more than flexibility when the diligence clock is running.

‍

The 5-Year Horizon Case

‍

For a group with a longer hold, Microsoft Fabric still delivers advantages beyond speed to value. The managed infrastructure reduces ongoing DevOps overhead. Built-in Power BI and OneLake eliminate separate licensing costs that compound over a multi-year hold. Engineering team cost must also enter the calculation. BLS data shows median annual wages of $135,980 for database architects, $112,590 for data scientists, and $133,080 for software developers. A lean three-role platform team implies roughly $381,650 in direct annual wage cost before benefits or contractors. That is the real comparison point against a managed platform subscription.

‍

VBC Contracts as a Gold Layer Design Requirement

‍

The HCPLAN 2024 measurement report found that 39.2% of commercial payments were tied to alternative payment models in 2023, with 21.6% in downside risk. For Medicare Advantage, 64.3% were tied to alternative models and 43.0% carried downside risk. Specialty groups that design their Gold layer only around fee-for-service metrics are building for yesterday's reimbursement environment.

‍

What VBC Arrangements Require the Provider Data Platform to Produce

‍

A shared savings contract requires utilization, cost of care, and quality performance reporting at the patient and provider level, reconciled against payer benchmarks. A bundled payment arrangement requires episode-level cost attribution across facility, professional, and post-acute spend. CMS Medicare Shared Savings Program 2024 results showed 476 ACOs serving more than 10.3 million beneficiaries generated $2.5 billion in net savings, with physician-led groups outperforming hospital-oriented ones. The Gold layer for a VBC-carrying specialty group must combine clinical, utilization, contract, and cost data at the patient and provider level.

‍

Why a Weak Gold Layer Undermines VBC Contract Management

‍

When the Gold layer cannot produce utilization, cost of care, and quality measure data segmented by contract, clinical and operational teams cannot tell which providers are driving excess utilization, which cohorts are trending above cost benchmarks, or which quality gaps are accumulating against bonus measures. The contract still runs. The financial exposure still accumulates. Without contract-aware Gold output, the group finds out at reconciliation, when it is too late to act.

‍

The Negotiation Risk: Unverifiable Payer-Reported Performance

‍

In most VBC contracts, the payer calculates performance and sends a report. The specialty group either accepts it or disputes it. Without the ability to run the same calculation independently, there is nothing to dispute. Specialty groups with a Gold layer capable of verifying utilization and quality metrics enter renewal conversations with leverage: they can confirm shared savings calculations, flag attribution errors, and negotiate from their own data rather than the payer's arithmetic.

‍

PHI Access Control Across Medallion Tiers

‍

A medallion architecture without PHI access controls is not production-ready. The HHS HIPAA Security Rule requires administrative, physical, and technical safeguards across all systems handling protected health information. OCR's 2022 report to Congress found that large breaches increased 107% from 2018 to 2022, and corrective action or civil monetary penalties were required in 80% of completed compliance reviews. The tiered medallion structure maps naturally to that compliance obligation.

‍

Bronze Layer Access Boundaries

‍

Bronze holds raw, identified patient data exactly as received. Access must be limited to automated pipeline credentials and a narrow set of senior platform engineers. No analyst, even a credentialed one, should query Bronze directly in production. Ad hoc query behavior against identified source data is where accidental PHI exposure originates, and broad Bronze access destroys the audit trail needed at exit.

‍

Silver Layer Access: Role-Based Controls for Clinical Data

‍

Silver access extends to analysts working with parsed clinical and claims records, scoped to the data domains their role requires. A revenue cycle analyst does not need access to clinical note structures. A population health analyst does not need financial claims detail beyond what cost modeling requires. Role-based Silver controls are not compliance overhead. They are what keeps a HIPAA breach from originating inside the analytics team.

‍

Gold Layer Access: De-Identification and HITRUST Mapping

‍

Gold is where broad organizational access becomes safe. De-identification and aggregation at the Gold layer let executives, department heads, and operational managers access the reporting they need without touching identifiable records. For multi-state specialty groups, de-identification logic must account for small location-level populations where aggregate data can still enable re-identification. HITRUST CSF control categories around access management, audit logging, and data classification map cleanly to this tiered structure, and a documented medallion access architecture accelerates HITRUST assessment cycles considerably.

‍

How Tiered Medallion Access Maps to HITRUST Controls

‍

HITRUST CSF control categories around access management, audit logging, and data classification map directly to the three medallion tiers. Bronze covers restricted access and full audit logging of unaltered source data. Silver covers role-based controls for analysts working with normalized clinical and claims records. Gold covers broad reporting access with PHI minimized through aggregation. For clinic groups in active PE portfolio management, where data moves across MSO structures, multiple legal entities, and third-party RCM vendors, a documented medallion access architecture provides a defensible control map that accelerates HITRUST assessment cycles. 

‍

Data Platform Maturity as a Due Diligence Asset

‍

When the sale process begins, the buyer's technical due diligence team is not evaluating the organization's Power BI dashboards. They are evaluating whether the numbers in those dashboards are real, reproducible, and auditable.

‍

What the Buyer's Due Diligence Team Looks For

‍

Experienced healthcare-focused PE buyers ask how a metric is defined, where the underlying data comes from, what the refresh cadence is, and whether the same calculation run last quarter produces the same result today. Most specialty clinic groups are unprepared for those questions. Arriving at diligence with disconnected SQL scripts and location-level exports is not a minor weakness. It is a direct discount on valuation.

‍

Bronze Auditability and Gold Metric Lineage as Key Signals

‍

The two features that most clearly signal data platform maturity are Bronze layer auditability and Gold layer metric lineage. Bronze auditability means the team can produce the original source file for any data point in the reporting stack. Gold metric lineage means any dashboard number can be traced back through Silver normalization logic to its Bronze source, with every transformation step documented. A specialty group that can demonstrate both is presenting itself as a data-credible acquisition target. One that cannot is presenting itself as a post-close cleanup project, and buyers price that accordingly.

‍

Reproducible Pipelines vs. Brittle SQL Scripts

‍

A reproducible medallion-based pipeline means any metric traces from the Gold domain through Silver normalization back to the raw Bronze source, with documented logic at every step. Brittle SQL scripts and disconnected location-level exports mean no two analysts agree on a revenue figure and the diligence team spends weeks reconciling numbers that do not tie. Buyers have seen both. They reflect the difference in their offers. 

‍

What the CIO Protects by Owning the Architecture

‍

What the CIO is protecting when they can walk a diligence team through the full medallion stack in one session is valuation. The contrast between a reproducible medallion-based pipeline from raw EMR ingestion to executive reporting and a collection of brittle SQL scripts stitched across location-level exports is a transaction-level distinction, not merely a technical one. No presentation-layer storytelling substitutes for showing a buyer exactly how a metric was produced, from what source, through what logic, with what governance controls in place.

‍

Final Thoughts

‍

Medallion architecture is not a new concept. What is new is applying it to the specific reality of a PE-backed outpatient specialty clinic group whose data estate expands every time the sponsor closes a new deal. The Bronze, Silver, and Gold structure gives CTOs and CIOs a framework that accommodates perpetual portfolio change without requiring full platform re-engineering at every acquisition. With administrative costs rising, denials trending higher, prior authorization still largely manual, and exit windows averaging three years, the groups that build this infrastructure early are not just solving a technical problem. They are building a due diligence asset that directly protects enterprise value at exit.

‍

FAQs

‍

How can Invene help a PE-backed specialty clinic group build a medallion architecture?

‍

Invene is a healthcare technology firm that helps PE-backed provider groups design and build the data infrastructure needed to support multi-site operations. For specialty clinic groups implementing medallion architecture, Invene handles the work that generic implementation guidance skips: EMR integration across heterogeneous source systems, Silver layer normalization for clinical and RCM data, and Gold layer domain design aligned to financial and operational reporting requirements. They bring both healthcare workflow expertise and engineering depth, which means the architecture they build reflects how outpatient specialty groups actually operate rather than how a generic data engineering team assumes they do.

‍

What is medallion architecture and why does it matter specifically for multi-site outpatient specialty clinic groups?

‍

Medallion architecture organizes data into three progressive layers: Bronze (raw ingestion), Silver (normalized and cleaned), and Gold (domain-specific analytics). For multi-site outpatient specialty groups, it matters because building reporting directly on raw EMR exports produces pipelines that break every time a source system changes or a new clinic joins the portfolio. The medallion structure absorbs those changes in Bronze and Silver without requiring the Gold reporting layer to be rebuilt from scratch.

‍

How should a CIO decide between Databricks and Microsoft Fabric for a medallion architecture in a specialty clinic group?

‍

The primary factors are the group's existing technology ecosystem, internal engineering team capacity, and time horizon to exit. Microsoft Fabric suits organizations already running on Azure and Microsoft 365, especially under a short exit timeline. Microsoft Fabric reduces time to production and lowers ongoing infrastructure burden for most specialty clinic groups. Databricks offers flexibility for groups with strong internal engineering resources and complex VBC analytics requirements. Neither delivers healthcare-specific Silver layer normalizations out of the box, so EMR integration work is required on either platform.

‍

What happens to Silver layer logic when a newly acquired clinic runs an EMR the organization has not onboarded before?

‍

A new EMR type requires a net-new Bronze-to-Silver mapping exercise: documenting how that system represents encounters, diagnosis codes, scheduling data, and claims, then building normalization transforms into the canonical Silver model. Mature organizations build Silver mapping templates for the EMR types they most commonly acquire, so a repeat onboarding of the same EMR becomes a configuration task rather than a new engineering project.

‍

Why is prior authorization data treated as a first-class concern in the Silver layer for outpatient specialty groups?

‍

Prior auth workflows produce three distinct record types: approvals, denials, and overturns, each with different operational and financial implications. The 2024 federal interoperability rule requires payers to report authorization turnaround and denial metrics publicly beginning in 2027. Specialty groups need structured prior-auth data in Silver to benchmark their own payer relationships and manage contract performance independently of payer-reported figures.

‍

How far in advance of a sale process should a PE-backed specialty clinic group begin building a medallion architecture?

‍

Earlier than most teams assume. Technical due diligence now routinely includes questions about data lineage, metric reproducibility, and platform governance. A medallion architecture in production for 12 to 18 months before the sale process gives the CIO time to document the stack, surface and resolve Silver-layer data quality issues, and demonstrate that Gold metrics have a stable, auditable history. Starting six months out typically means arriving at diligence with a credible architecture but thin historical depth, which still helps but not nearly as much as an early start.

‍

‍