Interview with Data Experts on Healthcare Data Modernization
Healthcare organizations generate enormous amounts of data every day: clinical records, operational data, financial information, and streams of real-time signals from dozens of systems. But transforming this data into trustworthy, accessible, and actionable insights is far from trivial. Zuyderland, one of the largest hospitals in the Limburg region (NL), is tackling this challenge head-on by modernizing its data landscape and laying the foundation for an AI-ready architecture.
To achieve this, Zuyderland is modernizing their traditional analytics focused data platform and moving to a modern all-encompassing platform powered by Microsoft Fabric, OneLake, FHIR, and OpenEHR. This marks a significant shift: instead of one central team handling all data ingestion and reporting, Zuyderland is gradually adopting a data mesh model where domains gain ownership of their own data products, while staying aligned through shared governance and platform standards.
From building a unified data foundation to enabling new use cases such as cloud cost monitoring (FinOps), ICU risk prediction, and secure patient-portal access logic, this journey is both technically complex and strategically transformative. To understand how this project evolved behind the scenes, what challenges were encountered, and how it is already creating value for the organization, we sat down with the data engineers directly involved in the work.
What follows is a conversation about architecture decisions, lessons learned, and how a modern data platform is enabling Zuyderland to build the next generation of healthcare innovation.
Q: What was the initial problem or opportunity that sparked this project with Zuyderland?
Zuyderland was facing the challenge that their data ecosystem was scattered across many different applications. There was no information on data quality, lineage or usage of data within the organization. That is why they wanted to build a centralized platform which they can use for all their data-related use cases.
Q: How did the idea of moving toward a data mesh architecture come into play?
They wanted to have a shared infrastructure for all their data workflows but still wanted to give the domain experts the ability to have ownership on their data.
Q: What was the biggest technical challenge you encountered during the platform implementation?
One of the biggest challenges is onboarding legacy technology into a modern, cloud-native platform while still meeting near real-time expectations. Zuyderland aims to make data available quickly across the organization, but many vendor solutions are not designed for real-time integration or modern event-driven architectures.
That means we often need to design innovative custom ingestion solutions, sometimes combining Fabric pipelines, and event-driven approaches, to reliably ingest and process data at the speed the domains need, without compromising security or governance.
Q: Were there any unexpected roadblocks in working with healthcare data standards or interoperability requirements?
One notable challenge was that OpenEHR is not yet widely adopted in mainstream industry applications. In practice, it is used more often in research settings, pilots, or proof-of-concepts than in large-scale production environments. This gives additional practical challenges on how to host and integrate OpenEHR in the Zuyderland ecosystem. In addition, the ecosystem of OpenEHR tools and platforms is still maturing. While there are several promising solutions, they differ in terms of scalability, tooling, operational maturity, and cloud readiness. Selecting the right OpenEHR framework therefore becomes a non-trivial architectural decision, requiring careful evaluation of long-term support, integration capabilities, and alignment with cloud based solutions.
Q: In your view, what makes a data mesh suitable for a healthcare organization like Zuyderland?
Healthcare is extremely domain-driven, clinical domains and operational departments all have different workflows, terminology, and priorities. A data mesh fits well because it allows domains to take ownership of their own data products close to the source of knowledge, without being blocked by a central team.
At the same time, healthcare requires strong centralized guardrails: governance, auditability, security, and interoperability standards. That’s why the hybrid model Zuyderland is using is so effective: domains can grow their independence, while central BI/DS teams and the platform team ensure that quality and compliance remain consistent across the organization.
Q: How does this new data platform add value for Zuyderland in the short term? And in the long term?
Short term, the platform creates clarity and momentum. Domain teams can start building within a secure and standardized environment without waiting for infrastructure or pipeline capacity. That means faster delivery of dashboards, reporting, and domain-specific analytics, and a measurable reduction in the bottleneck effect.
Long term, the value is even bigger: the platform becomes an interoperable, AI-ready foundation. With standardized clinical models (OpenEHR), governed analytical storage (OneLake), and full lineage (Purview), Zuyderland can scale advanced use cases such as predictive analytics, resource planning, and clinical decision support, with trust and auditability built in.
Q: How does this setup enable Zuyderland to adopt advanced AI use cases more easily?
AI in healthcare depends on high-quality, standardized, and well-governed data, otherwise models become unreliable or difficult to explain. This platform helps because it enforces consistent modeling and data product ownership at the domain level, while still providing central governance and security.
The medallion structure (bronze/silver/gold) makes it easier to produce reliable training datasets, and DTAP environments ensure reproducibility and controlled promotion to production. That combination reduces friction for experimentation while keeping the organization compliant and safe, which is essential when AI outputs influence clinical or operational decisions.
Q: Can you walk us through one concrete use case?
A good example is the FinOps use case. Cloud consumption data from Azure is ingested into the platform and made available in a dedicated Fabric workspace. From there, the BI department builds thebronze, silver, and gold datasets, including structured cost models and reporting layers. The finance domain uses these datasets to do their own analysis and build their own reports. This allows the team to monitor and optimize cloud spending independently, while still using the same governance, security, and architectural patterns as other domains.
Q: What role does Fabric play in enabling that use case?
Fabric acts as the unifying analytics and storage layer. It provides managed infrastructure, integrated tooling, and OneLake as a shared analytical storage layer. This allows data to be reused across domains without duplication and enables teams to focus on analytics and value creation rather than platform management.
Q: If you could start the project from scratch, is there anything you would do differently?
As this project spans over multiple years, lots of changes in the architecture and technology stack have been made. As we used new technologies, we ran into several limitations as our tech stack are still under heavy development. For example, we started with Databricks for ingestion as Fabric data connectors were not compatible yet with Zuyderland’s systems and network requirements. As the platform and tech stack became more mature, we were able to move more components to Fabric. We also removed some technologies as they either became redundant or turned out to not completely match our goal. Due to this project, we have a better understanding of which technologies to use for future projects.
Q: What skills or mindset are most important for success here?
A willingness to invest and learn new technologies was required for the success of this project. This is important as new technologies have some learning curves and also a creative mindset is needed to overcome limitations that go with new technologies.