Case study 04
Platform Integration Design
An example of designing an integration strategy to connect multiple legacy systems, with data flows, dependencies, and technical considerations.
- ScenarioA fictional clinic network bringing three result sources into one platform
- My roleProduct manager working with architects and engineering leads
- SystemsLab system, imaging system, legacy archive
- ChannelsWeb app, mobile app, provider portal
The problem
Point-to-point does not scale
Three source systems feed three applications. Connected one by one, that is nine interfaces, each with its own data format, error handling and owner. Every new channel would add three more.
- Codes do not match between systems, so the same test appears under different names.
- Results arrive late in some channels because the legacy archive only exports overnight.
- Failures are silent: when an interface breaks, users find out first.
My approach
Design the flow, then the parts
- InventoryListed every system, the data it owns, its volume and how it can share data today.
- Target architectureAn integration layer between sources and channels: API gateway, data transformation and mapping, and a standard FHIR API.
- Common data modelAgreed one representation of a result and its status, and a mapping table from each source's codes to it.
- Non-functional requirementsLatency, security, audit and failure handling, written as testable requirements, not left to the end.
- Phasing and dependenciesLab first (highest volume, cleanest interface), imaging second, the legacy archive last.
Sample artifact
Technical considerations
| Area | Requirement | Why |
|---|---|---|
| Latency | A final result is visible in every channel within 5 minutes | Clinicians act on results during the visit |
| Standards | Results exposed through a FHIR API | New channels and partners connect without custom work |
| Code mapping | Every source code mapped to one standard code; unmapped codes go to a review queue | One test, one name, everywhere |
| Security | Access by role and scope; every view is logged | Patient privacy and audit requirements |
| Failure handling | Automatic retry, alert after 3 failures, dashboard of message status | The team finds problems before users do |
| Legacy archive | Nightly batch until it is retired | Event support would cost more than the archive's remaining life |
Key decisions
Trade-offs I made
Hub instead of point-to-point
Options: keep adding direct interfaces, or build an integration layer.
Chose the hub. Six connections instead of nine today, and one more (not three) for each new channel.
Events where it matters, batch where it does not
Options: real-time for everything, or match the method to the need.
Mixed. Lab and imaging results flow as events; the legacy archive stays on a nightly batch until retirement.
Lab first
Options: start with the hardest system, or the one with most value.
Started with lab. Highest volume and a clean interface, so users saw value early and the team learned on the easier source.
Outcomes
How success is measured
- 9 to 6Interfaces to maintain
- < 5 minFrom final result to every channel
- < 0.5%Results with unmapped codes
- 0Silent failures: every error alerts the team
Illustrative targets. My role in a design like this is to make sure the architecture serves the user problem, that dependencies are known early, and that the technical requirements are clear enough to test.
Illustrative example. This work sample uses a fictional scenario and synthetic data. It does not contain confidential or proprietary employer information.