AI Healthcare Development Pitfalls: The Importance of HL7 and FHIR

Details

This article is an opinion column by a professional who has worked on medical IT product development. It argues that in developing AI healthcare platforms, data layer unpreparedness—rather than model performance itself—is the main cause of failure. Specifically, it discusses standards such as HL7 v2 (the messaging standard still running in most live EHR environments), FHIR (the structured data layer providing usable clinical context for AI systems), and SMART on FHIR (the authorization layer governing what an app can access). Crucially, the article links these standards not merely to abstract technical concepts but to concrete business outcomes like 'procurement delays' or 'failed compliance audits.' A particularly valuable section is a readiness checklist that adopts a 'pre-mortem' perspective. It raises practical questions—such as whether FHIR mapping has been tested against live EHR data, or if PHI controls are enforced at the data layer rather than bolted onto the application—reflecting how a CTO actually thinks during due diligence. The piece advocates treating interoperability not as a mere technical footnote but as a core business risk. It also provides a realistic view of timelines (e.g., framing a 90-day plan as a structure, not a guarantee).