Stop Chasing FHIR Shiny Objects and Fix the Semantics

Imagine you're the informatics lead at a mid-sized health system. Your CMIO just returned from HIMSS and wants to rip out the HL7 v2 interfaces that have run your lab and radiology feeds for two decades and replace them with FHIR APIs. You know the v2 pipes are ugly. You also know they rarely drop a result. So do you rip and replace, or do you fix what's actually broken?

We've watched this movie before. The interoperability conversation in health informatics has been hijacked by the transport layer. We treat FHIR as the finish line when it's really just the starting gun. Our position: the rate-limiting step in U.S. interoperability is no longer connectivity. It's semantics. If you have budget to spend this year, spend it on terminology governance and mapping, not on another API gateway.

Start with the actual state of adoption. As of 2021, 96 percent of U.S. non-federal acute care hospitals had adopted certified health IT, and about four in five office-based physicians had a certified EHR (ONC / HHS Report to Congress). In 2011, those numbers were 28 percent of hospitals and 34 percent of physicians. That's a staggering climb in a decade. The pipes are in the ground. The problem is what flows through them.

The semantic layer is where interoperability actually dies

Here's the concrete failure mode we see weekly. A patient gets a basic metabolic panel at an outside lab. The result flows into your EHR through a perfectly conformant HL7 v2 message. The lab used LOINC code 2823-3 for potassium. Your internal problem list and your analytics warehouse use a local code, or a different LOINC mapping, or a free-text string that says "K+". The message delivered. The data didn't. Clinically, you're blind.

This is not a transport problem. It's a vocabulary problem. LOINC, stewarded by the Regenstrief Institute, is the international standard for identifying health observations and measurements. Its version 2.82 release in February 2026 contains 109,325 total concepts, including 66,861 laboratory concepts and 28,635 clinical concepts (LOINC). That's a lot of surface area. If your organization has not invested in a curated LOINC map for your top 200 lab tests, you are accumulating semantic debt every single day.

We keep betting on the wrong fixes

The vendor and policy ecosystem rewards visible, demo-able work. FHIR resources are visible. A Patient resource, an Observation resource, a RESTful query that returns JSON — that's a demo. Mapping 800 local orderables to SNOMED CT and LOINC is not a demo. It's a slog. But it's the slog that determines whether a downstream clinician can actually see a diagnosis or a result.

We're not anti-FHIR. Far from it. FHIR R4, published December 27, 2018, was the first release with normative content, and R5, published March 26, 2023, defines 157 resources (HL7 FHIR). The US Core Implementation Guide, based on R4, gives us a shared floor for patient data access. That's real progress. But an API that returns a semantically garbled Observation is not interoperability. It's just faster confusion.

The regulatory tail is also wagging the dog. The HTI-1 final rule made USCDI v3 the baseline data standard in the ONC Health IT Certification Program as of January 1, 2026 (ONC / HHS HTI-1 Final Rule). USCDI v3 expanded from 52 data elements in 16 classes in v1 to 94 elements in 19 classes. That's a meaningful expansion of the required data surface. But notice what it doesn't do: it doesn't force your organization to map those elements consistently to your local workflows. Certification is a floor, not a strategy.

The strongest counter-argument, and why it still loses

The best argument against our position is that semantics are someone else's job. Vendors ship standard terminologies. The government mandates USCDI. Why should a health system's informatics team spend scarce analyst hours on terminology curation when the tools are supposed to handle it?

We reject that on two grounds. First, the tools don't handle it. A FHIR server will happily store whatever code you send it. RxNorm normalizes clinical drug names and links to multiple pharmacy vocabularies, but it doesn't know that your local formulary calls something by a different brand alias. Second, the cost of getting it wrong is now asymmetric. The HIPAA civil monetary penalty calendar-year cap rose to $2,190,294 in the January 2026 inflation adjustment, with a maximum per-violation penalty of $73,011 for Tiers 1 through 3 (Federal Register 2026 HIPAA CMP Adjustment). A breach involving mislabeled or misrouted PHI is not a theoretical risk. And the 60-day breach notification clock under 45 CFR 164.404 starts when you discover it, not when you finish your root-cause analysis.

So no, we don't think the semantic layer is someone else's job. It's the job.

What we actually recommend

Stop treating FHIR migration as the interoperability program. Treat terminology governance as the interoperability program, and treat FHIR as one delivery mechanism among several. In practice, that means three moves.

First, fund a terminology services function. Not a side project for an analyst who also does reporting. A named owner, a budget line, and a mandate to curate your top lab, medication, and diagnosis vocabularies against LOINC, RxNorm, and SNOMED CT. Second, measure semantic concordance, not just interface uptime. If your potassium result arrives with a code that doesn't match your reference lab's code, your interface is up and your interoperability is down. Third, keep your HL7 v2 interfaces where they earn their keep. HL7 v2 was first published in 1987 and remains the workhorse for high-throughput legacy workflows, while FHIR is favored for developer-facing and cloud-native applications (HL7 International). That division of labor is fine. Don't rip out the plumbing to install a new faucet.

One more thing: if you're evaluating AI features built into certified health IT, the HTI-1 rule now requires transparency for those algorithms. That transparency is only as good as the data feeding them. Garbage semantics in, biased or useless predictions out. The NIST AI Risk Management Framework is voluntary and useful, but it won't clean your LOINC map.

Bottom line

The single best move for most informatics teams this year is to stop chasing the next FHIR release and start funding a real terminology governance program. Connectivity got us to 96 percent hospital adoption. Semantics is what gets us to actual exchange. Pick the slog.

Sources

• ONC / HHS (Report to Congress) - https://healthit.gov/news/onc-outlines-health-it-interoperability-progress-report-congress/
• LOINC - https://loinc.org/
• HL7 International - https://www.hl7.org/fhir/
• ONC / HHS (HTI-1 Final Rule) - https://healthit.gov/regulations/hti-rules/hti-1-final-rule/
• Federal Register (2026 HIPAA CMP Adjustment) - https://www.federalregister.gov/documents/2026/01/28/2026-01688/annual-civil-monetary-penalties-inflation-adjustment