FHIR gaps hobble healthcare AI

Arinder Singh Suri, founder and chief executive of Taction Software, argued on August 5 that healthcare AI projects usually fail at scale because the underlying data are fragmented, inconsistently coded and hard to move between systems. In a commentary published by HIT Consultant, Suri said the recurring problem is “rarely the model” and instead sits in the data feeding it. The article lands as federal interoperability work continues to push FHIR-based exchange and TEFCA governance into wider use. The Office of the National Coordinator for Health Information Technology says TEFCA is meant to provide a common governance, policy and technical floor for nationwide exchange, while ONC’s recent materials also point to ongoing FHIR implementation and testing work. (hitconsultant.net) ### Why do AI pilots work in demos and then break in production? Suri wrote that many healthcare AI initiatives begin with a promising pilot and then fail when organizations try to expand them beyond a narrow use case. He said the obstacle is not the algorithm itself but the condition of the source data, which often arrive from multiple systems with different formats, identifiers and workflows. (healthit.gov) ONC has published similar evidence on the uneven state of API-based exchange. A July data brief said hospitals have increased use of HL7 FHIR APIs, but adoption still varies by use case, and hospitals use a mix of API and non-API approaches to share data with third-party technology. (hitconsultant.net) ### What does FHIR change if the records are still messy? FHIR is a standard for structuring and exchanging health data through application programming interfaces, but Suri’s argument was that standard pipes do not fix bad source data on their own. He said healthcare organizations still need normalization, identity matching and workflow discipline before AI systems can use information reliably across sites and departments. (healthit.gov) ONC’s interoperability materials make the same distinction in more formal terms. The agency says USCDI defines standardized data classes and elements for exchange, including laboratory data, while separate testing and implementation tools are needed to validate whether systems actually conform to FHIR requirements. (hitconsultant.net) ### Where does TEFCA fit into this? TEFCA is the federal framework designed to connect disparate health information networks under a common agreement. ONC says TEFCA is intended to simplify connectivity across organizations and enable secure nationwide exchange, and agency materials describe a FHIR roadmap and a facilitated FHIR implementation process inside that broader framework. (isp.healthit.gov) Suri pointed to that governance layer as part of what healthcare AI needs. His argument was that AI systems require not only access to data, but also rules for how information is exchanged, trusted and reused across organizations. ### Why does this matter more for laboratories than it may sound? (healthit.gov) Laboratory data are among the most structured pieces of the medical record, but they still depend on consistent coding, correct patient matching and clear result status. ONC’s Interoperability Standards Platform lists laboratory tests, values and result status among the standardized elements needed for exchange, and its laboratory materials tie those elements to diagnosis, treatment and public-health reporting. (hitconsultant.net) For labs, that means LIS and LIMS work becomes part of AI readiness. A model that reads lab results can misfire if the same test is coded differently across sites, if results are linked to the wrong patient, or if audit trails do not show where a value originated. That conclusion is an inference from Suri’s argument and ONC’s interoperability specifications. (isp.healthit.gov) ### What should readers watch next? ONC’s public interoperability work gives the clearest markers for what comes next: TEFCA implementation updates, FHIR testing tools, and continued expansion of standardized data elements and exchange pathways. HIT Consultant’s commentary did not announce a product or financing event; it made the case that health systems will need to spend on “data plumbing,” standardized APIs and operational workflows before AI can spread safely through care delivery. (hitconsultant.net) ONC’s Lantern dashboard and certification resources provide public signals of where FHIR endpoint availability and standardized API testing are moving, while local progress will depend on named participants inside hospitals — EHR vendors, interface teams, laboratory systems staff and compliance officers. (lantern.healthit.gov) (healthit.gov)