FHIR and TEFCA interoperability still failing

Arinder Singh Suri, founder and chief executive of Taction Software, argued on August 5 that many healthcare AI projects fail for reasons that appear before any model is trained: bad timestamps, inconsistent schemas, weak normalization and brittle data handoffs. In an article published by HIT Consultant, Suri said pilots often look promising in demos and then break when teams try to scale them across real clinical systems. (hitconsultant.net) He identified FHIR APIs and TEFCA-enabled interoperability as the infrastructure needed to make preprocessing reproducible and data movement more reliable. ONC and the Sequoia Project’s Recognized Coordinating Entity describe TEFCA as a nationwide exchange framework and say FHIR is being built into that exchange model. (rce.sequoiaproject.org) ### Why are AI teams still getting stuck before the model stage? Suri wrote on August 5 that the recurring failure point is “rarely the model” and instead the data feeding it. He said healthcare organizations still struggle with source systems that label the same concept differently, omit acquisition times, or package records in ways that are hard to normalize consistently. FHIR, or Fast Healthcare Interoperability Resources, is intended to give systems a common API-based structure for exchanging clinical data. But ONC’s TEFCA materials say earlier TEFCA exchange relied mainly on existing IHE-based exchange and C-CDA content, with FHIR added later as standards for multi-network exchange matured. (hitconsultant.net) ### What does TEFCA actually fix, and what does it not fix? The Sequoia Project’s “TEFCA on FHIR” page says FHIR is designed to let organizations conduct transactions directly with other TEFCA participants without a prior one-off connection. That addresses one part of the problem: network-level connectivity and a common exchange pathway. (hitconsultant.net) Suri’s article described a separate layer of failure inside the data itself. Even when records move successfully, he said AI programs can still break if timestamps are missing, units are inconsistent, fields are mapped differently across systems, or preprocessing steps cannot be reproduced from one dataset to the next. (rce.sequoiaproject.org) ### Why do timestamps and schemas matter so much? Missing timestamps change the meaning of clinical data. A lab value, medication event or bedside reading can be clinically useful only if a downstream system knows when it was captured and how it relates to other events in the record. Suri wrote that absent or inconsistent timing data makes it harder to build reliable pipelines for training, validation and operational use. (rce.sequoiaproject.org) Clear schemas matter for the same reason. If one source uses different field names, coding systems or units than another, teams can produce different preprocessing outputs from what should be the same underlying event. Suri said that breaks reproducibility, which is a basic requirement for scaling healthcare AI beyond a pilot. (hitconsultant.net) ### What does this mean for medtech builders and engineers? Suri’s practical test was simple: prove that the plumbing works. He pointed to timestamped acquisition, explicit schemas and validation tooling as evidence that a team can manage real clinical data rather than a cleaned-up demo dataset. (hitconsultant.net) For device and software teams, Python validation scripts are one way to show that discipline. A script that checks dropped samples, malformed fields, unit mismatches and missing timestamps does not solve interoperability on its own, but it documents the assumptions in the pipeline and makes failures visible. That is consistent with Suri’s argument that the bottleneck is often data engineering, not model selection. (hitconsultant.net) ### Where is the interoperability work headed next? ONC’s interoperability pages and TEFCA materials say the framework is intended to support nationwide data sharing across participants, while the FHIR roadmap lays out how API-based exchange is being incorporated into that system. (hitconsultant.net) The Sequoia Project’s TEFCA materials direct implementers to the roadmap and “TEFCA on FHIR” guidance for the next phase of exchange design. HIT Consultant published Suri’s article on August 5, and the ONC and Sequoia Project pages remain the main public references for how FHIR-based TEFCA exchange is being structured. For teams building healthcare AI now, those documents set the technical direction, while Suri’s warning centers on the work that still has to happen inside the data pipeline itself. (hitconsultant.net) (rce.sequoiaproject.org)