News
FHIR-native AI agents for clinical trial screening reduce eligibility review from days to hours
A NEJM AI perspective published in August 2026 outlines a pragmatic trial operations framework for AI agents that read patient FHIR data and match it against clinical trial eligibility criteria in real time. Early deployments at academic medical centers report reducing eligibility review from two to three days of manual chart review to under four hours per candidate, with false-negative rates comparable to trained research coordinators.
- FHIR
- clinical-trials
- AI-agents
- eligibility-screening
- NEJM-AI
- interoperability
Clinical trial enrollment has been one of the most consistently cited bottlenecks in drug development. Roughly 80% of trials fail to meet enrollment targets on schedule; the median trial takes twice as long to enroll as projected; and the manual chart review required to identify eligible patients is expensive, slow, and dependent on research coordinator capacity that most health systems lack.
The FHIR-native AI agent approach addresses the identification step specifically: automatically reading structured patient data (diagnoses, medications, lab values, procedures) against trial eligibility criteria expressed in machine-readable format. The 510(k) status of these systems varies — some are positioned as research tools outside FDA jurisdiction, others as clinical decision support under the Cures Act exclusion, and some may eventually attract device scrutiny if they are positioned as influencing treatment decisions.
What makes the NEJM AI framework notable is the explicit operationalization: not just “AI can do eligibility screening” but a specific playbook for how to run a pragmatic trial that validates the AI agent’s performance within a clinical environment, using the existing trial as both the deployment context and the evaluation dataset. This is the right epistemic approach for clinical AI — building the evidence alongside the deployment rather than certifying the AI in isolation and then deploying.
The interoperability layer matters for scalability. FHIR-native agents that can read eligibility criteria from standardized formats (CDISC, mCODE for oncology) and patient data from any FHIR-compliant EHR can be deployed without custom integration for each site. Health systems that have completed FHIR R4 implementation as required by ONC interoperability rules are positioned to run these agents without significant additional infrastructure investment.
Primary source: Read the full original on NEJM AI ↗