TITLE:
AI-LEx: An AI-Driven Interoperability Architecture for Laboratory Exception Management, Patient Safety, and Public Health
AUTHORS:
Sai Krishna Puli, Priya Bandhavi Pichika
KEYWORDS:
Laboratory Information Systems, Semantic Interoperability, HL7 FHIR, Artificial Intelligence, Exception Management, Risk Scoring, Public Health Surveillance
JOURNAL NAME:
International Journal of Intelligence Science,
Vol.16 No.4,
October
23,
2026
ABSTRACT: AI-LEx introduces a risk-aware architecture that transforms laboratory interoperability exceptions from technical failures into clinically and publicly meaningful risk signals, by combining deterministic validation, machine-learning anomaly detection, semantic knowledge, and human-governed resolution. A missing specimen source or an unmapped code on a reportable pathogen looks, now it occurs, like a routine interface problem; left unresolved, it can instead become a missed diagnosis or an invisible gap in disease surveillance. Recent empirical work tracing real laboratory transmissions across health information exchanges documents exactly this pattern, and the FDA’s decade-long SHIELD initiative and a 2024 federal data brief both identify laboratory-data harmonization as an unresolved patient-safety and surveillance concern. AI-LEx addresses it with a three-stage architecture: Detection combines deterministic validation with a calibrated machine-learning anomaly classifier; Interpretation grounds each detected irregularity in semantic knowledge via a knowledge graph and retrieval-augmented generation; and Risk & Resolution computes a Risk Score—Consequence × Propagation × Urgency—kept strictly independent of the AI’s own confidence, and acts on the result only through human-governed resolution: an explicit decision matrix rather than a blended score. A synthetic, illustrative evaluation comparing this hybrid architecture against single-method baselines shows an F1 of 0.983 against 0.278 - 0.833, showing that an OR-combined detector recovers signal that no single channel could—under a generator deliberately constructed around independent channels, not as a general claim about hybrid detection on operational laboratory data. We conclude with the architecture’s implications, its limitations, and the empirical work required before clinical deployment.