Data-quality and AI-readiness risk — hidden costs for EHR-based AI
Key Questions
What data quality issues affect AI readiness in EHR systems?
Between 40-60% noise or gaps in GP records create significant hidden costs and risks for EHR-based AI. Techniques such as RAG are recommended to reduce these issues.
How can benchmarks support evaluation of AI models for Irish EHR contexts?
The BRIDGE multilingual benchmark assesses LLMs on clinical text across nine languages and offers a tool for testing AI performance in Irish settings while addressing data quality and governance.
What solutions exist for structuring unstructured EHR data for AI use?
xCures has raised $46M to develop AI that converts unstructured EHR data into actionable intelligence from over 300 million records with source traceability, providing a relevant benchmark for improving AI readiness.
Clinical data quality, missing GP information, unstructured records and weak testing environments remain barriers to safe EHR-based AI. Recent EMR-AI analysis reinforces standardisation, data refinement and privacy safeguards, while the NHS evidence dispute underscores the need for provenance, validation and clinically supervised deployment rather than rapid experimentation.