Skip to content

Data, testing and AI evaluation

  • partial National UK synthetic patient population: NHS England's UK Synthea adaptation covers primary care in the South West, Simulacrum covers cancer, and there is PDS test data. There is no maintained synthetic population covering all care settings, including hospitals, in UK Core FHIR with SNOMED CT and dm+d codes.
  • missing Open UK Core conformance test suite: validators and examples exist, but there is no shared, automated conformance suite for UK Core and the national APIs, so each supplier proves conformance its own way.
  • partial Open de-identification for UK clinical text: open tools exist, such as CRATE, but there is no maintained tool tuned for UK records with a shared corpus to measure it against.
  • restricted Openly licensed UK clinical language models and corpus: open-source clinical NLP tools exist, but SNOMED CT licensing restricts their models, and there is no openly licensed corpus of UK clinical text.
  • missing Open evaluation datasets for ambient voice technology: NHS England keeps a self-certified supplier registry for ambient scribes, but there is no consented test set of UK consultations, in UK accents, and no shared way to compare products' accuracy and omissions.