Relay race

Shared oncology case – AI compiles a report for the MDT from the medical records, citing the source for each value; the doctor confirms it, and the GP views the result.

The Relay is a shared oncology case that links the general practitioner (GP), the KOC coordinator, the KOC doctor and the multidisciplinary team (MDT). It addresses the VZP’s mandate ‘The patient’s journey without loss of information’: documents are not rewritten, it is clear who is responsible for what and by when, and the GP is informed of the MDT’s decision.

A 6-step example (fictitious patient):

1. The GP uploads a request; the Relay system pre-fills the patient’s details and diagnosis; the GP selects the KOC and confirms their consent.
2. The case coordinator takes over; the application lists, with citations, the reports mentioned in the request that are missing and requests them in a single step.
3. The uploaded reports are used to create 15 sections of supporting documentation for the MDT. Each value has a source and a verbatim citation; it is indicated what is missing, what is contradictory (e.g. the date of histology collection versus the date of issue) and what is out of date. The doctor responsible for the section will confirm, amend or leave it marked as missing.
4. The ‘Pancreatic cancer – diagnosis’ set creates steps with an owner, location and deadline; histology results past the deadline are escalated.
5. The documentation for the start of the MDT contains only the documents available at that time (subsequent reports are marked ‘After MDT’). The server assesses readiness; a doctor may waive a missing item, stating a reason. The minutes are downloaded as a DOCX file.
6. The doctor records the MDT conclusion; the coordinator sends it to the GP (an email with no clinical content, just a link); the GP opens it in the portal and everything is recorded in the audit trail.

Technology: InterSystems IRIS for Health as an FHIR R4 repository (Patient, EpisodeOfCare, Task, DocumentReference, Composition, Provenance, AuditEvent, etc.) and vector search in IRIS; backend: Python/FastAPI; frontend: React + TypeScript; LLM via Azure (EU).

Security: The AI merely suggests and cites; it does not make inferences or provide clinical recommendations; any result without a verifiable citation is rejected. The doctor makes the decision and confirms it; every confirmation is logged (who, when, what). Prototype, fictitious data, not intended for clinical decision-making.

Next steps: integration with real-world systems (HL7 v2/FHIR), authenticated login, an overview for the insurance company’s reviewing doctor, and validation across multiple cases.