eCoordination

From 10 to 15 reports from seven centres, he will compile background material for the oncology consultation, citing the source for each piece of information, and highlight any omissions or contradictions.

A web application for the coordinator of a comprehensive cancer centre. It manages a single cancer case from the GP’s referral request through to the handover of the consultation report, in accordance with the four steps set out in the VZP guidelines.

The GP uploads their letter and reports (Word, PDF, RTF, text, photo or scan; text from images is recognised by OCR in the browser). Patient and doctor details are populated from the letter, with a source sentence indicated for each entry. The patient, along with their reports, will appear in the coordinator’s list.

The rules in the browser populate 20 fields for the consultation in accordance with the Handover_obsah template. Each value includes a verbatim quote and a link to the report. Missing or conflicting data are flagged, along with what needs to be verified and who to contact. The coordinator completes the details, records the verification, requests further information, or leaves it unverified with a reason.

The overview shows what is the coordinator’s responsibility, what is pending for others, the next consultation and any new reports. The workflow plan monitors the sequence of steps, deadlines and who is responsible for each task. The Consultations section sorts patients by consultation date, along with the status of the supporting documents. Everything is recorded in the coordination log.

The consultation documentation follows the structure of the MDT_zapis template, and a PDF can only be generated once all points have been resolved and confirmed by the doctor. The consultation conclusion is sent to the practitioner with the status ‘sent’, ‘delivered’ or ‘read’. Along with the conclusion, a summary is sent to the patient’s insurance company’s reviewing doctor, and the coordinator is notified of the doctor’s request for additional information as a task. The AI assistant answers questions about the data in the application and cites the source.

Test results on the test datasets compiled by our team: for Alena, 20 out of 20 items, 17 out of 17 test errors and 0 made-up values. For the second fictitious patient: 20 out of 20 and 13 out of 13 after the rules were adjusted (previously 18 out of 20 and 6 out of 13). A blind test on a third patient yielded 10 out of 20 items and 1 out of 4 errors.