All hematological entities that are typically discussed at the tumor boards, for example malignant lymphomas, plasmacytoma/multiple myeloma, leukemias, other malignant neoplasms of lymphoid/hematopoietic tissue, as well as polycythemia vera, myelodysplastic syndromes, and other myeloproliferative neoplasms. C81-C96
Conditions
Interventions
Group 1: Therapy recommendation generated by the LLM agent HemaGuide based on the original tumor board input data. Blinded to the clinical tumor board
no clinical intervention.
Group 2: Documented therapy decision of the regular hematology tumor boards, as documented at the time of the original patient care and recorded in the hospital information s
Sponsors
Universitätsklinikum Heidelberg, Klinik für Innere Medizin V: Hämatologie, Onkologie und Rheumatologie
Eligibility
Sex/Gender
All
Age
18 Years to No maximum
Inclusion criteria
Inclusion criteria: Patients whose hematological-oncological case was presented and discussed at least once at a hematology tumor board between 2024 and 2025 and for whom a documented tumor board therapy recommendation is available.
Exclusion criteria
Exclusion criteria: Excluded were tumor board cases in which the primary question concerned radiological image interpretation (e.g., pure restaging presentations without a resulting therapy decision), cases with incomplete or non-reconstructible documentation of the tumor board decision, cases without a documented therapy recommendation, as well as cases with incomplete clinical findings required for a reproducible therapy decision.
Design outcomes
Primary
| Measure | Time frame |
|---|---|
| Proportion of cases (%) in which the recommendation generated by the LLM agent agrees with the documented tumor board decision (reference standard). For this, the LLM agent is applied to de-identified tumor board cases in routine clinical care. Agreement is assessed per case using a specifically developed Tumor Board Concordance Score (TBCS, 10 points); concordance is defined as a total score = 8/10. Assessment is performed by two independent hematologists - with a third independent rater adjudicating in cases of disagreement. The concordance rate is calculated across all included cases. | — |
Secondary
| Measure | Time frame |
|---|---|
| 1. Reduction in hallucination rate across all evaluated cases. 2. Classification of discordant cases into clinically interpretable categories. 3. Stratification by guideline conformity, patient-context integration, reasoning transparency, clinical utility, and run-to-run consistency. 4. Median and p95 inference time and token count per case on standard hardware, stratified by decision mode. | — |
Countries
Germany
Contacts
Public ContactMirco Friedrich
Universitätsklinikum Heidelberg, Klinik für Innere Medizin V: Hämatologie, Onkologie und Rheumatologie
Outcome results
None listed