Mahal Systems CliBench
Fits coding, order-set or discharge-summary assistants that output sets of standard codes. Borrow the multi-granular scoring that maps predicted codes to ontology ancestors and reports precision, recall and F1 at each level.
What it asks the agent to do
- list discharge diagnoses as ICD-10-CM codes from an admission record
- identify procedures performed as ICD-10-PCS codes
- order the lab tests a clinician would order (LOINC)
- prescribe medications as ATC codes
What to borrow from it
grading designtaxonomy
Caveats
Zero-shot generation from a record, not agentic; the evaluation set (about 1,000 admissions per task) must be rebuilt from raw PhysioNet credentialed MIMIC-IV, nothing ships; the data license means models trained on it should be treated as containing sensitive data.
Details
| Workloads | healthcare, knowledge qa |
|---|---|
| Interaction | single turn |
| Environment | static |
| End state checked | No |
| Tools | none |
| Grading | programmatic check |
| Size | Not published |
| Data | real public |
| Personal data | possible |
| License | Apache-2.0 |
| Harness | own repo |
| Status | active |