ClinicalBench (Can LLMs Beat Traditional ML Models in Clinical Prediction?)

Fits teams tempted to use an LLM for risk prediction over structured records. Borrow the matched-cohort comparison against 11 traditional ML baselines and the rule that unparseable answers count as wrong.

What it asks the agent to do

• predict length of stay from an admission record
• predict in-hospital mortality
• predict readmission within two weeks

What to borrow from it

baselinegrading design

Caveats

This is the Chen et al. KDD 2026 ClinicalBench, the only benchmark by that name we could verify; prediction from prompts, not agentic; prompts must be rebuilt from PhysioNet credentialed MIMIC-III and MIMIC-IV; headline finding is that XGBoost and SVM beat every LLM, so it is a sanity check rather than a target.

Details

Links

• Paper
• Github