Treatment Outcome Classification on MIMIC-IV
98F1 ScoreL2D-Clinical
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| L2D-ClinicalDeferral strategy=Learned Deferral2026.04 | 98 | 98.6 | 16.8 | |
| Fixed θ = 0.95Confidence threshold (θ)=0.95, Deferral strategy=Confidence-based2026.04 | 97.1 | — | 26.2 | |
| GPT-5-nano onlyDeferral strategy=Always Defer (LLM Only)2026.04 | 96.7 | 96.8 | 100 | |
| ClinicalBERT onlyDeferral strategy=None (Base Model Only)2026.04 | 88.7 | 94.6 | 0 |