Streaming EHR Clinical Action Prediction on MIMIC-IV (test)
59.86Medication Recall@5TRACE
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| TRACEBackbone=Llama-3.1-70B2026.02 | 59.86 | 41.76 | 41.84 | 3.56 | 92.8 | 9.42 | |
| TRACEBackbone=Meditron-70B2026.02 | 57.23 | 48.72 | 42.19 | 3.91 | 94 | 5.27 | |
| TRACEBackbone=Mixtral-8x22B-v0.12026.02 | 55.3 | 57.27 | 37.18 | 3.85 | 92.1 | 5.15 | |
| TRACEBackbone=Qwen-2.5-72B-Instruct2026.02 | 53.27 | 43.32 | 38.17 | 3.71 | 93.6 | 7.13 | |
| TRACEBackbone=DeepSeek-V3.12026.02 | 33.48 | 43.23 | 30.89 | 3.96 | 94.7 | 6.23 | |
| Monolithic AgentBackbone=MemGPT2026.02 | 32.24 | 13.6 | 33.13 | 3.28 | — | — | |
| Static RAGBackbone=Contriever2026.02 | 32.21 | 12.45 | 28.27 | 3.1 | — | — | |
| Long-ContextBackbone=Llama-3-70B2026.02 | 31.92 | 21.19 | 35.11 | 2.95 | — | — |