Mortality Prediction on MIMIC-III (AUROC, AUPRC, F1)
84.19AUROCGraph-Walker
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Graph-WalkerBackbone=Qwen3-14B, Paradigm=Ours2026.04 | 84.19 | 59.46 | 58.24 | — | — | |
| Graph-WalkerBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Ours2026.04 | 76.71 | 35.09 | 50.13 | — | — | |
| SPELLBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 75.37 | 32.23 | 50.32 | — | — | |
| EHR-embBackbone=Qwen3-14B, Paradigm=Data Perspective2026.04 | 75.24 | 29.52 | 42.49 | — | — | |
| HoT-SSM (v1)spatial encoder=HConv2026.06 | 74.2 | 34.4 | — | — | — | |
| HoT-SSM (v2)spatial encoder=HGAT2026.06 | 73.2 | 32.8 | — | — | — | |
| GraphCARE2026.06 | 70.3 | 16.7 | — | — | — | |
| SFTBackbone=Qwen3-14B, Paradigm=Supervised2026.04 | 68.34 | 17.48 | 30.43 | — | — | |
| SFTBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Supervised2026.04 | 67.41 | 26.41 | 35.73 | — | — | |
| LMS3Backbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 66.54 | 26.92 | 39.07 | — | — | |
| RandomBackbone=Qwen3-14B, Paradigm=Vanilla2026.04 | 66.33 | 31.04 | 43 | — | — | |
| EHR-embBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Data Perspective2026.04 | 65.78 | 24.3 | 41.89 | — | — | |
| Zero-shotBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Vanilla2026.04 | 65.35 | 21.72 | 35.75 | — | — | |
| CONEBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 64.69 | 27.9 | 38.82 | — | — | |
| Semantic-embBackbone=Qwen3-14B, Paradigm=Data Perspective2026.04 | 64.43 | 21.78 | 33.6 | — | — | |
| Zero-shotBackbone=Qwen3-14B, Paradigm=Vanilla2026.04 | 64.09 | 17.58 | 31.62 | — | — | |
| Time-seriesBackbone=Qwen3-14B, Paradigm=Data Perspective2026.04 | 62.28 | 29.62 | 39.34 | — | — | |
| GradSelBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 62.09 | 32.2 | 36.68 | — | — | |
| InfluenceBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 61.73 | 31.8 | 38.39 | — | — | |
| GRPO——advanceType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 61.54 | 15.74 | 18.91 | 10.05 | 26.56 | |
| StageNet2026.06 | 61.5 | 12.4 | — | — | — | |
| GRU2026.06 | 61.3 | 11.8 | — | — | — | |
| K2KType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 61.05 | 15.22 | 18.55 | 10.22 | 26.26 | |
| Deepr2026.06 | 60.8 | 13.2 | — | — | — | |
| InfluenceBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 60.54 | 27.2 | 36.1 | — | — | |
| GRAM2026.06 | 60.4 | 11.4 | — | — | — | |
| IDSBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 59.89 | 28.4 | 36.11 | — | — | |
| w/o retrieverType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 59.4 | 11.61 | 16 | 8.69 | 23.92 | |
| RETAIN2026.06 | 59.4 | 9.6 | — | — | — | |
| IDSBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 59.14 | 19.73 | 32.97 | — | — | |
| Prompt-basedType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 58.72 | 10.78 | 15.05 | 8.13 | 23.17 | |
| GRPOType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 58.4 | 13.8 | 16.9 | 9.1 | 24.55 | |
| Delta-KNNBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 57.42 | 21.48 | 33.61 | — | — | |
| Transformer2026.06 | 57.2 | 10.1 | — | — | — | |
| Semantic-embBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Data Perspective2026.04 | 56.9 | 22.32 | 35.15 | — | — | |
| KAREType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 56.65 | 9.72 | 18.01 | 9.9 | 23.57 | |
| Time-seriesBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Data Perspective2026.04 | 56.38 | 24.86 | 37.16 | — | — | |
| KAREType=LLM, Base Model=Meditron3-Qwen2.5-7B, Evaluation Protocol=Fine-tuned2026.04 | 56.19 | 10.38 | 11.67 | 6.19 | 21.11 | |
| SPELLBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 55.43 | 20.92 | 34.51 | — | — | |
| RETAINType=ML2026.04 | 54.86 | 9.57 | 13.73 | 7.37 | 21.38 | |
| K2KType=LLM, Base Model=Meditron3-Qwen2.5-7B, Evaluation Protocol=Fine-tuned2026.04 | 54.56 | 10.29 | 17.27 | 9.45 | 22.89 | |
| Prompt-basedType=LLM, Base Model=Meditron3-Qwen2.5-7B, Evaluation Protocol=Fine-tuned2026.04 | 54.29 | 12.01 | 15.44 | 8.36 | 22.52 | |
| Standard RAGType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 54.19 | 9.34 | 11.94 | 6.34 | 20.45 | |
| Standard RAGType=LLM, Base Model=Meditron3-Qwen2.5-7B, Evaluation Protocol=Fine-tuned2026.04 | 54.11 | 8.43 | 16.35 | 8.9 | 21.95 | |
| GRUType=ML2026.04 | 53.5 | 8.03 | 13.87 | 7.45 | 20.71 | |
| RandomBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Vanilla2026.04 | 52.04 | 18.81 | 28.9 | — | — | |
| LMS3Backbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 51.8 | 16.41 | 29.37 | — | — | |
| DeeprType=ML2026.04 | 51.66 | 7.58 | 13.87 | 7.45 | 20.14 | |
| Delta-KNNBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 51.52 | 15.66 | 25.22 | — | — | |
| AdaCareType=ML2026.04 | 50.69 | 7.8 | 12.9 | 6.89 | 19.57 | |
| CONEBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 50.4 | 18.93 | 32.79 | — | — | |
| w/o retrieverType=LLM, Base Model=Meditron3-Qwen2.5-7B, Evaluation Protocol=Fine-tuned2026.04 | 48.9 | 7 | 13.78 | 7.4 | 19.27 | |
| GradSelBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 47.26 | 17.32 | 31.23 | — | — | |
| StageNetType=ML2026.04 | 47.14 | 7.1 | 9.97 | 5.25 | 17.37 | |
| TCNType=ML2026.04 | 45.81 | 6.76 | 11.28 | 5.97 | 17.45 | |
| RePrompTBackbone=Llama 3.1 1B2026.04 | 0.646 | 0.152 | — | — | — | |
| COCONUTBackbone=Llama 3.1 1B2026.04 | 0.612 | 0.136 | — | — | — | |
| Prompt-TuningBackbone=Llama 3.1 1B2026.04 | 0.611 | 0.132 | — | — | — | |
| Zero-ShotBackbone=Llama 3.1 1B2026.04 | 0.501 | 0.069 | — | — | — |