Readmission Prediction on MIMIC-IV
0.798AUC-ROCHyMaTE
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| HyMaTE2025.09 | 0.798 | 0.624 | — | — | — | — | |
| DuETT2025.09 | 0.772 | 0.592 | — | — | — | — | |
| Graph-WalkerBackbone=Qwen3-14B, Paradigm=Ours2026.04 | 0.7591 | 0.6061 | 58.01 | — | — | — | |
| EHR-Mamba2025.09 | 0.752 | 0.585 | — | — | — | — | |
| MUSE+AUMAUM-enhanced=true2026.01 | 0.7281 | 0.5003 | — | — | — | — | |
| DT-BEHRT2026.03 | 0.7208 | 0.8485 | 84.18 | — | — | — | |
| GRAPE+AUMAUM-enhanced=true2026.01 | 0.7197 | 0.4939 | — | — | — | — | |
| STraTS2025.09 | 0.718 | 0.544 | — | — | — | — | |
| MUSEAUM-enhanced=false2026.01 | 0.7152 | 0.467 | — | — | — | — | |
| GRAPEAUM-enhanced=false2026.01 | 0.7085 | 0.4551 | — | — | — | — | |
| M3CareAUM-enhanced=false2026.01 | 0.7067 | 0.4532 | — | — | — | — | |
| RePrompTBackbone=Llama 3.1 1B2026.04 | 0.706 | 0.728 | — | — | — | — | |
| HGMFAUM-enhanced=false2026.01 | 0.7005 | 0.4421 | — | — | — | — | |
| MTAUM-enhanced=false2026.01 | 0.6901 | 0.4375 | — | — | — | — | |
| SMILAUM-enhanced=false2026.01 | 0.6894 | 0.4368 | — | — | — | — | |
| HEART2026.03 | 0.6893 | 0.8207 | 83.68 | — | — | — | |
| Med-BERT2026.03 | 0.6851 | 0.8189 | 83.19 | — | — | — | |
| GraphCARE2026.06 | 0.685 | 0.696 | — | — | — | — | |
| Oracletraining_mode=trained on target data2025.06 | 0.6847 | 0.6937 | — | — | — | — | |
| ExBEHRT2026.03 | 0.6841 | 0.8186 | 83.21 | — | — | — | |
| Graph-WalkerBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Ours2026.04 | 0.6832 | 0.4637 | 51.54 | — | — | — | |
| CM-AEAUM-enhanced=false2026.01 | 0.6817 | 0.4324 | — | — | — | — | |
| UDONCAREtraining_mode=trained on source data2025.06 | 0.6726 | 0.6561 | — | — | — | — | |
| Prompt-basedType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 0.6702 | 0.6889 | 61.02 | 43.91 | 60.21 | — | |
| KAREType=LLM, Base Model=Meditron3-Qwen2.5-7B, Evaluation Protocol=Fine-tuned2026.04 | 0.6676 | 0.685 | 64.62 | 47.74 | 61.91 | — | |
| Standard RAGType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 0.6664 | 0.6851 | 60.95 | 43.84 | 59.98 | — | |
| K2KType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 0.6647 | 0.6867 | 63.75 | 46.79 | 61.42 | — | |
| GRAM2026.06 | 0.663 | 0.661 | — | — | — | — | |
| K2KType=LLM, Base Model=Meditron3-Qwen2.5-7B, Evaluation Protocol=Fine-tuned2026.04 | 0.6628 | 0.676 | 67.11 | 50.5 | 62.87 | — | |
| GRU2026.06 | 0.662 | 0.661 | — | — | — | — | |
| StageNet2026.06 | 0.662 | 0.661 | — | — | — | — | |
| HoT-SSMvariant=v12026.06 | 0.662 | 0.663 | — | — | — | — | |
| AdaCare2026.06 | 0.661 | 0.659 | — | — | — | — | |
| GRASP2026.06 | 0.661 | 0.663 | — | — | — | — | |
| HypEHR2026.03 | 0.6607 | 0.805 | 82.8 | — | — | — | |
| w/o retrieverType=LLM, Base Model=Meditron3-Qwen2.5-7B, Evaluation Protocol=Fine-tuned2026.04 | 0.6586 | 0.6776 | 56.92 | 39.79 | 57.58 | — | |
| w/o retrieverType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 0.6583 | 0.6786 | 61.3 | 44.2 | 59.8 | — | |
| Standard RAGType=LLM, Base Model=Meditron3-Qwen2.5-7B, Evaluation Protocol=Fine-tuned2026.04 | 0.6578 | 0.6768 | 65.09 | 48.25 | 61.7 | — | |
| KAREType=LLM, Base Model=BioMistral-7B, Evaluation Protocol=Fine-tuned2026.04 | 0.6544 | 0.6709 | 61.75 | 44.67 | 59.74 | — | |
| Deepr2026.06 | 0.654 | 0.656 | — | — | — | — | |
| G-BERT2026.03 | 0.6538 | 0.7149 | 82.13 | — | — | — | |
| Transformer2026.06 | 0.653 | 0.657 | — | — | — | — | |
| RETAIN2026.06 | 0.653 | 0.662 | — | — | — | — | |
| SLDGtraining_mode=trained on source data2025.06 | 0.6529 | 0.6155 | — | — | — | — | |
| Prompt-basedType=LLM, Base Model=Meditron3-Qwen2.5-7B, Evaluation Protocol=Fine-tuned2026.04 | 0.6484 | 0.6691 | 57.71 | 40.55 | 57.5 | — | |
| ManyDGtraining_mode=trained on source data2025.06 | 0.6407 | 0.6348 | — | — | — | — | |
| SeFT2025.09 | 0.638 | 0.377 | — | — | — | — | |
| EHR-embBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Data Perspective2026.04 | 0.6375 | 0.4632 | 49.81 | — | — | — | |
| HoT-SSMvariant=v22026.06 | 0.632 | 0.661 | — | — | — | — | |
| PCLtraining_mode=trained on source data2025.06 | 0.6304 | 0.6103 | — | — | — | — | |
| BEHRT2026.03 | 0.6232 | 0.7823 | 82.76 | — | — | — | |
| CondAdvtraining_mode=trained on source data2025.06 | 0.6229 | 0.6278 | — | — | — | — | |
| MLDGtraining_mode=trained on source data2025.06 | 0.6224 | 0.602 | — | — | — | — | |
| IRMtraining_mode=trained on source data2025.06 | 0.6184 | 0.6181 | — | — | — | — | |
| DANNtraining_mode=trained on source data2025.06 | 0.6122 | 0.6089 | — | — | — | — | |
| Basetraining_mode=trained on source data2025.06 | 0.6056 | 0.5846 | — | — | — | — | |
| COCONUTBackbone=Llama 3.1 1B2026.04 | 0.581 | 0.611 | — | — | — | — | |
| EHR-embBackbone=Qwen3-14B, Paradigm=Data Perspective2026.04 | 0.5779 | 0.3273 | 44.87 | — | — | — | |
| Prompt-TuningBackbone=Llama 3.1 1B2026.04 | 0.575 | 0.608 | — | — | — | — | |
| GRUType=ML2026.04 | 0.5658 | 0.5738 | 59.28 | 42.13 | 53.84 | — | |
| LMS3Backbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 0.5477 | 0.3095 | 45.41 | — | — | — | |
| Zero-shotBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Vanilla2026.04 | 0.544 | 0.2988 | 41.24 | — | — | — | |
| Time-seriesBackbone=Qwen3-14B, Paradigm=Data Perspective2026.04 | 0.5346 | 0.2697 | 40.55 | — | — | — | |
| Time-seriesBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Data Perspective2026.04 | 0.5345 | 0.2767 | 42.28 | — | — | — | |
| SFTBackbone=Qwen3-14B, Paradigm=Supervised2026.04 | 0.5323 | 0.2744 | 39.36 | — | — | — | |
| InfluenceBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 0.5285 | 0.2855 | 43.19 | — | — | — | |
| GradSelBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 0.5233 | 0.287 | 42.17 | — | — | — | |
| CONEBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 0.5167 | 0.2942 | 42.73 | — | — | — | |
| Delta-KNNBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 0.5123 | 0.265 | 41.13 | — | — | — | |
| Zero-ShotBackbone=Llama 3.1 1B2026.04 | 0.512 | 0.545 | — | — | — | — | |
| DeeprType=ML2026.04 | 0.5044 | 0.5227 | 68.13 | 51.66 | 55.62 | — | |
| AdaCareType=ML2026.04 | 0.5038 | 0.5212 | 47.96 | 31.54 | 45.5 | — | |
| RandomBackbone=Qwen3-14B, Paradigm=Vanilla2026.04 | 0.4998 | 0.2481 | 40.77 | — | — | — | |
| InfluenceBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 0.4964 | 0.284 | 42.65 | — | — | — | |
| TCNType=ML2026.04 | 0.4962 | 0.5133 | 53.32 | 36.35 | 47.66 | — | |
| RETAINType=ML2026.04 | 0.4961 | 0.5144 | 66.77 | 50.12 | 54.48 | — | |
| SPELLBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 0.4882 | 0.2437 | 40.03 | — | — | — | |
| StageNetType=ML2026.04 | 0.4867 | 0.5074 | 48.11 | 31.67 | 44.8 | — | |
| RandomBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Vanilla2026.04 | 0.4863 | 0.2706 | 39.85 | — | — | — | |
| SPELLBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 0.4811 | 0.2503 | 40.09 | — | — | — | |
| IDSBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 0.4741 | 0.2543 | 42.68 | — | — | — | |
| Delta-KNNBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 0.4711 | 0.2441 | 40 | — | — | — | |
| Semantic-embBackbone=Qwen3-14B, Paradigm=Data Perspective2026.04 | 0.4576 | 0.2775 | 41.89 | — | — | — | |
| CONEBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 0.4576 | 0.2775 | 41.89 | — | — | — | |
| GradSelBackbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 0.4573 | 0.2539 | 41.83 | — | — | — | |
| SFTBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Supervised2026.04 | 0.4537 | 0.2294 | 37.25 | — | — | — | |
| IDSBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Model Perspective2026.04 | 0.4525 | 0.2486 | 42.65 | — | — | — | |
| LMS3Backbone=Qwen3-14B, Paradigm=Model Perspective2026.04 | 0.4434 | 0.2468 | 42.28 | — | — | — | |
| Semantic-embBackbone=LLaMA-3.1-8B-Instruct, Paradigm=Data Perspective2026.04 | 0.4355 | 0.2369 | 42.37 | — | — | — | |
| Zero-shotBackbone=Qwen3-14B, Paradigm=Vanilla2026.04 | 0.4111 | 0.2711 | 41.66 | — | — | — | |
| Backbone onlyReasoning module=-2025.05 | — | — | 25.87 | — | — | 39.46 | |
| ProMedStrategy=S#1+GRPO2025.08 | — | — | — | — | — | 50.4 | |
| ProMedStrategy=S#1+22025.08 | — | — | — | — | — | 76.2 | |
| UniRReasoning module=Readmission2025.05 | — | — | 26.77 | — | — | 52.91 | |
| UniRReasoning module=Readmission + LOS2025.05 | — | — | 28.79 | — | — | 44.88 | |
| Vanilla2025.08 | — | — | — | — | — | 41.6 |