Length of Stay Prediction (LOS) on MIMIC-IV (test)
81.88ROC AUCMERGE
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| MERGE2025.09 | 81.88 | — | 74.43 | — | — | — | |
| FuseMoE2025.09 | 81.74 | — | 75.18 | — | — | — | |
| ReToP_LlamaBackbone=Llama3-8B2026.01 | 80.62 | 69.05 | — | — | — | — | |
| ModernBERTMethod Category=Serialized Classifiers2026.01 | 80.4 | 68.42 | — | — | — | — | |
| ReToP_QwenBackbone=Qwen2.5-7B2026.01 | 80.38 | 68.91 | — | — | — | — | |
| REMedMethod Category=EHR-oriented models2026.01 | 80.15 | 68.06 | — | — | — | — | |
| I2MoE2025.09 | 79.88 | — | 74.36 | — | — | — | |
| pi_mrmr_dMethod Category=Rewrite-then-Predict models2026.01 | 79.48 | 68.53 | — | — | — | — | |
| self-genMethod Category=Rewrite-then-Predict models2026.01 | 79.02 | 67.04 | — | — | — | — | |
| CondIps (sample missing rate)=0.2, pw (within-sample missing rate)=0.22026.04 | 78.32 | — | 68.75 | — | — | — | |
| CondIps (sample missing rate)=0.8, pw (within-sample missing rate)=0.22026.04 | 78.32 | — | 64.14 | — | — | — | |
| PEPSYps (sample missing rate)=0.2, pw (within-sample missing rate)=0.22026.04 | 78.16 | — | 66.33 | — | — | — | |
| CondIps (sample missing rate)=0.2, pw (within-sample missing rate)=0.82026.04 | 77.68 | — | 71.35 | — | — | — | |
| CondIps (sample missing rate)=0.8, pw (within-sample missing rate)=0.82026.04 | 77.68 | — | 71.35 | — | — | — | |
| BioMedBERTMethod Category=Serialized Classifiers2026.01 | 77.59 | 64.42 | — | — | — | — | |
| pi_v_oMethod Category=Rewrite-then-Predict models2026.01 | 77.51 | 63.41 | — | — | — | — | |
| MISTS2025.09 | 77.49 | — | 73.86 | — | — | — | |
| mTAND2025.09 | 77.34 | — | 73.45 | — | — | — | |
| pi_rfe_dMethod Category=Rewrite-then-Predict models2026.01 | 76.84 | 63.22 | — | — | — | — | |
| MulT2025.09 | 76.68 | — | 72.52 | — | — | — | |
| pi_gu_edMethod Category=Rewrite-then-Predict models2026.01 | 76.13 | 60.57 | — | — | — | — | |
| FedMMps (sample missing rate)=0.2, pw (within-sample missing rate)=0.22026.04 | 76.08 | — | 69.91 | — | — | — | |
| Transformer2025.09 | 75.46 | — | 72.31 | — | — | — | |
| LlemrMethod Category=EHR-oriented models2026.01 | 75.36 | 62.21 | — | — | — | — | |
| LlamaMethod Category=Serialized Classifiers2026.01 | 75.09 | 62.42 | — | — | — | — | |
| PEPSYps (sample missing rate)=0.8, pw (within-sample missing rate)=0.22026.04 | 74.48 | — | 60.84 | — | — | — | |
| FedMMps (sample missing rate)=0.2, pw (within-sample missing rate)=0.82026.04 | 74.45 | — | 68.41 | — | — | — | |
| FedAvgps (sample missing rate)=0.2, pw (within-sample missing rate)=0.22026.04 | 74.23 | — | 66.82 | — | — | — | |
| FedAvgps (sample missing rate)=0.2, pw (within-sample missing rate)=0.82026.04 | 73.9 | — | 66.47 | — | — | — | |
| PEPSYps (sample missing rate)=0.2, pw (within-sample missing rate)=0.82026.04 | 73.86 | — | 67.3 | — | — | — | |
| PEPSYps (sample missing rate)=0.8, pw (within-sample missing rate)=0.82026.04 | 73.86 | — | 67.3 | — | — | — | |
| QwenMethod Category=Serialized Classifiers2026.01 | 73.78 | 60.75 | — | — | — | — | |
| FedMMps (sample missing rate)=0.8, pw (within-sample missing rate)=0.22026.04 | 73.51 | — | 67.02 | — | — | — | |
| pi_f_sMethod Category=Rewrite-then-Predict models2026.01 | 73.33 | 59.29 | — | — | — | — | |
| RNNMethod Category=EHR-oriented models2026.01 | 73.02 | 58.01 | — | — | — | — | |
| RETAINMethod Category=EHR-oriented models2026.01 | 72.91 | 58.21 | — | — | — | — | |
| FedMMps (sample missing rate)=0.8, pw (within-sample missing rate)=0.82026.04 | 71.9 | — | 34.5 | — | — | — | |
| FedAvgps (sample missing rate)=0.8, pw (within-sample missing rate)=0.22026.04 | 71.81 | — | 65.58 | — | — | — | |
| FedAvgps (sample missing rate)=0.8, pw (within-sample missing rate)=0.82026.04 | 71.77 | — | 65.71 | — | — | — | |
| GRASPMethod Category=EHR-oriented models2026.01 | 70.48 | 53.85 | — | — | — | — | |
| pi_h_kMethod Category=Rewrite-then-Predict models2026.01 | 70.04 | 54.74 | — | — | — | — | |
| pi_b_vMethod Category=Rewrite-then-Predict models2026.01 | 66.63 | 50.79 | — | — | — | — | |
| CTGANModel=GRU, Observation window=48h, View=CTGAN synthetic2026.03 | — | — | — | 63 | 37 | — | |
| Gaussian noiseModel=GRU, Observation window=48h, View=Gaussian noise (simple noise baseline)2026.03 | — | — | — | 65 | 40 | 100 | |
| RawModel=GRU, Observation window=48h, View=Raw (upper bound without privacy)2026.03 | — | — | — | 70 | 44 | 40 | |
| T1+T2Model=GRU, Observation window=48h, Privacy strength (α)=0.5, View=T1+T2@α = 0.52026.03 | — | — | — | 69 | 44 | 39 |