Clinical prediction on MIMIC-DOS
100Valid RateSingle-pass
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Single-passWorkflow=Single-pass, Local LLM=GPT-OSS2026.04 | 100 | 19.8 | 83.4 | 51.6 | 40.64 | 0.0415 | 1,155.47 | |
| ConfMADWorkflow=ConfMAD, Local LLMs=GPT-OSS + Qwen + LLaDA2026.04 | 100 | 33.6 | 66.2 | 49.9 | 47.16 | -0.0021 | 20,458.34 | |
| CAREWorkflow=CARE (ours), Local LLM=GPT-OSS, Proprietary LLM=GPT-52026.04 | 100 | 52.2 | 57 | 54.6 | 54.55 | 0.0921 | 7,771.52 | |
| CAREWorkflow=CARE (ours), Local LLM=Qwen, Proprietary LLM=GPT-52026.04 | 100 | 65.2 | 35.6 | 50.4 | 48.18 | 0.0084 | 7,168.53 | |
| CAREWorkflow=CARE (ours), Local LLM=LLaDA, Proprietary LLM=GPT-52026.04 | 99.9 | 62.32 | 36.6 | 49.46 | 47.76 | -0.0111 | 7,246.58 | |
| Single-passWorkflow=Single-pass, Local LLM=Qwen2026.04 | 99.8 | 38 | 60.04 | 49.02 | 47.77 | -0.0201 | 745.42 | |
| RSMADWorkflow=RSMAD, Local LLMs=GPT-OSS + Qwen + LLaDA2026.04 | 99.7 | 27.51 | 74.35 | 50.93 | 45.23 | 0.021 | 12,912.39 | |
| Majority votingWorkflow=Majority voting, Local LLMs=GPT-OSS + Qwen + LLaDA2026.04 | 99.4 | 42.89 | 55.96 | 49.1 | 46.97 | -0.0197 | 2,739.65 | |
| Single-passWorkflow=Single-pass, Local LLM=LLaDA2026.04 | 98.1 | 93.88 | 6.92 | 50.4 | 25.5 | 0.0162 | 834.67 |