Research map / Large language models / Language, safety and interpretability
REACT-LLM: A Benchmark for Evaluating LLM Integration with Causal Features in Clinical Prognostic Tasks
Linna Wang, Zhixuan You, Qihui Zhang, Jiunan Wen, Ji Shi, Yimin Chen, Yusen Wang, Fanqi Ding, Ziliang Feng, Li Lu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use LLMs instead of traditional ML models | LLM direct prompting · 7 clinical outcomes, MIMIC-III/IV | AUROC/AUPRC: LLMs lag ML baselines by about 10-20% | Hurt |
| use Gemini-2-Flash instead of best ML model on EarlyICU | Gemini-2-Flash · MIMIC-III EarlyICU | AUPRC: 0.8473 -> 0.8540 | Improved |
| add in-context learning prompting (ablation) | LLMs · 7 clinical outcomes | AUROC/AUPRC/F1: surpasses baseline 41 times, best in 22 | Improved |
| add self-reflection prompting (ablation) | LLMs · 7 clinical outcomes | AUROC/AUPRC/F1: exceeds baseline in 38 cases, leads in 19 | Improved |
| add ICL to Gemini-2-Pro on Readmit30 (ablation) | Gemini-2-Pro · Readmit30 | AUROC: +0.077 | Improved |
| add CoT to DeepSeek-R1 (ablation) | DeepSeek-R1 · Readmit30 | AUROC: +0.0912 AUROC, +0.0271 AUPRC | Improved |
| add any prompting strategy (ablation) | LLMs · 7 clinical outcomes | AUROC | No clear effect |
| use JSON instead of RCF input format (ablation) | Qwen3-235b · MIMIC-III | AUROC/AUPRC/F1: up to +0.064 | Improved |
| use narrative serialization instead of RCF input format (ablation) | Gemini-2-Flash · MIMIC-III | AUROC/AUPRC: +0.051 AUROC, +0.026 AUPRC | Improved |
| use RCF format instead of NS or TBNL (ablation) | GPT-o3-mini · MIMIC-III | AUROC/AUPRC/F1 | Improved |
| restrict input to causal-discovery features (DirectLiNGAM) (ablation) | LLMs · MIMIC-III and MIMIC-IV | AUROC | Hurt |
| restrict input to GES causal features (ablation) | DeepSeek-R1, Qwen3-8b, Qwen3-235b · MIMIC-III and MIMIC-IV | AUROC | Improved |
Closest papers by meaning
- LLM-Driven Treatment Effect Estimation Under Inference Time Text Confounding (NeurIPS 2025 · similarity 0.87)
- Risk Horizons: Structured Hypothesis Spaces for Longitudinal Clinical Prediction (NeurIPS 2026 · similarity 0.87)
- Reasoning-Enhanced Healthcare Predictions with Knowledge Graph Community Retrieval (ICLR 2025 · similarity 0.86)
- End-To-End Causal Effect Estimation from Unstructured Natural Language Data (NeurIPS 2024 · similarity 0.86)
- MedJourney: Benchmark and Evaluation of Large Language Models over Patient Clinical Journey (NeurIPS 2024 · similarity 0.85)
- Memorize and Rank: Elevating Large Language Models for Clinical Diagnosis Prediction (AAAI 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
