Research map / Reinforcement learning / Visual and representation RL
Medical Dead-ends and Learning to Identify High-Risk States and Treatments
Mehdi Fatemi, Taylor W. Killian, Jayakumar Subramanian, Marzyeh Ghassemi
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 DeD to flag potentially detrimental administered treatments | Septic patients in MIMIC-III · MIMIC-III septic patients | detrimental administered treatments identified 24 hours before death: more than $12\%$; $0.6\%$ false positive rate | Improved |
| Use DeD to identify unavoidable dead-end trajectories | Septic patients in MIMIC-III · MIMIC-III septic patients | non-surviving cases identified up to 48 hours before expiration: $2.7\%$; $0.2\%$ false positive rate | Improved |
| Use red-flag membership as a mortality-risk signal | DeD on septic patient trajectories · MIMIC-III survivors and nonsurvivors | red flags maintained for more than eight hours: 0.5\% of survivors; 32.6\% of nonsurvivors | Mixed |
| Use median Q values to flag states rather than maximum Q values | DeD · MIMIC-III septic patients | robustness of state flagging to extreme approximation error | Improved |
| Use looser thresholds for yellow flags | DeD · MIMIC-III septic patients | sensitivity of early deterioration flags: $\delta_{D}=-0.15$ and $\delta_{R}=0.85$ | Mixed |
| Use a separate rescue-value network alongside the dead-end-value network | DeD · MIMIC-III septic patients | confirmation of presumed dead-end states | Improved |
| Use value-function trends to detect deterioration | DeD · MIMIC-III septic patients | lead time before clinical interventions: 4 to 8 hours | Improved |
| Compare administered treatments with the 20% most secure treatments | DeD · MIMIC-III first-flag analysis | security of administered treatments: 20% most secure treatments (5 out of 25) | Mixed |
| Replace DeD's double-DQN learner with a contemporary offline RL method | DeD · MIMIC-III septic patients | prediction quality | No clear effect |
Closest papers by meaning
- Offline Guarded Safe Reinforcement Learning for Medical Treatment Optimization Strategies (NeurIPS 2025 · similarity 0.84)
- Benchmarking Reinforcement Learning Algorithms for ICU Ventilator Settings: An Interpretable and Probabilistic Patient Environment for Doctor Agents (AAAI 2026 · similarity 0.84)
- Contextualized Policy Recovery: Modeling and Interpreting Medical Decisions with Adaptive Imitation Learning (ICML 2024 · similarity 0.83)
- EpiCare: A Reinforcement Learning Benchmark for Dynamic Treatment Regimes (NeurIPS 2024 · similarity 0.83)
- Evaluating and Learning Optimal Dynamic Treatment Regimes under Truncation by Death (NeurIPS 2025 · similarity 0.82)
- Distribution-Free Uncertainty Quantification in Mechanical Ventilation Treatment: A Conformal Deep Q-Learning Framework (AAAI 2025 · similarity 0.82)
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.
