You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Auditable EHR data infrastructure for patient world models and clinical agents: source-linked conversion of hospital exports into OMOP CDM 5.4 and MEDS, with 55 checks that read the published artifacts
Early warning for sepsis with a calibrated alert budget. It states how often it will alert and how often it will be right, then checks both on a hospital it never trained on. Open data, no credentials, every number reproducible.
Open, reproducible implementation of the CMS Hospital-Wide Readmission (HWR) measure on MIMIC-IV: faithful 30-day unplanned-readmission labels, a calibrated gradient-boosted baseline with fairness and drift monitoring, and a leakage-controlled test of whether clinical-note LLM embeddings add signal.
Independent research project testing how reliable feature-importance rankings are in machine learning models that predict cognitive impairment from sleep recordings.
SAIL: detect state-action information leakage in offline reinforcement learning for sepsis — construction, reconstruction, temporal overlap, timing and persistence checks.
Retrospective research on risk within NHAMCS emergency triage strata. Bilingual overview, reproducible notebook and later-year holdout; not validated for clinical use.
Retrospective feasibility study and research pipeline for short-horizon ICU hypotension forecasting using ABP, ECG, PPG, and EHR context, with leakage-aware evaluation and sensor-missingness robustness tests.
Selective prediction for fetal ultrasound plane classification by pairing a ResNet-18 with diffusion reconstruction uncertainty and softmax confidence for calibrated human-AI deferral. Includes an honest ablation of uncertainty methods.
ICU mortality on the PhysioNet/CinC Challenge 2012: XGBoost on summary features vs LSTM, GRU and GRU-D on irregular hourly sequences, tested on the hidden set C against the published challenge scores; trees win (AUROC 0.882 vs 0.871) and measurement timing alone reaches 0.79.
A-ICF: Auditing, Not Predicting — A Causal Bias-Decomposition Framework for Clinical Fairness. Code, Figures, and Tables for OMLET 2026 (Paper ID: 596).