eicu

An interpretable foundation model of the patient clinical timeline: event forecasting, calibrated time-to-event alerts, and concept-level interpretability, benchmarked head-to-head against tuned GBMs, tabular foundation models, and survival baselines on MIMIC-IV, eICU, and GEMINI via the MEDS standard.

OMOP standardization pipeline for ICU databases

Resources for the Society of Critical Care Medicine (SCCM) Datathon

REMed: Retrieval-Enhanced Medical prediction model

Materials for the BIDMC datathon (Sat 29 Feb 2020)

Early longer-stay risk for the renal cohort — a leakage-audited, calibrated, externally-validated clinical-ML case study (dbt + DuckDB + scikit-learn, MIMIC-IV + eICU).

Analysis code: heart-rate complexity (AC1 & Euler characteristic) as early-warning signals for septic shock — MIMIC-IV v3.1 discovery + eICU-CRD v2.0 external validation

Docker container building scripts for MIMIC-IV, MIMIC-III, and eICU.

Official codebase for LAADAN-AC: admissibility-aware offline actor-critic learning for sepsis treatment policies under benchmark-defined action constraints.

Reproducibility package for SAFE-CAP RL: paper policy files and methods code for CAP antibiotic-policy evaluation with local MIMIC-IV/eICU data.

Analysis code and de-identified local validation cohort for a GCS-free long-term mortality model in critically ill patients with acute brain injury

Site-aware temporal EHR validation for ARDS/AHRF using MIMIC-IV and eICU

GNRI + NLR prediction of postoperative ICU transfer in older orthopedic surgery patients (MIMIC-IV / eICU / INSPIRE) — analysis code, de-identified derived cohort, and EN/ZH web calculator.

Analysis code and public aggregate results for a baseline laboratory-based risk-ranking model for in-hospital mortality in rheumatoid arthritis (INSPIRE/MIMIC-IV/eICU); offline risk-ranking web calculator

Analysis code for NPSLE ICU phenotyping via latent class analysis with cross-database replication (MIMIC-IV, eICU, NWICU).

Multi-module clinical ML framework for real-time ICU mortality prediction, multimorbidity risk scoring, and waveform event detection with automated MLOps drift monitoring and ensemble fusion.

Shock-anchor-aware dynamic AMI-CS risk modelling across MIMIC-IV and eICU

Sharp performance bounds for early-warning evaluation with interval-uncertain event onset.

Analysis code for a three-database evaluation (MIMIC-IV, eICU-CRD, NWICU) of administrative neuropsychiatric phenotyping in critically ill SLE: diagnostic-confidence tiers, sepsis co-occurrence, a three-layer quantitative bias framework, and model transportability. Code only; no patient-level data (PhysioNet DUA).

Multicohort machine-learning workflow and SHAP calculator for early high-risk lactate-to-albumin ratio trajectories in sepsis.