Reproducibility

Every extraction query is public. Patient-level data is not — MIMIC-IV is credentialed-access, and this project treats that boundary as a hard rule, not a suggestion.

The pipeline

00
MIMIC-IV
01
Cohort
02
Variables
03
Dose extraction
04
Aggregation
05
State variants
06
Mechanism analysis
07
Evidence

Extraction queries

Five standalone files, safe to publish — code isn't data. Each tab below fetches the actual file live from this same site, not a copy-pasted excerpt — it can never silently go stale the way a hardcoded snippet can. See queries/README.md for the full run order and a substitution script that avoids the placeholder bugs this project hit early on.

loading…
Raw file

This site is an interface to a reproducible research project — not a replacement for the repository. Full source →

Browse all queries →

Data governance

MIMIC-IV requires PhysioNet credentialing. This repository publishes queries and aggregate-only results — never row-level or patient-level data, enforced by an explicit .gitignore policy, not just intention.

Safe to publish
SQL queries, analysis code, aggregate statistics reported in the manuscript
Never published
Any file with one row per patient or per decision-point — enforced by an explicit gitignore block, verified against full git history before every push
Full governance policy →

Getting your own access

  1. Complete CITI "Data or Specimens Only Research" training
  2. Apply for credentialed MIMIC-IV access via PhysioNet
  3. Request BigQuery access to the physionet-data project
  4. Run scripts/check_setup.sh to verify your environment before touching any query
← Evidence Papers →