Sepsis Resuscitation Cohort

This Zenodo record hosts a free evaluation sample of the full dataset — data only, not the generator, and not the complete dataset. The full package (10,000 patients, 21,630,000 hemodynamic profile rows, plus the complete generator source) is available at: https://sentineldata.com.ua/dataset/sepsis-resuscitation-cohort Sample contents: 100 patients (SEPSIS_00000–SEPSIS_00099) × 3 counterfactual arms = 300 regimens, 216,300 hemodynamic profile rows, across 4 Apache Parquet files (zstd), 9.3 MB total. --- Dataset Architecture & Specifications The Sepsis Resuscitation Cohort ships as both a ready-to-use dataset and the complete generator that built it. Ten thousand septic patients are each run through three resuscitation strategies — liberal, conservative, and a phenotype-aware oracle arm — starting from one shared underlying physiology, so the same patient's outcome can be compared directly across arms. All 66 model parameters carry a provenance tier: 26 are verified against a peer-reviewed source located and checked during production, the remaining 40 are estimates anchored to a documented calibration target. Nothing in the model is an unsourced guess. Synthetic Patients: 10,000 (SEPSIS_00000–SEPSIS_09999) Counterfactual Arms: 3 per patient — liberal, conservative, phenotype_guided (30,000 regimens total) Hemodynamic Profile Rows: 21,630,000 (721 points/regimen, 0–72 h at 6-min ... Full description with all tables is on the Wiki page of this project. Full dataset and purchase: https://sentineldata.com.ua/dataset/sepsis-resuscitation-cohort-generator . Zenodo copy: https://doi.org/10.5281/zenodo.22547576