Wave 2 · KHIDI RFP 11-1 [6-1] (p.61–64) · TRL 3 → 5
S-PUB
✓ PASS
Solo Reached: 85/100 ·
+CRO/Clinical: 93/100
Multi-Compartment Pulmonary V/Q Shunt Ventilator Twin & Real MIMIC-IV ICU Cohort Stratifier
MIMIC-IV First-24h Subject-Locked ICU Benchmark (Holdout AUROC 0.744 vs Shock-Index 0.655) + 3-Compartment V/Q Shunt Ventilator Twin
2. Solver-Free Recomputed Benchmark Metrics
Target: Zero post-24h timestamp leakage; beats admission shock-index baselineMIMIC-IV First-24h Subject-Locked ICU Deterioration AUROC (n = 66)
0.7442 AUROC
95% CI [0.6057, 0.8615]
Baseline Comparison (Single-Vital Admission Shock Index (HR/SBP))
0.6552 AUROC
ICU Sepsis-ARDS Severity & Prognosis
MIMIC-IV Holdout Probability Calibration Brier Score
0.1825 Brier
MIMIC-IV Holdout Expected Calibration Error (10-Bin)
0.1086 ECE
Pulmonary Shunt O2 Flux Conservation Max Residual
0 mL/dL
Post-24h Chart/Lab Events Strictly Excluded by Temporal Lock
58619 events
Verified First-24h ICU Chart & Lab Events Integrated
22931 events
Interactive Domain Simulation & Cohort Evidence Canvas
PASSfinalPfRatioMmHg
248.00
finalRiskProb
0.27
finalShuntPct
16.90
initialPfRatioMmHg
136.00
maxDrivingPressureCmH2o
13.80
Mathematical / Biophysical Methodology & Upgrade vs APOSTOLUS v1
Integrates an exact 3-compartment alveolar gas & Severinghaus-Hill pulmonary shunt solver (with ΔP > 15 cmH2O barotrauma interlock) and strict <=24h timestamp-locked feature extraction across 140 real Beth Israel Deaconess ICU stays.
Advancement Over Legacy APOSTOLUS v1
Replaces Phase I vaccine trial RFP 10-1; pairs West-zone pulmonary shunt & driving-pressure mechanics with real Beth Israel Deaconess ICU stays, labs, and chartevents (MIMIC-IV Demo 2.2).