Wave 1 · KHIDI RFP 11-1 [3-2] (p.61–64) · TRL 3 → 5
S-PUB
✓ PASS
Solo Reached: 87/100 ·
+CRO/Clinical: 95/100
0D–1D Cardiopulmonary RK4 Hemodynamic Digital Twin & Counterfactual Therapy Engine
Closed-loop 4-compartment RK4 cardiopulmonary Twin calibrated on 138 real PhysioNet MIMIC-IV ICU stays achieves 0.8392 AUROC for held-out hypotensive decompensation (vs 0.6536 static baseline) and reduces forward MAP RMSE from 12.78 to 8.44 mmHg with 3.00e-12 mL blood volume drift.
2. Solver-Free Recomputed Benchmark Metrics
Target: AUROC >= 0.78 & MAP RMSE < Static Windkessel with Volume Drift < 1e-5 mLHeld-Out ICU Hypotensive Instability (MAP < 65 mmHg) Prediction AUROC (n = 137)
0.8392 AUROC
95% CI [0.7604, 0.9145]
Baseline Comparison (Early Window Static Shock-Index / Linear Windkessel Baseline)
0.6536 AUROC
Cardiovascular Digital Twin
Held-Out Subsequent Window MAP Prediction RMSE (Closed-Loop RK4 Twin)
8.4449 mmHg
Baseline Comparison: Naive Static Linear Windkessel MAP RMSE
12.78 mmHg
Held-Out Arterial Pulse Pressure Prediction MAE
6.0484 mmHg
Closed-Loop 4-Compartment Blood Volume Conservation Residual
0 mL
RK4 Timestep Halving Convergence Error |MAP(2ms) - MAP(1ms)|
0.001 mmHg
Counterfactual HFrEF Vasodilator (-25% SVR) Stroke Volume Gain
5.363 mL/beat
Interactive Domain Simulation & Cohort Evidence Canvas
PASSbaselineEfPct
23.13
baselineMapMmHg
89.00
deltaSvMl
5.36
pcwpDropMmHg
-0.61
therapyEfPct
27.08
therapyMapMmHg
81.30
Mathematical / Biophysical Methodology & Upgrade vs APOSTOLUS v1
Integrates Suga-Sagawa time-varying LV elastance E_lv(t), nonlinear exponential EDPVR, diode aortic/mitral valves, 3-element arterial Windkessel (R_c, C_sa, R_svr), systemic venous reservoir, and pulmonary capillary wedge coupling via 4th-order Runge-Kutta (dt = 2 ms).
Advancement Over Legacy APOSTOLUS v1
Newly introduced flagship track replacing wet-lab animal regeneration RFP 1-1; solves closed-loop LV/RV time-varying elastance ODEs calibrated against real PhysioNet MIMIC-IV ICU hemodynamics.