Introduction/Background: Despite numerous medical advances, 30% of patients with Fontan circulation require heart transplantation by 21. Current risk stratification relies on clinical metrics that lack strong correlations to patient outcomes, leading to a high waitlist mortality. Hypothesis: Integration of physics-based hemodynamic modeling with interpretable machine learning will improve prediction of patient outcomes (listing for heart transplant and/or cardiac death) by quantifying physiologic mechanisms inaccessible through clinical measures alone. Goals/Aims: 1. Develop a lumped-parameter mechanistic model (LPMM) of Fontan circulation 2. Identify patient clusters and predictors of patient outcomes 3. Validate model-informed classifiers against clinical-only approaches Methods/Approach: A LPMM incorporating ventricular sarcomere mechanics, atrial hemodynamics, and vascular viscoelastic effects was optimized to cardiac catheterization and echocardiography data from 51 patients (age 25±8 years). Thirty-eight parameters (ventricular inotropy, atrial stiffness, pulmonary elastance, etc.) were calibrated via surrogate optimization, minimizing least-squares error versus 11 clinical targets. Partial least-squares regression with leave-one-out cross-validation compared clinical-only versus model-informed classifiers of outcomes. Results/Data: The LPMM achieved 2% mean error across hemodynamic targets, with CVP and pulmonary pressures matching clinical measurements within 1%. Model-derived minimum atrial pressure (4.46±1.78 vs 2.68±0.87 mmHg, p<0.05) outperformed all clinical variables in discriminating patient outcomes. Integration of LPMM parameters increased AUROC from 0.67 to 0.78 for patient outcome prediction. Conclusions: Physics-based modeling identifies ventricular passive stiffness, atrial end-diastolic volume, and minimum atrial pressure, along with patient age (a clinical value) as critical determinants of Fontan failure. Combined mechanistic-machine learning approaches enable phenotype-specific risk stratification, supporting earlier referral for advanced therapies. This methodology establishes a framework for precision management of complex congenital circulations.
Schenk et al. (Fri,) studied this question.