Global sensitivity analysis through causal discovery for an electromechanical cardiac model

Area of Science:

• Computational Biology
• Biomedical Engineering
• Cardiovascular Research

Background:

• Advanced cardiac imaging and modeling increase complexity, necessitating understanding parameter-output relationships.
• Personalizing cardiac models for specific pathologies requires robust sensitivity analysis.

Purpose of the Study:

• To develop an interpretable global sensitivity analysis method for electromechanical cardiac models using causal discovery.
• To quantify the joint effect of model parameters on clinical biomarkers across different cardiac geometries (healthy, hypertrophic cardiomyopathy, dilated cardiomyopathy).

Main Methods:

• Leveraging causal discovery to perform multi-output global sensitivity analysis on electromechanical cardiac models.
• Comparing parameter-biomarker relationships in healthy hearts versus hypertrophic cardiomyopathy and dilated cardiomyopathy.
• Utilizing an additive noise model (ANM) for validation.

Main Results:

• Causal discovery identified geometry-dependent sensitivities and a reduced set of influential parameters for cardiac biomarkers.
• The approach provided stable and interpretable results, outperforming classical methods like Sobol and Pawn, especially with limited simulations.
• Parameter-biomarker relationships were explored for ejection fraction, max(dP/dt), isovolumic relaxation time, and early passive filling.

Conclusions:

• Causal discovery offers a powerful, reliable alternative for sensitivity analysis in complex cardiac models, particularly in data-limited scenarios.
• The findings provide actionable guidance for model calibration and pathology-informed personalization.
• The workflow is publicly available to advance cardiac modeling research.