Introduction:
Effective management of pulmonary arterial hypertension (PAH) requires understanding disease progression based on infrequent clinical measurements. Patient-specific digital twins could enhance decision-making by enabling disease trajectory prediction and in silico therapy evaluation. Physics-based cardiovascular models are a key component of these twins.
Detailed four-chamber cardiovascular models contain many unidentifiable parameters that are typically fixed to literature values, potentially inducing bias. Therefore, we investigate whether a simplified two-chamber model offers a better trade-off between parameter identifiability and predictive accuracy for PAH digital twin applications.
Methodology:
To estimate the parameters of our model, we use clinical indices from the ASPIRE registry data: cardiac output; mean, systolic, and diastolic pulmonary and systemic arterial pressures from right-heart catheterization; and systolic and diastolic ventricular volumes from cardiovascular magnetic resonance (CMR).
First, parameter identifiability is assessed using Bayesian inference with Markov chain Monte Carlo (MCMC) and profile likelihood analysis for different parameter and clinical index sets. Parameters are considered identifiable if, starting from uniform priors, their posterior contains the ground truth and has a standard deviation <10% of that value, and exhibits a finite, closed likelihood-ratio confidence interval.
Second, a global sensitivity analysis evaluates parameter influence on pulmonary arterial pressure predictions across varying heart rates. Parameters with a total Sobol index >0.2 for at least one heart rate are deemed influential.
Results:
Using synthetic two-chamber model data, we identify minimal and maximal ventricular elastances, systemic vascular resistance, and pulmonary vascular resistance and compliance as essential parameters. Their identifiability requires all considered clinical indices. Finally, we demonstrate the model's ability and limitations in reproducing pulmonary arterial pressure waveforms after calibration to synthetic data from a four-chamber model and ASPIRE clinical data across scenarios.
Conclusion:
The two-chamber model is a promising complementary approach to four-chamber-based workflows, retaining clinical relevance for well-defined tasks while improving parameter identifiability.