Background: Many cardiovascular 0D models replicate available data (e.g. pressure or flow) with reasonable accuracy but they cannot explain the underlying physiology. Such models may be overparameterized (i.e. overfit the data), yielding parameters and predictions that are not well determined. Models which are identifiable both replicate the data and have well-determined parameters and predictions. In the context of model personalisation, identifiability is crucial because estimated parameters should be identifiable if they are to be representative of the individual.
Methods: We collected structural (vessel radii/lengths) and functional (blood flow, time varying chamber volumes) cardiovascular ultrasound data for term and late preterm babies at two time points. These data inform personalised 0D closed-loop cardiovascular models. Models were parameterised using patient-specific data (e.g. arterial radii) and unknown parameters (e.g. terminal resistance and compliance) were optimised with a genetic algorithm. Models were reduced in an iterative manner until all estimated parameters were identifiable, as demonstrated by sensitivity analysis and MCMC.
Results: There were no differences in cardiovascular ultrasound measures between the term and preterm groups. The personalised models compared extremely well to clinical measurements, median error -0.1% (IQR 2.2, 1.4) for mean blood pressure, 1.8% (IQR 0.5, 7.4) for end-systolic volumes and -9.3% (IQR -19.8, -0.7) for end-diastolic volumes. Although estimated parameters were similar for the term and preterm group at birth, by three to six weeks of age, the preterm group had significantly higher lower body resistance compared to the term group.
Conclusions: We have demonstrated an automated model calibration workflow to produce practically identifiable 0D cardiovascular models and applied and validated these in a rich clinical dataset. Preterm babies developed greater vascular resistance and, as this parameter was shown to be practically identifiable, it is valid to use it to make inferences about the underlying physiology and suggests possible impaired vascular elastogenesis.