Bayesian parameter inference and uncertainty-informed sensitivity analysis in a 0D cardiovascular model for...

Area of Science:

• Computational physiology and biomedical engineering.
• Application of advanced statistical methods in healthcare.
• Clinical decision support systems.

Background:

• Computational cardiovascular models aid clinical decisions, especially for intraoperative hypotension (IOH).
• Traditional calibration methods struggle with parameter non-identifiability and lack uncertainty quantification, limiting clinical use.
• Bayesian approaches offer a solution for parameter inference, sensitivity analysis, and uncertainty quantification.

Purpose of the Study:

• To apply Bayesian Markov chain Monte Carlo (MCMC) for parameter estimation in cardiovascular lumped parameter models (LPMs) for IOH scenarios.
• To demonstrate the impact of parameter non-uniqueness on sensitivity and improve parameter reliability.
• To introduce uncertainty-aware sensitivity analysis and compare it with classical methods for enhanced clinical utility.

Main Methods:

• Utilized Bayesian Markov chain Monte Carlo (MCMC) for parameter estimation of a cardiovascular lumped parameter model (LPM).
• Incorporated clinical knowledge and measurement uncertainties to enhance parameter reliability.
• Implemented sequential parameter updating for continual model learning and uncertainty-aware sensitivity analysis.

Main Results:

• MCMC distinguished between different IOH scenarios (e.g., impaired contractility vs. hypovolemia), unlike classical optimization.
• Parameter uncertainty reduced significantly with additional data (70%) and sequential updating (94%).
• Uncertainty-aware sensitivity analysis yielded more stable parameter rankings and tighter credible intervals than classical approaches.

Conclusions:

• Bayesian inference with sequential updating and sensitivity analysis improves LPM reliability and identifiability.
• This approach enhances clinical utility for therapy guidance in IOH and similar complex conditions.
• The Bayesian framework provides more stable parameter estimates and insights into model sensitivity across different patient states.