Multiscale model calibrated with a Bayesian algorithm predicts long-term outcomes of treatments for mitral valve regurgitation in animal models.

In primary mitral valve regurgitation (MR), a failing valve induces volume overload (VO) of the left ventricle (LV), triggering neurohormonal responses and myocardial remodeling. Surgical valve repair is the most effective therapy for MR, improving symptoms and in some patients reversing cardiac hypertrophy. However, at least 20% of patients develop postoperative impairment of LV function with detrimental effects on long-term outcomes, with no known cause. Many MR patients receive pharmacological treatment to alleviate symptoms, with drugs that alter cardiac function, hemodynamics, and molecular pathways that influence cardiac remodeling. To support the advancement of medical management of MR, we developed a multiscale model of cardiac hypertrophy that links ventricular mechanics, cardiovascular hemodynamics, and cardiomyocyte molecular signaling. We calibrated the model using the Markov chain Monte Carlo algorithm, a machine learning method, to integrate data from 76 studies of experimental VO in animals. The model's predictions showed good agreement with experimental data on MR in dogs, including the chronic effects of four drugs on MR-induced hypertrophy. When simulating valve repair, our model suggests that full reversal of LV hypertrophy requires both the mechanical elimination of VO and the restoration of normal neurohormonal receptor activity. Finally, we explored selecting subsets of a large pool of pre-computed simulations to personalize model predictions for a small population or individual where insufficient data are available to refit the entire model.