Data-driven Optimization of Therapy for Heart Failure

DATA-DRIVEN OPTIMIZATION OF THERAPY FOR HEART FAILURE This program's moonshot goal is a computer model of the heart that can be personalized to a specific patient and deployed to optimize their therapy. The strategy's ultimate success will be determined by a future randomized clinical trial that tests whether patients treated with model-optimized therapies have better outcomes than patients who receive standard care. This application lays the foundation for the future trial. The investigators described this vision in a recent position paper titled “Closing the therapeutic loop” [23]. The manuscript describes how biomedical scientists have worked to define many of the molecular and cellular processes that regulate cardiovascular function. The paper also describes how advances in other areas of science can be leveraged to improve clinical care. Specifically, computer modeling, artificial intelligence, and machine learning have advanced to the point that they can be harnessed to test therapeutic strategies and optimize the treatment of patients who have heart failure. This project builds on these innovative ideas and integrates three aims that will advance translational cardiovascular science and move the field closer to a pivotal clinical trial. The team comprises 3 basic scientists and 2 cardiologists and is uniquely equipped to integrate physiological, computational, and clinical concepts. Aim 1: Develop the PyMyoVent framework as a testbed for implementing baroreflex control and myocardial growth. The PyMyoVent framework simulates a single ventricle pumping blood around a closed circulation [18]. The contraction algorithm mimics molecular biophysics but the ventricle has an idealized geometry. This allows the simulations to run in near real-time on a laptop. Specifically, the team will use PyMyoVent to develop and test algorithms that implement baroreflex control and myocardial growth. Specifically, the extended model will (1) regulate the rate and strength of contraction to maintain arterial pressure, and (2) allow the heart to change size (dilate/constrict and/or change thickness) in response to physiological demands. Aim 2: Create and validate patient-specific biventricular finite element models that incorporate growth and functional remodeling. This Aim uses clinical data collected over 6 months from 100 patients at the University of Kentucky who have advanced heart failure. Personalized biventricular finite element models [59] will be generated from echocardiographic and/or magnetic resonance imaging performed as part of standard care. The models will then be run forward in time to predict long-term changes in structure and function. The baroreflex and growth algorithms for the training set (60 patients) will be optimized using swarm intelligence. As recommended for prior studies of atrial fibrillation [49], data from another 20 patients will be used for model refinement. The remaining data (20 patients) will be used for final testing. The framework will be evaluated using a composite metric that compares the predictions for regional ventricular geometry (dimensions, thicknesses), ejection fraction, and arterial pressures to the corresponding clinical data at 6 months. Aim 3: Deploy the patient-specific models to Objective 1 Objective 2 Objective 3 predict optimal therapies for patients Algorithm Patient-specific Investigating who have advanced heart failure. development simulations therapies This Aim tests how potential interventions change PyMyoVent long-term growth and function by re-running the simulations from Aim 2 with different combinations + + of parameter adjustments. The interventions range from lowering the baroreflex setpoint Baroreflex Growth (mimicking blood pressure control) to modulating and functional myosin transition rates (mirroring the effects of the and new myotropes omecamtiv mecarbil [54] and remodeling mavacamten[3]). Random forest [35] and gradient Growth boosting techniques [10] will be used to rank the interventions by their ability to improve ejection Molecular biophysics, physiology, engineering fraction, myocardial strain, and myocardial energy use. Algorithm development Patient-specific modeling Clinical data and interpretation Figure 1: Overview of the project. Page 1 of 1