The objective of this Faculty Early Career Development (CAREER) Program award is to develop improved methods for optimization via simulation, and to apply them to decision-making problems in cardiovascular medicine. Optimization via simulation is required whenever we wish to find the best among several options, and evaluating the quality of an option requires running a stochastic simulation. Two types of improved methods will be developed. First, methods for ranking and selection will be developed using a link between Bayesian average-case and frequentist worst-case performance. These methods promise to produce solutions more quickly than do existing methods, with frequentist statistical guarantees on solution quality that are tighter and whose form is more natural for engineering applications. Second, multistart gradient-based methods will be developed using Bayesian value of information analysis to efficiently allocate simulation effort across starts, allowing those starts more likely to have high-quality local optima to converge first. These methods promise to produce high-quality solutions more quickly than do existing approaches for allocating simulation effort across starts. The new methods will be demonstrated via application to two problems in cardiovascular medicine: the design of post-operative surveillance strategies for patients undergoing endovascular aneurysm repair; and the design and placement of grafts in patients undergoing bypass surgery.
If successful, this research will both serve as an enabling technology within operations research, and will have specific benefits within cardiovascular medicine. Better post-operative surveillance will allow problems occurring after surgery to be detected more quickly, reducing the risk of aneurysm rupture. Better graft design will improve the reliability and longevity of surgical repairs, improving patient health. More broadly, as an enabling technology, the improved simulation optimization methods developed with this award will allow practitioners and researchers to solve a wider variety of optimization via simulation problems with greater speed and accuracy.
