Data-Driven Learning and Resource Allocation in Healthcare Operations Management

Abstract

The tremendous advances in machine learning and optimization over the past decade have immensely increased the opportunity to personalize and improve decisions for a plethora of problems in healthcare. This brings forward several challenges and opportunities that have been the primary motivation behind this dissertation and its contributions in both practical and theoretical aspects. This dissertation is broadly about sequential decision-making and statistical learning under limited resources. In this area, we treat sequentially arriving individuals, each of which should be assigned to the most appropriate resource. Per each arrival, the decision-maker receives some contextual information, chooses an action, and gains noisy feedback corresponding to the action. The aim is to minimize the regret of choosing sub-optimal actions over a time horizon. We provide data-driven and personalized methodologies for this class of problems. Our data-driven methods adaptively learn from data over time to make efficient and effective real-time decisions for each individual, when resources are limited. With a particular focus on high-impact problems in healthcare, we develop new online algorithms to solve healthcare operations problems. The theoretical contributions lie in the design and analysis of a new class of online learning algorithms for sequential decision-making and proving theoretical performance guarantees for them. The practical contributions are to apply our methodology to solve and provide managerial and practical insights for problems in healthcare, service operations, and operations management in general.