Learning-Based Planning and Scheduling in Healthcare

Abstract

With countries such as the USA, Canada, and the UK devoting a substantial portion of their GDP to healthcare, the efficient allocation of expensive healthcare resources becomes crucial. Decision-making in healthcare involves intricate coordination and planning, amidst conflicting objectives and uncertainty. Operations research and deep learning provide promising solutions to healthcare decision-making problems, capable of enhancing performance metrics. This dissertation centers on three distinct healthcare operations management problems and introduces novel methodologies to address them.

Chapter 2 investigates staff scheduling in residential care facilities during pandemic conditions. Leveraging the COVID-19 pandemic as a case study, we utilize public health guidelines to derive state-of-the-art policies that help mitigate the spread of respiratory-prone diseases in residential care facilities. After formulating the staff scheduling problem as a mixed-integer programming model, we integrate the policies into the model as new objective functions and constraint sets. To achieve efficient solutions, we propose a population-based heuristic algorithm, which demonstrates superior performance when compared to two benchmark algorithms. Chapter 3 addresses a distributed multi-appointment, multi-resource ambulatory care scheduling problem that considers uncertain patient arrivals and emergency department utilization. The chapter introduces an approximate dynamic programming algorithm, enhanced by integrating a fully-connected neural network. Through numerical analyses, our proposed methodology demonstrates intelligent scheduling policies that efficiently minimize patient wait time, deferral, and transfer costs. This approach equips booking clerks with decision rules that are otherwise challenging to identify through trial and error in real-time. Chapter 4 introduces an appointment system for outpatient clinics, which tackles uncertainties in service duration, patient punctuality, and no-shows. We propose a learning-based optimization approach that integrates a predictive model, a convolutional neural network, into the objective function of a deterministic mixed-integer programming model. This model optimizes resource utilization, patient wait times, and uncertainty costs (estimated by the predictive model). Simulation results demonstrate that our proposed methodology provides practical and effective solutions.

These chapters make valuable contributions to healthcare operations management by offering novel approaches to address real-world challenges. The proposed algorithms aim to improve scheduling and resource allocation in healthcare settings, ultimately enhancing the overall healthcare operations.