IMPROVING HOSPITAL DISCHARGES AND CARE TRANSITIONS: PREDICTIVE AND OPTIMIZATION INSIGHTS

Abstract

Improving discharge planning and long-term care capacity is essential to reducing delayed discharges and enhancing healthcare system efficiency. Delayed discharge patients, particularly those designated as Alternate Level of Care, contribute significantly to hospital overcrowding and poor patient outcomes. This thesis addresses these challenges through two interconnected themes: developing tools to predict patients at risk of delayed discharge and their expected length of stay, and designing mathematical models to optimize long-term care capacity and patient flow. The first theme begins with a systematic literature review that classifies existing studies based on the statistical and machine learning methods used to predict discharge destination, length of stay, and discharge volume. Identified gaps in discharge-related predictions motivate the development of data-driven models. The second and third studies introduce machine learning–based decision support tools built on patient health records from Nova Scotia. The second study predicts Alternate Level of Care designation at the time of admission, enabling early identification and intervention. The third study estimates the length of stay for Alternate Level of Care patients using key predictors such as discharge destination, patient service type, and season, offering insights to understand complex discharges. In the second theme, the fourth study presents a multi-period mathematical optimization framework for long-term care capacity expansion and patient assignment. Two models are proposed: one for planning expansions and assignments, and another that incorporates interfacility transfers. Applied to a real case study, results show that incorporating interfacility transfers reduces system costs while improving flexibility and equity in resource allocation. Additionally, column generation and rolling time horizon algorithms are developed to enhance scalability. Together, this thesis offers integrated predictive and prescriptive tools, built on real-world data, to support timely patient discharges and inform strategic planning for long-term care across health systems.