A prediction-optimization approach to surgery prioritization in operating room scheduling

Abstract

ABSTRACT This study proposes a mixed-integer programming model to optimize daily schedules based on surgery priority. Stacking ensemble learning is employed to predict surgery priority. The stacking algorithm is composed of K-nearest neighbor, multi-nominal logistic regression, decision tree, multi-layer perceptron, and ensemble learning. Then, the predicted priorities are fed into an optimization model. Six patient-related variables are used to predict surgery priority: surgery type, patient acuity, patient age, number of delayed days a surgery is postponed, patient age, and surgery time. The study contribution comes from integrating machine learning and optimization to propose a priority-based decision model for optimally sequencing surgeries daily. The experimental results show that the proposed approach is better than the current practice in handling unscheduled surgeries, while the scheduling cost remains nearly unchanged. We show the effectiveness of the proposed approach for handling the surgery cancellation problem in operating room systems with high surgery demands. Graphical abstract