Feature‐driven robust surgery scheduling

Abstract

Patient features such as gender, age, and underlying disease are crucial to improving the model fidelity of surgery duration. In this paper, we study a robust surgery scheduling problem augmented by patient feature segmentation. We focus on the surgery‐to‐operating room allocations for elective patients and future emergencies. Using feature data, we classify patients into different types using machine learning methods and characterize the uncertain surgery duration via a feature‐based cluster‐wise ambiguity set. We propose a feature‐driven adaptive robust optimization model that minimizes an overtime riskiness index, which helps mitigate both the magnitude and probability of working overtime. The model can be reformulated as a second‐order conic programming problem. From the reformulation, we find that minimizing the overtime riskiness index is equivalent to minimizing a Fano factor. This makes our robust optimization model easily interpretable to healthcare practitioners. To efficiently solve the problem, we develop a branch‐and‐cut algorithm and introduce symmetry‐breaking constraints. Numerical experiments demonstrate that our model outperforms benchmark models in a variety of performance metrics.