An analytics-driven optimization framework for nurse scheduling under uncertainty

Nurse scheduling in hospitals is a complex optimization problem influenced by uncertain staff absences, regulatory constraints, and the need to ensure fair workload distribution. Traditional scheduling models often overlook the unpredictability of last-minute absences, leading to operational disruptions and compromised care quality. This study introduces an analytics-driven optimization framework that integrates predictive modeling with mathematical optimization to create more resilient nurse schedules. First, a mixed-integer linear programming (MILP) model incorporates historical absence data to compute shift-level criticality scores, enabling proactive reinforcement of high-risk shifts through targeted allocation of surplus nurses. Second, a real-time reassignment heuristic supported by a virtual overflow unit dynamically reallocates nurses during unforeseen absences to maintain coverage continuity. Third, a two-part hurdle model predicts shift-level absence probabilities, facilitating realistic robustness testing under simulated absenteeism scenarios. Using real hospital data, extensive computational experiments validate the framework’s effectiveness, demonstrating significant improvements in absence coverage and operational reliability compared to conventional scheduling methods. • Develop an analytics framework integrating predictive modeling and optimization for nurse scheduling. • Apply criticality-based prioritization to proactively reinforce high-risk shifts in scheduling. • Implement a real-time heuristic to dynamically reassign nurses during unforeseen absences. • Use a predictive hurdle model to simulate realistic absenteeism scenarios and test schedule robustness. • Validate the proposed framework using real hospital data and extensive computational experiments.