Team Twente's Hybrid AI Approach Takes Third in Healthcare Timetabling Competition
Published on November 10, 2025 at 05:00 AM
A team from the University of Twente, known as Team Twente, has achieved third place in the Integrated Healthcare Timetabling Competition (IHTC) 2024 with their innovative hybrid solution. The competition focused on finding efficient approaches to the integrated planning of patient admissions, room allocation, nurse assignment, and operating theater planning within healthcare systems.
Faced with the complexity of the integrated planning problem, especially for larger instances, Team Twente developed a three-phase decomposition approach. This approach combines Mixed-Integer Linear Programming (MILP), Constraint Programming (CP), and Simulated Annealing (SA). The method decomposes the problem into subproblems that are solved iteratively and in parallel.
The solution breaks down into: patient admission, deciding admission days; room assignment, allocating rooms for the patient's stay; operating theater assignment; and nurse assignment, assigning nurses to patient rooms during shifts.
Phase 1 optimizes initial patient admission and calculates lower bounds of combinations using information from other subproblems. Phase 2 uses a warm start based on computed lower bounds to iteratively solve patient-day admission, patient-room assignment, patient-theater assignment, and nurse-room assignment. Phase 3 improves nurse assignment via an exact approach after a heuristic approach in Phase 2.
The team's solution runs in parallel, utilizing all four available threads to maintain a pool of partial and complete solutions throughout the runtime. Whenever a feasible solution to a subproblem is found, it's used as input for the next subproblem.
The team also shared insights gained during the development of their solution and, for the first time, provided lower bounds on optimal solution values for the benchmark instances.
The implementation of the solution approach is available on GitHub. The code, written in Python, uses the MIP solver Gurobi 12 and the open-source CP solver from Google’s OR-Tools.
Researchers also noted that there is improvement potential in finding admission schedules that lead to feasible room assignments and improving the efficiency of the nurse assignment. Feedback mechanisms and cutting planes could also improve the nurse scheduling.