Long wait times for elective surgery are a persistent challenge in publicly funded health systems, where hospitals must coordinate limited capacity before, during, and after the operation under considerable uncertainty. We study how a network of collaborating hospitals, such as the University Health Network in the City of Toronto, can centralize intake and jointly schedule same-day elective surgeries to improve access and efficiency under uncertainty. We formulate the problem using stochastic Mixed-Integer Programming (MIP) and Constraint Programming (CP) to capture the many linked resources used across a patient’s full surgical episode, from pre-operative preparation through the operating room (OR) to post-operative recovery. Network-wide completion times of surgical episodes are controlled by incorporating coherent risk measures in the objective function, such as the risk-averse Conditional Value-at-Risk (CVaR). To solve this large-scale problem, we develop logic-based Benders decomposition methods that integrate MIP and CP in two- and three-level schemes, strengthened with problem-specific cuts, valid inequalities, and a graph-based timing reformulation. Our experiments show that the decomposition design matters more than the choice of solver. The multi-level decomposition improves scalability, with the three-level approach giving the most reliable performance. We also find that uncertainty in ORs can propagate and intensify downstream, causing congestion that models limited to OR scheduling might miss. These results show the importance of risk-aware network-level scheduling, especially in publicly funded healthcare systems where capacity, efficiency, and equitable access must be balanced.