Spatial optimization of vaccine supply networks in complex terrains: balancing economic feasibility and public health equity

Background Vaccine supply planning in hazard-prone terrains faces compounding geographic, socioeconomic, and operational barriers. Conventional models relying on planar distances and static metrics often fail to capture true accessibility disparities, often driving inefficient infrastructure allocation and exacerbating health inequities. Purpose This study develops an integrated decision-making framework coupling spatial equity, economic feasibility, and multi-scenario resilience to systematically reconfigure the vaccine supply network across 183 counties in Sichuan, China. Methods Using high-resolution road network data from the AMap Application Programming Interface (API), we model actual geographic impedance based on travel time. An adapted two-step floating catchment area (2SFCA) method, embedded with socioeconomic vulnerability indicators, pinpoints baseline accessibility disparities. Candidate hubs are selected via entropy-weighted Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) combined with spatial-equity quotas, and a candidate-pool capacity-sufficiency check (α coefficient) determines the final candidate-pool size. A multi-objective mixed-integer linear programming (MO-MILP) model then co-optimizes hub configuration, capacity scaling, and cross-regional dispatch, endogenously balancing operational efficiency, equity penalties, and capacity expansion costs. Results Under entropy-weighted TOPSIS ranking with spatial-equity quotas, the initial 61-hub pool (including 21 existing CDC anchors) provides aggregate capacity equal to 82% of total demand. The capacity-sufficiency rule therefore expands the pool to C* = 103 hubs that satisfy Cap(C*) ≥ 1.2 × total demand. The optimized baseline network (103 candidate hubs, 99 with positive flows) achieves zero unmet demand through endogenous capacity expansion and single-source assignment, with a total objective value of 18,510,357. Sensitivity analysis across α = 1.0–1.4 shows that higher sufficiency coefficients increase the candidate pool (76 → 150 hubs) while reducing required capacity expansion by 63%. Fixed-topology resilience tests indicate zero shortages under all structural failure scenarios when emergency expansion is permitted, whereas zero-expansion stress tests bound shortage-free surge tolerance between +10% and +20% demand growth. Conclusion Integrating spatial equity and capacity sufficiency at the candidate-pool stage limits the risk that pool-level capacity deficits are later compensated through node-level expansion. The α coefficient serves as a planning parameter for balancing network redundancy and expansion needs. The α = 1.2 configuration provides a robust and equitable vaccine supply network for mountainous regions under the model’s cost assumptions.