RESUMEN
OBJECTIVE: To evaluate a p-median model for health care services accessibility based on decentralization and optimal allocation of Primary Health Care Units in the State of São Paulo, Brazil. METHODS: Using geographical data of Primary Health Care Units located in the State of São Paulo, potential support and supply facility allocations were simulated by means of a random approach. Several constraints were then imposed on the system to simulate different scenarios. Results were assessed according to geographic disposition. RESULTS: Using a fixed number of supply facilities, ten as a constraint, the p-median approach allocated three facilities near the state capital (the area with the highest concentration of Primary Health Care Units), while remaining facilities were spread throughout the west of the state. A second round of tests assessed the impact of fixed costs alone on optimization, ranging from 71 optimal locations with a fixed unit cost to six optimal locations at a cost 300-fold higher. This finding was relevant to decision-making, since it encompassed scenarios in which only the final number of facilities or only the budget was known. A third set of simulations contemplates an intermediate scenario. CONCLUSION: The p-median approach was capable of optimizing a wide range of scenarios with an average running time of less than 2 hours and 30 minutes while considering a large dataset of more than 4,000 locations. In spite of some shortcomings, such as estimation of Euclidean distances, the method is simple yet powerful enough to be considered a useful tool to assist decision makers in the distribution of resources, and facilities across large areas with high number of locations to be supplied.