In this research we present an architectural design model of building an enterprise integrated database by using the approach of partial replication of data to extract and aggregate the databases dispersed at different locations to effectively provide the required information for business management and decision making. We take the design and creation of an experimental central database for National Immunization Information System (NIIS), Center for Disease Control (CDC) of Department of Health (DOH), as a case of practical study to investigate how an integrated central database can be built by making use of the distributed database located at each county's and city's health bureau so that the capabilities of building decision-making applications and the administration for the operations of national immunization can be supported. In order to verify the true value of being able to more effectively support NIIS decision support through the implementation of central database, in this research we also attempt to create a prediction model for yearly national vaccine procurement as a basis for CDC to purchase the best-fit amount of vary types of vaccine so that the current approach of using human experience to compute the yearly vaccine purchasing amount can be substituted. In this study of building the application of decision making, in the first stage we use Grey Prediction Theory and Back-Propagation Neural Network separately to build the prediction model to predict the number of immunization population for the next year based on the major factors of the number of immunization, population and the completion rate, and the objective immunization population of next year of each city's bureau of health. In the second stage, we take the prediction result of the next-year from the first stage and further consider the yearly vaccine waste rage, the amount of reservation stock, and last year stock amount to compute the amount of yearly national vaccine procurement. After completing the experimental system building of central database system, the experimental result shows that taking the data resources from the central database to build the prediction model for the yearly national vaccine procurement which can be more accurately to compute the next year's vaccine procurement, and it also can efficiently generate the information and report for the management statistical analysis.