In this paper we study the problem of applying data mining methods for prognosis analysis. We present a two-phase approach that first mines the entire dataset to discover a discriminating sub-dataset according to the classified outcome. Then the dataset is partitioned by the outcomes and is used to be further mined for another value to be predicted associating with the outcome. During each phase of classification and prediction, the mining process exploits an integrative framework that combines several methods. In data mining one of the key problems is selection of the most appropriate data-mining method. One advantage of the integrative framework that we propose is that the highest accuracy is always ensured, while any biased result can be eliminated, by using multi-attribute utility theory. We believe that such mining framework would be useful for medical prognosis.