Medical researchers seek to identify and predict profit (or effectiveness) potential in a new medicine B against a specified disease by comparing it to an existing medicine A, which has been used to treat the disease for many years, called medicine assessment. Applying traditional data mining techniques to the medicine assessment, one can discover patterns, such as A.X=a à B.Y=b, which are identified at the attribute-value level. These patterns are useful in predicting associated behaviors at the attribute-value level. However, to evaluate B against A, we have to obtain globally useful relations between B and A at an attribute level. Therefore, this paper proposes a group interaction approach for multiple data source discovery. Group interactions include, such as rules, differences, and links between datasets. These group interactions are discovered at the attribute level. For example, R(A.X, B.Y), where R is a relationship, or a predication. Some examples are presented for illustrating the use of the group interaction approach.