Domain knowledge may be used in a medical application to avoid wrong decisions, e.g., decisions raising contraindications. The contribution of this paper is twofold. First, it presents an approach for exploiting domain knowledge in a case-based decision support system in the domain of oncology. This approach is based on the so-called conservative adaptation that provides a solution necessarily consistent with the domain knowledge. Second, this paper describes an approach for the evolution of this domain knowledge when the expert rejects a proposed solution as being inconsistent with his/her knowledge. This inconsistency is characteristic of a difference between the domain knowledge of the system and the expert knowledge; from an interactive analysis, a piece of knowledge to be added to the domain knowledge is pointed out. This approach to domain knowledge evolution is implemented in a prototype called FrakaS.