When a new patient does not comply with the guidelines, experts base their therapeutic decisions on similar patients' cases they have treated in the past. The case-based reasoning decision-support system we propose will automatically compare the patient's case to the structured guideline, then to other cases, and retrieve similar cases. The general structure of the system is presented here, the domain of application being oncology. As the patients' records are not stored in the current database in a format directly exploitable, an object-oriented model is proposed, that includes prognosis factors currently tested in clinical trials, well-established ones, and a description of the illness' episodes. Such system does not exist in the literature. It can be viewed as a data warehouse. Future work consists in defining the similarity measures, and connecting the system to the current database.