Chronic illness is characterized by a long response time during which many external influences can modify the system. Accurate characterization of such a system may require a large number of attributes. Such "data" complexity, in both the time and the feature domain, introduces a number of operational difficulties into the methodology for the scientific investigation of such systems. One such problem is the investigator's changing viewpoint and understanding of the system during the course of a study. Research data bases with a static attribute content are in conflict with this learning process and are of limited use when incorporated in investigation of chronic illness. Data element refinement is a tool for dealing with this problem, by providing a means for modifying an attribute's definition during the temporal life of a data base. We have been investigating means for supporting refinement of data elements in a large data base for coronary artery disease. For certain data types, we have developed a means of introducing "refined" views of particular data elements into a data base such that programs written prior to the refinement execute properly without modification during the lifetime of the data base. We will describe a data structure we have found suitable for capturing the information base from a broad range of clinical investigations, and our approach to the problem of data element refinement. Examples from ongoing clinical studies will be used to illustrate these ideas.