Use of expert systems in medical research data analysis : The POSCH AI project

Artificial Intelligence (AI), especially expert systems technology, has very promising possibilities for more fully automating the analysis of medical research data. The Program on the Surgical Control of the Hyperlipidemias (POSCH), a national multi-centered clinical trial has been experimenting with using AI in its data analyses for several years. Three projects are described in this paper. The first experiment, designed to automate the clinical judgment used to evaluate the data from serial graded electrocardiograms has been a success. Early efforts to automate one step of the evaluation of serial coronary arteriography data has also been successful, but major difficulties must be overcome to extend the work. The initial motivation for the POSCH AI Project was to build an expert system that can search for interesting relationships among, and between, the POSCH variables. The problems encountered are different from those of other investigators who are attempting similar projects. Efforts to segment and/or collapse the problem into smaller units are explained. Issues related to the data structure and query languages have been identified. Statistical and other logically based reasoning methods, fundamental to the entire project, are discussed in conjunction with the heuristic methods used by expert systems. * The work of J.R. Slagel, M.R. Wick, and Erach Irani is partially supported by NSF Grant DCR8512857 and by the Microelectronics and Information Sciences Center of the University of Minnesota.