Extending the Medical Concept of Reference Intervals Using Fuzzy Predicates

Expert systems for medical applications have to deal with medical concepts such as "normal range", "elevated", or "reduced". These concepts, although backed by a profound medical background based on reference intervals, are defined manually by physicians using interval-based representation. This approach is usually not feasible in large- scale applications. In the present study we describe a method to generate fuzzy-logic-based predicates founded on historic medical data, using a combination of established statistical methods and cluster analyses to generate concepts that correspond to established laboratory standards and the physician\´s interpretation. We also describe visualization techniques which help the physician to analyze and adapt the results according to clinical needs. Finally, a case study using actual laboratory data from 562 hepatitis patients is presented.