Experienced physicians and automatic generation of decision rules from clinical data

Abstract

Clinical Decision Support Systems embed data-driven decision models designed to represent clinical acumen of an experienced physician. We argue that eliminating physicians' diagnostic biases from data improves the overall quality of concepts, which we represent as decision rules. Experiments conducted on prospectively collected clinical data show that analyzing this filtered data produces rules with better coverage, certainty and confirmation. Cross-validation testing shows improvement in classification performance. © 2010 Springer-Verlag Berlin Heidelberg.