Mining Clinical Data using Minimal Predictive Rules

Modern hospitals and health-care institutes collect huge amounts of clinical data. Those who deal with such data know that there is a widening gap between data collection and data comprehension. Thus, it is very important to develop data mining techniques capable of automatically extracting useful knowledge to support clinical decision-making in various diagnostic and patient-management tasks. In this paper, we develop a new framework for rule mining based on minimal predictive rules (MPR). Our goal is to minimize the number of rules in order to reduce the information overhead, while preserving and concisely describing the important underlying patterns. We develop an algorithm to efficiently mine these MPRs and apply it to predict Heparin Platelet Factor 4 antibody (HPF4) test orders from electronic health records.