Abstract

A novel methodological approach for identifying clusters of similar medical incidents by analyzing large databases of incident reports is described. The discovery of similar events allows the identification of patterns and trends, and makes possible the prediction of future events and the establishment of barriers and best practices. Two techniques from the fields of information science and artificial intelligence have been integrated—namely, case based reasoning and information retrieval—and very good clustering accuracies have been achieved on a test data set of incident reports from transfusion medicine. This work suggests that clustering should integrate the features of an incident captured in traditional form based records together with the detailed information found in the narrative included in event reports.