Analysing Content of Patient-physician Messages with Self-organizing Maps

This paper is a part of an extensive research effort, in which our overall objective is to understand the impacts of patient-physician messaging service on provider efficiency and use of resources. In this poster presentation we evaluate the usefulness of Self Organizing Maps (SOM) technology for analysis of unstructured patient-originated electronic messages. A one-by-one hand analysis of a subset of the messages was performed in order to evaluate the accuracy of the topographic method. This study looks at two specific research questions: 1) What broad categories of requests are submitted by the patients 2) how can the requests be understood in the context of health care information provisioning by primary care physicians to patients with a specific chronic disease. Topographic mapping of message types confirmed the patterns seen through manual analysis. Sub-mapping of messages by symptom themes allowed more granular understanding of the clinical decision processes involved. SOM technology has potential for analyses of large data sets of unstructured messages for research purposes.