Automatic adaptation of information in electronic patient records for patients

Abstract

In this work we present results of research, design and implementation of prototype system aimed at the task of automatically identifying and explaining difficult medical words in Electronic Patient Records. A patient record is often a jungle of medical words, ordinary and irregular abbreviations and acronyms. Patient records are usually entered in a high tempo and have a lot of spelling errors. Healthcare workers have to put in an enormous effort, when trying to explain its content to ordinary people. Therefore it is quite urgent to automate the process of presentation and explanation of patient record to a patient. Different methods from Information Retrieval and Natural Language Processing fields were evaluated during work on this project. Several alternative solutions are studied in this thesis. Our approach is based on consequent filtering of terms by different algorithms, starting with the most accurate and fastest one and ending with the least accurate and slowest. Among the algorithms used for the filtering are modified Porter stemmer algorithm for Norwegian and set of transformation rules for translating Latin words, possibly misspelled, to Norwegian. It is a novel solution that provides for fully automated and reliable identifying of medical terms in mixed multi-lingual texts with a lot of irregularities. After such terms are recognized, an automated search is performed in limited set of dedicated electronic information sources. Results of this search are then stored in the form that provides for efficient access to the explanations of medical terms. We implemented prototype Web application that uses our own approach for automatic medical terms recognition and then performs search for explanations in thesaurus built from Norwegian Electronic Medical Handbook. The search is index-based and uses three indices dynamically generated for thesaurus.