Results of Data Mining Technique Applied to a Home Enteral Nutrition Database

Abstract

This chapter presents the results of data mining (ID3 and A PRIORI) techniques applied to a Health (nutrition) database that was originated from a knowledge management point of view. As participants of a Graduate Program in Surgery, we developed a knowledge management strategy and operationalized it by an information system, called SINPE, which is able to manipulate a large database (actually with 500,000+). Data items are organized into data collecting protocols that store data in a relational database. The system offers tools to retrieve and analyze data, with some basic (descriptive) statistics and graphic charts generated automatically. An interface allows to apply data mining algorithms into collected items. A main feature of the system is that it was developed from the usability point of view, because the primary users are physicians with low domain of information systems. Thus the system is very easy to manage. The aim of this work was to apply these data mining techniques to a home enteral nutrition database in order to identify features not previously suspected. The data mining technique was applied to a large health database protocol, with 1592 specific (nutrition) collected items. After the selection of interesting items the ID3 and APRIORI algorithms were applied to 111 patients, 58 females and 53 males, between 19 and 92 years old. These data were analyzed and presented in graphics and tables. Two questionnaires were answered by the users to validate the tool and its results. All operations were performed by physicians with low knowledge of data mining techniques, who were assisted by a BS in Computer Science professional. After mining the database, obtained results were compared with the international literature and the overall results met our expectations. Among other results, the data mining technique applied to the home enteral nutrition database identified an unexpected high incidence of malnutrition among patients that were receiving home enteral nutrition. Readmission after treatment was also higher than expected, reaching a 50% rate. Physicians who used the system approved it. No discrepancy was observed while using the system, but there are some parameters that must be better explained. The application of data mining techniques to a large medical (nutrition) database allowed us to identify nutrition features not previously known, which helped to improve public nutrition policies in this specific area. This user friendly system was proved useful when