Abstract:
The increasing number of hospital infections with considerable morbidity, mortality and economic burden attracts the attention of not only the health-care environment, but also the whole society. In some medical centers, hospital infections are traced with more controlled and extensive surveillance methods, which adopt data mining applications. Data mining methods are applied to find the outbreaks, which cannot be determined easily by infection control teams. This study presents an application of data mining methods for hospital infection detection in a newborn intensive care unit. The data set is provided by Department of Clinical Microbiology and Infectious Diseases, Eskişehir Osmangazi University, Faculty of Medicine. Decision tree, neural network and logistic regression classification models are built with holdout sampling and cross validation. In model comparison, accuracy and sensitivity measures are taken into consideration. Bagging and boosting methods are applied on neural network and decision trees in order to increase the performance of these models. According to the results, antibiotics and urinary catheter usage, peripheral catheter duration, enteral and total parenteral nutrition durations, and birth weight for gestational age are prominent risk factors. Among the models, neural network performs well on hospital infections detection representing 83% accuracy and 30% sensitivity on test data set. Furthermore, the sensitivity is improved with boosted neural network model to 44%.
