Healthcare data mining: Predicting hospital length of stay of dengue patients

Abstract

Dengue is regarded as the most important mosquito-borne viral disease. Recently dengue has emerged as a public health burden in Southeast Asia and other tropical countries. At times when dengue re-emerges as an epidemic, hospitals are required to be able to handle patient flow fluctuation while maintaining their performance. This research applied a data mining technique to build a model that can predict in-patient hospital length of stay from the time of admission, which can be useful for effective decision-making that may lead to better clinical and resource management in hospitals. Using the C4.5 algorithm and a decision tree classifier, an accuracy of 71.57% and an area under the receiver operating characteristic (ROC) curve value of 0.761 were obtained. The decision tree showed that only 7 out of 21 input attributes affect the hospital length of stay prediction of dengue patients. The attribute with the highest impact was monocytes, followed by diastolic blood pressure, hematocrit, leucocytes, systolic blood pressure, comorbidity score, and lymphocytes. In this research also a prototype of a prediction system using the resulting model was developed.