Application of a Bayesian Network to Predict Hospitalization among HIV Adults

Lee, Sun Mi

Abstract

OBJECTIVE
The purpose of this study was to explore the potential application of a Bayesian network, an emerging data mining technique, in predicting outcomes using large healthcare databases.
METHODS
The HIV Cost and Services Utilization Study(HCSUS) dataset, consisting of 2,864 HIV positive adults in the US, was used. A total of 35 variables were selected including one output variable defined as more than one hospitalization in six months representing a sub-optimal pattern of healthcare utilization in HIV care. The HUGIN Researcher 6.2(TM) was used to develop a Bayesian network model with two learning algorithms: 1) Necessary Path Condition(NPC) to construct a Bayesian network structure, and 2) Expectation-Maximization(EM) algorithm to estimate parameters.
RESULTS
The area under the Receiver Operating Characteristic(ROC) curve was .72. The accuracy of the prediction model was .66. Sensitivity and specificity were .65 and .66, respectively.
CONCLUSION
The Bayesian network showed fair performance in this prediction problem. This study provides researchers new insight into working with large sets of data, which continue to grow in a number of cases and variables. The repeated testing and refinement of the Bayesian network modeling techniques with other health outcomes in large databases is recommended.