To explore the feasibility of using the Bayesian network approach to study health outcomes and evaluate its predictive performance.

The Human immuno-deficiency virus Cost and Services Utilization Study (HCSUS) baseline dataset consisting of 2,864 human immuno-deficiency virus positive adults was used. The Hugin Researcher 6.2TM was used to develop the Bayesian network and Na?ve Bayes models. The SAS/STAT PROC LOGISTIC was used to develop the logistic regressions.

The area under the receiver operating characteristic curve of the Bayesian network model was statistically higher than that of the Na?ve Bayes model, but no higher than that of the logistic regression model using the 8 variables from a previous study. In a second analysis using the 10 most influential predictors discovered by the Bayesian network approach, the Na?ve Bayes and the logistic regression performance improved.

The BN approaches contributed to the discovery of additional influential predictors that lead to an increase of the models' predictive performance. When attempting to discover unknown relationships that might be missed by traditional analysis methods alone, the use of the Bayesian network as complementary methods may add value.