[Comparison of multiple prediction models for hypertension (Neural networks, logistic regression and flexible discriminant analyses)].
OBJECTIVE: In this study, we compared performances of logistic regression analysis (LR), flexible discriminant analysis (EAA) and neural networks (SA) in prediction of primary hypertension. METHODS: Predictor variables were family history, lipoprotein A, triglyceride, smoking and body mass index. The data were collected from Cardiology Clinic of Trakya University Medical Faculty in Turkey, 2001. Logistic regression analysis, flexible discriminant analysis and neural networks were used for prediction of control and hypertension groups. Comparison of the performance of all models was done using receiver operating characteristic (ROC) curve analysis. RESULTS: All models had areas under the ROC curve in the range of 0.793-0.984 and SA had sensitivity, specificity, and accuracy greater than 90% at ideal threshold. ROC curve areas of SA and LR, and SA and EAA were statistically different (p<0.001 and p<0.001 respectively), while ROC curve areas of EAA and LR did not differ (p>0.05). CONCLUSION: We concluded that family history, lipoprotein A, triglyceride, smoking and body mass index variables can be used for prediction of control and hypertension groups with statistically better performance of SA over LR and EAA.