Comparison of the Predictive Powers of Phenotypes Combined by Anthropometric Index and Triglyceride for Hypertension Diagnosis Based on Data Mining

Prediction and classification in data mining are commonly used to identify subjects with a particular disease in medical informatics. The hyper triglyceridemic waist (HW) phenotype is an important predictor of cardiovascular disease and hypertension. However, no study has evaluated the predictive power of phenotypes combined with individual anthropometric indices and triglyceride (TG) levels for hypertension diagnosis. The objectives in this study are to assess the association of hypertension with the HW phenotype and to compare the predictive powers of the phenotypes combined with individual indices and TG for hypertension diagnosis using data mining techniques. Based on 4906 Korean adult men, binary logistic regression (LR) was used for the evaluation of significant differences between normal and hypertensive conditions, and the naive Bayes algorithm (NB) and logistic regression (LR) were used to assess which index or phenotype showed better predictive power for hypertension. Among all variables, the presence of the HW phenotype showed the strongest association with hypertension, after adjustment for age and region, the phenotype still showed the best association (p = <;0.0001, OR = 0.555, adjusted OR = 0.58). RibC was the best predictor of hypertension among the single indices (p = <;0.0001, OR = 1.466, adjusted OR = 1.428, AUC = 0.609). Although the presence of HW had the strongest association with hypertension, the predictive power of the HW phenotype of waist circumference (WC) + TG with real values (not the presence of HW) was somewhat lower than that of the rib circumference (RibC) + TG phenotype. Therefore, the addition of TG to WC for the identification of hypertension is meaningless in the case of the use of real values of TG and WC. Our finding can provide useful information for the development of primary screening tools to identify subjects with hypertension.