Predicting hypertension without measurement: A non-invasive, questionnaire-based approach

Early detection of hypertension contributes to the prevention and reduction of the onset of cardiovascular diseases. Since lifestyle choices are linked to the occurrence and development of hypertension, determining hypertension risk factors and further establishing a predictive model with these factors will facilitate the early prevention and effective management of hypertension and improve individual health conditions. This study attempts to construct a prediction model based on the hybrid use of logistic regression and artificial neural networks (ANNs) for hypertension detection in a non-invasive, questionnaire-based way. First, the binary logistic regression model was used to select risk factors significant to hypertension. Second, after detailing the selection of ANNs architecture and the setting of relevant parameters, we constructed a multi-layer perception neural network model with back propagation learning algorithms to predict hypertension. Then, to mitigate the biased prediction results caused by a potentially unbalanced training set, we proposed an effective under-sampling technique and adopted it to balance the dataset prior to the training of the predictive model. To evaluate the performance of the proposed approach, we conducted extensive experiments on the questionnaires collected from Behavior Risk Factor Surveillance System. Experimental results show that ANN-based prediction model obtains over 72.0% accuracy and an area under the receiver-operator curve of 0.77 and achieves good stability in comparison with the logistic regression-based model. Further, the proposed approach obtains balanced prediction performance with the under-sampling technique. The results demonstrate the practicability of hypertension prediction with simple demographic data rather than with clinical tests and genomic data and of developing a hypertension surveillance system for a large scale of population in a non-invasive and economical way. Also, we actually provide a general framework for the simultaneous identification of risk factors and prediction of other chronic diseases.