Background: Cardiovascular disease, such as coronary heart disease (CHD), are the main cause of mortality and morbidity worldwide. CHD is not entirely predicted by classic risk factors; however, they are preventable. Facing this major problem, the development of novel methods for CHD risk prediction is of practical interest. The purpose of our study was to construct an artificial neural networks (ANNs)-based diagnostic model for CHD risk using a complex of clinical and haemodynamics factors of this disease and aortic pulse wave velocity (PWV) index.
Methods: A total of 437 patients were included from 2012 to 2017: 99 CHD and 338 non-CHD patients. Theoretical PWV was calculated, on 93 patients free of hypertension, diabetes and CHD, according to age, blood pressure, sex and heart rate. The results were expressed as an index [(measured PWV - theoretical PWV)/theoretical PWV] for each patient. The original database for ANNs included clinical, haemodynamic and laboratory characteristics. Multilayered perceptron ANNs architecture were applied. The performance of prediction was evaluated by accuracy values based on standard definitions.
Results: By changing the types of ANNs and the number of input factors applied, we created models that demonstrated 0.63-0.93 accuracy. The best accuracy was obtained with ANNs topology of multilayer perceptron with three hidden layers for models, parameters included by both biological factors, carotid plaque and PWV index.
Conclusion: ANNs models including a PWV index could be used as promising approaches for predicting CHD risk without the need for invasive diagnostic methods and may help in the clinical decision.