Patient classification based on pre-hospital heart rate variability

Heart rate variability (HRV) is a non-invasive measurement that has shown promise as an indicator of cardiovascular, respiratory and metabolic dynamics. In this study, three different classification techniques, i.e. extreme learning machine (ELM), support vector machine (SVM) and back-propagation based neural network (BP), were investigated to classify HRV signals obtained from electrocardiograms (ECGs) of critically ill patients seen at the emergency department of a large hospital. HRV parameters were found to be better predictors of patient outcome than traditional vital signs. It was also found that the length of the ECG segment used affects the predictive ability of the classifiers and a windowing scheme was implemented to enhance performance.