Prediction of congestive heart failure in patients using artificial intelligence: proof of concept

Abstract

Introduction: Due to the aging of the population we are confronting ourselves with an increased number of patients with chronic heart failure which stands a prevalence of over 37,7 million cases worldwide, being a major public health issue, first by the substantial morbidity and mortality that carries and second by the economic burden it brings annually upon the health-care system. The main purpose of this study is to test a machine learning algorithm which can predict, through voice analysis the acutisation of heart failure, considering the particularities of the patient's voice. Methods: The data we have used has been collected from a total of 16 patients, 9 men and 7 women, ages between 65 and 91 years old, who have agreed to take part in the study. The selective criteria of inclusion has been the cause of hospitalization, selecting only the patients presented with cardiogenic acute pulmonary edema, regardless the precipitation cause or other known cardiovascular comorbidities. With the same electronic device we have recorded each patient's voice, twice a day, starting from the day one of hospitalization, when their general status was critical, until the day of discharge, when they were clinically stable. Each voice recording containing specific keywords has been associated to the most used classification system for heart failure, the New York Heart Association Functional Classification and introduced into the machine learning algorithm. Results: After integrating the information from 15 patients, the algorithm has classified correctly the 16th patient into the third NYHA stage, based only on his voice recording. Conclusion: Voice is a cheap and easy way to monitor a patient's health status. The algorithm we have used for analyzing the voice provides high accuracy results, but for a larger dataset it might not be the best choice as it is computationally expensive. We are looking to obtain larger datasets and to compute more complex voice analyzer algorithms.