(Original news at the web of the Barcelona Center for New Medical Technologies)
Although cardiac resynchronization therapy reduces morbidity and mortality in patients with heart failure, heart failure syndrome affects particularly heterogeneous groups of patients, which can limit the treatment’s effectiveness.
Based on the aim of personalized medicine to optimize treatment adaptation to specific patients to maximize treatment response, the authors of a study recently published in the European Journal of Heart Failure have implemented an algorithm that, using complex ecocardiography data and clinical parameters, could identify precisely which patients with heart failure may benefit from cardiac resynchronization therapy.
The study has been carried out by Sergio Sánchez Martínez, who has recently defended his PhD thesis at the Department of Information and Communication Technologies (DTIC), directed by Bart Bijnens, ICREA-DTIC research professor and coordinator of the Sensing in Physiology and Biomedicine (PhySense) research group, of UPF’s BCN MedTech and by Gemma Piella, researcher and adjunct professor with the DTIC. This study is the result of collaboration with Maja Cikes, of the University Hospital of Zagreb, and Scott D. Solomon, a professor at the Harvard Medical School (Boston), and has been partly funded by Fundació “La Caixa” and Fundació La Marató de TV3.
The algorithm grouped patients according to their similarity
In a clinical trial, 1,106 patients diagnosed with heart failure were studied, of whom 677 received cardiac resynchronization therapy and 429 were treated only with an implantable cardioverter defibrillator. When analysing cardiac function along with demographic, clinical and medication parameters, the algorithm was able to group together similar patients using unsupervised clustering techniques, and to analyse the characteristics of each group in order to relate them with the therapeutic responses obtained.
Unsupervised learning algorithms characterize precisely the patients from a large number of clinical and echocardiographic parameters, in order to overcome the limitations of current diagnostic guidelines.
Technique based on unsupervised machine learning
Although the focus of machine learning has been applied to the diagnosis, classification, and recommendation of treatments in patients with heart failure, as well as to identify different phenotypes in various heart disorders, the authors propose a technique based on unsupervised machine learning that characterizes individuals precisely on the basis of a large number of clinical parameters, in order to overcome the limitations of clinical guidelines that do not manage to diagnose accurately and predict the prognosis of a group of patients with heart failure. Integrating clinical data with multiple ecocardiographic parameters such as myocardial deformation and volume changes in the left ventricle could overcome some of the limitations of the traditional approach.
The algorithm proposed in this work includes more than 1,600 parameters per patient that include demographic, clinical and medication data, and also complex patterns of shape and deformation of the left ventricle, which are impossible to describe with traditional methods
Data on the structure and function of the heart provided by echocardiography contain a large amount of information on the cardiac cycle that has traditionally not been taken advantage of. The algorithmproposed in this work includes more than 1,600 parameters per patient that include demographic, clinical and medication data, and also complex patterns of shape and deformation of the left ventricle, which are impossible to describe with traditional methods.
According to the authors, the results of this study “serve as a proof of concept that the integration of all these clinical parameters using unsupervised machine learning algorithms can provide a clinically significant classification of a group of patients with heart failure, which can help optimize the response rate to treatments such as cardiac resynchronization”.
Reference work:
Maja Cikes, Sergio Sanchez-Martinez, Brian Claggett, Nicolas Duchateau, Gemma Piella, Constantine Butakoff, Anne Catherine Pouleur, Dorit Knappe, Tor Biering-Sørensen, Valentina Kutyifa, Arthur Moss, Kenneth Stein, Scott D. Solomon, and Bart Bijnens ( 2018),“Machine learning-based phenogrouping in heart failure to identify responders to cardiac resynchronization therapy”, European Journal of Heart Failure, doi: 10.1002 / ejhf.1333.