Antiarrhythmic drugs therapies are currently going through a turning point. The high risk that exists during the treatments has led to an ongoing search for new non-invasive toxicity risk biomarkers.
We propose the use of spatial biomarkers obtained through the quaternion algebra, evaluating the dynamics of the cardiac electrical vector in a non-invasive way in order to detect abnormal changes in ventricular heterogeneity. In groups of patients with and without history of Torsade de Pointes undergoing a Sotalol challenge, we compute the radius and the linear and angular velocities of QRS complex and T-wave loops. From these signals we extract significant features in order to compute a risk patient classifier.
Using machine learning techniques and statistical analysis, the combinations of few indices reach a pair of sensitivity/specificity of 100%/100% when separating patients with arrhythmogenic substrate. Several biomarkers not only measure drug-induced changes significantly but also observe differences in at-risk patients outperforming current standards.
Alternative biomarkers were able to describe pre-existing risk of patients. Given the high levels of significance and performance, these results could contribute to a better understanding of the torsadogenic substrate and to the safe development of drug therapies.