Independent estimates of the axis of arrhythmia for assessing pro-arrhythmic drugs

Abstract

Many classes of drugs can induce potentially lethal cardiac arrhythmias. International safety guidelines (ICH-S7B) therefore recommend that all new drugs be tested for pro-arrhythmic risk prior to conducting human trials. The principal biomarker of drug-induced arrhythmia is prolongation of the ventricular cardiac action potential. Prolongation is known to be caused by drug-block of the hERG channel, hence that channel is of primary interest in safety testing. However, more accurate predictions of pro-arrhythmic risk can be achieved by considering the effect of the drug across multiple ion channels. The axis of arrhythmia is a new metric for predicting the pro-arrhythmic risk directly from a drug's effect on four key ion-currents (ICaL, IKr, INaL, IKs). In a previous study, we derived the axis of arrhythmia from computer simulations of a large population of ventricular cardiomyocytes. There, the axis was defined by the most potent combination of ion-channel blocks that shifted the simulated electrophysiology towards a pro-arrhythmic regime that was characterized by early-afterdepolarizations. In the present study, we derive the axis of arrhythmia using an entirely different method. Here, the axis is derived directly from a dataset of drugs (n = 109) without modeling the action potential. We statistically analyzed the drug potencies for each ion channel to find the linear boundary that optimally segregates the pro-arrhythmic drugs from the benign drugs, according to their clinical risk labels. The orthogonal line to that linear boundary corresponds to the axis of arrhythmia defined in our original study. The agreement between the two methods is remarkable. The axis derived from the simulated cardiomyocytes predicted the pro-arrhythmic risk of n = 109 test drugs with 88.1 %–90.8 % accuracy. In comparison, the axis derived from the drug dataset predicted the pro-arrhythmic risk with 89.9 %—90.8 % accuracy (cross-validated). The difference is negligible. Further analysis revealed that those accuracy rates could be improved to 90.8 %—92.7 % and 89.9 %—91.7 %, respectively, by excluding the IKs current. Such accuracy rates are comparable to the best biophysical model in contemporary computational cardiology. From a theoretical perspective, the independent derivations of the axis of arrhythmia represent convergent findings from complex biophysical models and simple statistical models.