Predicting the onset of paroxysmal atrial fibrillation: the Computers in Cardiology Challenge 2001

The advent of pacing techniques for preventing the onset of atrial arrhythmias motivates the development of accurate predictors of these arrhythmias, and of paroxysmal atrial fibrillation (PAF) in particular The goals of the second annual Computers in Cardiology Challenge were to determine if segments of ECG that do not include PAF contain information sufficient (1) to distinguish subjects at risk of PAF from others not at risk, and (2) to predict imminent PAF in at-risk subjects. Via PhysioNet, 18 teams of participants studied training and test databases containing two half-hour ECG recordings from each of 100 subjects (of whom 53 experienced PAF immediately following one of the two recordings). The results indicate that roughly 80% of the subjects can be correctly classified (as at-risk or not), and that imminent PAF can be predicted in roughly 80% of subjects at risk. The most successful approaches were based on analyses of the incidence of premature atrial complexes (PACs) and P-wave variability