Introduction: Tachyarrhythmia prediction prior to its onset in patients with ICD could allow delivery of preemptive therapies that alter the arrhythmia substrate. Our aim was to evaluate a novel mathematical method for tachyarrhythmia prediction.
Methods: Data of 30 ICD patients with ventricular tachyarrhythmia from the HAWAI registry (Biotronik Ltd) was analyzed. Each patient had several baseline & pre-shock files and each file was ~ 30 minutes long. For each baseline file, three 10 min. long segments were extracted & form each pre-shock file the 10 min. segment prior to delivered ICD shock was used. Overall 100 different baseline and 83 pre-shock examples were utilized. A discriminator was computed using and analyzing only the heart rate. Analysis was done by the Orthogonal Matching Pursuit (OMP) algorithm with an over-complete Fourier/Wavelet dictionary. This analysis included extraction of the energy & other features of the calculated Fourier or Wavelet coefficients and a classifier using logistic regression was used to discriminate between a baseline and a pre-arrhythmia example. 70% of the data was randomly used for training the discriminator algorithm and 30% was used for testing. This process was repeated 3 times and an ROC curve based on the percentage of recognition and the percentage of false positive recognition was built.
Results: The ROC curve showed the following: there was correct prediction of tachyarrhythmia of 79% with 8% false positive (FP), 84% correct prediction with 23% FP and 90% correct prediction with 44% FP.
Conclusions: An OMP based discriminator achieved detection of 78% of imminent shock needs with an 8% FP rate in a pooled, non-personalized data base with significant interpersonal clinical variations. There is a trade-off of prediction and false negative. We believe this initial analysis method for predictive tachyarrhythmia modeling can achieve improved recognition rates.
