Variations of heart rate variability parameters prior to the onset of ventricular tachyarrhythmia and sinus tachycardia in ICD patients. Results from the heart rate variability analysis with automated ICDs (HAWAI) registry

Abstract

The HAWAI registry evaluated the role of heart rate variability in predicting the occurrence of ventricular tachycardia and fibrillation (VT/VF) and sinus tachycardia in patients with an implantable cardioverter-defibrillator (45 patients with 155 RR recordings). A significant decrease of the mean value of all RR intervals (MeanNN) was observed in the period starting 20 and 40 min prior to VT/VF and sinus tachycardia, respectively. The standard deviation of RR intervals (SDNN) and the power at low frequency (LF) were the only parameters with significant changes prior to VT/VF. For sinus tachycardia, the root mean square of successive differences of all successive RR intervals (r-MSSD) and the power at low and high frequency (HF) decreased, whereas SDNN and the power at very low frequency increased. Comparison of RR recordings preceding VT/VF and sinus tachycardia revealed significant differences of the MeanNN, SDNN, r-MSSD, LF and HF. Based on a classification and regression tree analysis, MeanNN, SDNN and r-MSSD showed a sensitivity of 94.4% and a specificity of 50.6% as predictors of VT/VF. Our results suggest that the temporal changes in heart rate before an arrhythmic event can be used to predict the occurrence of VT/VF. These parameters may be used to optimize pacing therapies designed to prevent VT/VF recurrences as well as for improving device-based discriminators for VT/VF and sinus tachycardia.

Ventricular Tachyarrhythmia Prediction based on Heart Rate Variability and Genetic Algorithm

“…With 5 minutes of HRV signal length prior to VTA onset, our method achieves 79.41% of accuracy, which is higher than the accuracy reported by Joo, et al, [11] (best previous work) and Wollman, et al, [14] respectively. Although the accuracy levels were not reported by Thong & Raitt [12] and Rozen, et al, [13], their prediction sensitivity and specificity were not balanced with low sensitivity rate (53 and 50%).…”

“…With the additional types of HRV features and GA based optimization process, prediction performance of our method outperforms all previous works with 77.94%, 80.88% and 79.41 % for sensitivity, specificity and accuracy respectively even though we use stricter approach to evaluate our method. Prediction sensitivity and specificity of our method are more balanced when compared to previous works [12][13][14]. Furthermore, in contrast to most of previous works [12][13][14] that used more than 10 minutes of HRV signal, our method only uses 5 minutes HRV signal, which end immediately prior to VTA onset, for feature extraction.…”

“…Prediction sensitivity and specificity of our method are more balanced when compared to previous works [12][13][14]. Furthermore, in contrast to most of previous works [12][13][14] that used more than 10 minutes of HRV signal, our method only uses 5 minutes HRV signal, which end immediately prior to VTA onset, for feature extraction. Finally, optimal HRV feature subset selected by GA for prediction is also reported for reference in future works.…”

“…Furthermore, their method used significantly shorter HRV signal (5 minutes only) to achieve higher prediction accuracy. In contrast, other previous works [12][13][14] utilized more than 10 minutes of HRV signal in length prior to VTA event for prediction. The main reasons for such higher prediction performance were utilization of more comprehensive types of HRV features than other previous works and employment of more advanced supervised classifier -artificial neural network (ANN).…”

“…With decision rule based system based on multipole analysis, Rozen, et al, [13] achieved 50% of sensitivity and 91.6% of specificity. Wollman, et al, [14] proposed a method, which performed the VTA onset prediction based on time domain features and regression tree classifiers, obtained 70.9% of accuracy. Joo, et al, [11] proposed a prediction method that attained the most balanced and highest prediction performance among previous works with 77.3% of sensitivity, 73.8% of specificity and 75.6% of accuracy.…”