Slope filtered pointwise correlation dimension algorithm and its evaluation with prefibrillation heart rate data.

Abstract

Various studies have shown that a low variability in heart rate is associated with increased risk of ventricular fibrillation. Low chaotic (correlation) dimension in the heart rate also appears to predict fibrillation risk. However, these results have been based on intergroup comparisons and have not been found useful for predicting when a patient may fibrillate with any degree of sensitivity, specificity, or temporal accuracy. There are two primary limitations in using dimensional analysis to predict imminent fibrillation. The first is that the standard algorithms (for correlation dimension) assume stationarity of the system. The second limitation is that these algorithms require 10,000-50,000 data points to achieve good accuracy. Thus, even if stationarity were not an issue, there would be a lag of 2.4-12 hours to warn of impending fibrillation. An algorithm has been developed to calculate an accurate pointwise correlation dimension of heart rate data. The slope filtered pointwise correlation dimension algorithm requires as few as 1,000 points of data. Using this algorithm, it was found that the correlation dimension dropped from 2.50 +/- 0.81 to 1.07 +/- 0.18 in the minute before fibrillation in conscious pigs with an occluded coronary artery. In clinical studies, Holter tapes from patients that had suffered fatal fibrillation were also analyzed along with healthy controls and nonfibrillation ventricular patients. The fibrillation patients all had excursions of low dimension (less than 1.5), while the majority of the others did not. In the minutes before fibrillation, the correlation dimension dropped to a steady range of 0.8-1.3. Drops in the slope filtered pointwise correlation dimension appear to predict fibrillation in animals and patients.