A nonparametric surrogate-based test of significance for T-wave alternans detection

We present a nonparametric adaptive surrogate test that allows for the differentiation of statistically significant T-wave alternans (TWA) from alternating patterns that can be solely explained by the statistics of noise. The proposed test is based on estimating the distribution of noise-induced alternating patterns in a beat sequence from a set of surrogate data derived from repeated reshuffling of the original beat sequence. Thus, in assessing the significance of the observed alternating patterns in the data, no assumptions are made about the underlying noise distribution. In addition, since the distribution of noise-induced alternans magnitudes is calculated separately for each sequence of beats within the analysis window, the method is robust to data nonstationarities in both noise and TWA. The proposed surrogate method for rejecting noise was compared to the standard noise-rejection methods used with the spectral method (SM) and the modified moving average (MMA) techniques. Using a previously described realistic multilead model of TWA and real physiological noise, we demonstrate the proposed approach that reduces false TWA detections while maintaining a lower missed TWA detection, compared with all the other methods tested. A simple averaging-based TWA estimation algorithm was coupled with the surrogate significance testing and was evaluated on three public databases: the Normal Sinus Rhythm Database, the Chronic Heart Failure Database, and the Sudden Cardiac Death Database. Differences in TWA amplitudes between each database were evaluated at matched heart rate (HR) intervals from 40 to 120 beats per minute (BPM). Using the two-sample Kolmogorov-Smirnov test, we found that significant differences in TWA levels exist between each patient group at all decades of HRs. The most-marked difference was generally found at higher HRs, and the new technique resulted in a larger margin of separability between patient populations than when the SM or MMA were applied to the same data.