Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Editor,—I was intrigued by the paper “Prognostic significance of electrical alternans versus signal averaged electrocardiography in predicting the outcome of electrophysiological testing and arrhythmia-free survival” by Armoundaset al.1 In a time of statistics and prediction of important outcome events the reader has to pay much attention to the literature.
The conclusion “T wave alternans was a highly significant predictor of the outcome of electrophysiological testing and arrhythmia-free survival” is wrong; the positive predictive values are very low with very large confidence limits. The conclusion should be “There is a significant association between T wave alternans (TWA) and the outcome of electrophysiological (EP) testing. With single sided tests of significance there was a difference in arrhythmia-free survival for EP negative v EP positive patients, and for TWA negative v TWA positive patients.”
There are also some possible literal errors and problematic calculations.
In the tables accuracy = (A + D)/n. In table 2 line 5 PV− is 80%; is this correct?
In table 3, are there three or four arrhythmic events? If there are three events it is impossible to recalculate the other parameters. Figures 2 and 3 show four events; in the text it states that three patients developed ventricular tachycardia or fibrillation.
In table 3 line 3 accuracy is 65% (calculated from (3 + 17)/31); n = 31 but the values compatible with your calculations are A = 3; B = 7; C = 1; D = 17; n = 28. The value of D is probably wrong: if n = 31 it should be 20; in this case other parameters will change.
With a little effort the authors could have illustrated the original numbers of outcome events, and not just the calculation of indices.
This letter was shown to the authors, Dr Armoundis replies as follows:
The statement “T wave alternans was a highly significant predictor of the outcome of electrophysiological testing and arrhythmia-free survival” is an accurate characterisation of the results of our study as reflected by the p values for these two end points (0.0001 and 0.03).
The positive predictive values for T wave alternans as a predictor of the outcome of electrophysiological testing (64%) and 20 month arrhythmia-free survival (43%) are not low from a clinical perspective. For example, identification of a patient with a 43% probability of having cardiac arrest or sudden death within 20 months would certainly be adequate to justify therapeutic intervention. In contrast, the signal averaged ECG generally has been reported to have a positive predictive value on the order of 16%. It is true that our paper is an analysis of a small number of patients. We note that the T wave alternans results are confirmed in the larger group of patients in the original study1-1 from which the patients in our study were drawn. The intriguing results of our pilot study merits prospective examination of the comparative predictive accuracy of T wave alternans compared with other risk stratifiers. Several recent prospective studies have reported confirmatory results.1-2-1-4
I am grateful to Dr Ector for detecting one transcription error and one arithmetic error. In table 2, the entry in the row labelled TWA + SAECG-II and the column labelled PV− should be changed from 80% to 89%. In table 3 in the row labelled SAECG-I, the entry in the column labelled PV+ should be changed from 30% to 23%, and in the column labelled RR it should be changed from 5.4 to 4.2. Correction of these errors, while small in magnitude, tends further to support the conclusions of our paper.
Table 3 is computed from predicted 20 month arrhythmia-free survival values as measured from the computed Kaplan-Meier curves. The number of predicted events were rounded to the nearest integer value in each stratum. This accounts for the variation in the total number of predicted events between 3 and 4, which may be imputed from the different rows of the table.